| 1 | --- a/include/linux/netfilter_ipv4/ip_nat.h |
| 2 | +++ b/include/linux/netfilter_ipv4/ip_nat.h |
| 3 | @@ -121,5 +121,13 @@ extern int ip_nat_used_tuple(const struc |
| 4 | extern u_int16_t ip_nat_cheat_check(u_int32_t oldvalinv, |
| 5 | u_int32_t newval, |
| 6 | u_int16_t oldcheck); |
| 7 | + |
| 8 | +/* Call input routing for SNAT-ed traffic */ |
| 9 | +extern unsigned int ip_nat_route_input(unsigned int hooknum, |
| 10 | + struct sk_buff **pskb, |
| 11 | + const struct net_device *in, |
| 12 | + const struct net_device *out, |
| 13 | + int (*okfn)(struct sk_buff *)); |
| 14 | + |
| 15 | #endif /*__KERNEL__*/ |
| 16 | #endif |
| 17 | --- a/include/linux/rtnetlink.h |
| 18 | +++ b/include/linux/rtnetlink.h |
| 19 | @@ -234,6 +234,8 @@ struct rtnexthop |
| 20 | #define RTNH_F_DEAD 1 /* Nexthop is dead (used by multipath) */ |
| 21 | #define RTNH_F_PERVASIVE 2 /* Do recursive gateway lookup */ |
| 22 | #define RTNH_F_ONLINK 4 /* Gateway is forced on link */ |
| 23 | +#define RTNH_F_SUSPECT 8 /* We don't know the real state */ |
| 24 | +#define RTNH_F_BADSTATE (RTNH_F_DEAD | RTNH_F_SUSPECT) |
| 25 | |
| 26 | /* Macros to handle hexthops */ |
| 27 | |
| 28 | --- a/include/net/ip_fib.h |
| 29 | +++ b/include/net/ip_fib.h |
| 30 | @@ -162,7 +162,8 @@ static inline int fib_lookup(const struc |
| 31 | |
| 32 | static inline void fib_select_default(const struct rt_key *key, struct fib_result *res) |
| 33 | { |
| 34 | - if (FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) |
| 35 | + if ((FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) || |
| 36 | + FIB_RES_NH(*res).nh_scope == RT_SCOPE_HOST) |
| 37 | main_table->tb_select_default(main_table, key, res); |
| 38 | } |
| 39 | |
| 40 | @@ -174,6 +175,7 @@ extern struct fib_table * fib_tables[RT_ |
| 41 | extern int fib_lookup(const struct rt_key *key, struct fib_result *res); |
| 42 | extern struct fib_table *__fib_new_table(int id); |
| 43 | extern void fib_rule_put(struct fib_rule *r); |
| 44 | +extern int fib_result_table(struct fib_result *res); |
| 45 | |
| 46 | static inline struct fib_table *fib_get_table(int id) |
| 47 | { |
| 48 | @@ -275,5 +277,6 @@ static inline void fib_res_put(struct fi |
| 49 | #endif |
| 50 | } |
| 51 | |
| 52 | +extern rwlock_t fib_nhflags_lock; |
| 53 | |
| 54 | #endif /* _NET_FIB_H */ |
| 55 | --- a/include/net/route.h |
| 56 | +++ b/include/net/route.h |
| 57 | @@ -49,6 +49,8 @@ struct rt_key |
| 58 | { |
| 59 | __u32 dst; |
| 60 | __u32 src; |
| 61 | + __u32 lsrc; |
| 62 | + __u32 gw; |
| 63 | int iif; |
| 64 | int oif; |
| 65 | #ifdef CONFIG_IP_ROUTE_FWMARK |
| 66 | @@ -128,6 +130,7 @@ extern void ip_rt_advice(struct rtable |
| 67 | extern void rt_cache_flush(int how); |
| 68 | extern int ip_route_output_key(struct rtable **, const struct rt_key *key); |
| 69 | extern int ip_route_input(struct sk_buff*, u32 dst, u32 src, u8 tos, struct net_device *devin); |
| 70 | +extern int ip_route_input_lookup(struct sk_buff*, u32 dst, u32 src, u8 tos, struct net_device *devin, u32 lsrc); |
| 71 | extern unsigned short ip_rt_frag_needed(struct iphdr *iph, unsigned short new_mtu); |
| 72 | extern void ip_rt_update_pmtu(struct dst_entry *dst, unsigned mtu); |
| 73 | extern void ip_rt_send_redirect(struct sk_buff *skb); |
| 74 | @@ -148,6 +151,15 @@ static inline int ip_route_output(struct |
| 75 | } |
| 76 | |
| 77 | |
| 78 | +static inline int |
| 79 | +ip_route_output_lookup(struct rtable **rp, |
| 80 | + u32 daddr, u32 saddr, u32 tos, int oif, u32 gw) |
| 81 | +{ |
| 82 | + struct rt_key key = { dst:daddr, src:saddr, gw:gw, oif:oif, tos:tos }; |
| 83 | + |
| 84 | + return ip_route_output_key(rp, &key); |
| 85 | +} |
| 86 | + |
| 87 | static inline void ip_rt_put(struct rtable * rt) |
| 88 | { |
| 89 | if (rt) |
| 90 | --- a/net/ipv4/fib_frontend.c |
| 91 | +++ b/net/ipv4/fib_frontend.c |
| 92 | @@ -54,6 +54,8 @@ |
| 93 | struct fib_table *local_table; |
| 94 | struct fib_table *main_table; |
| 95 | |
| 96 | +#define FIB_RES_TABLE(r) (RT_TABLE_MAIN) |
| 97 | + |
| 98 | #else |
| 99 | |
| 100 | #define RT_TABLE_MIN 1 |
| 101 | @@ -71,6 +73,7 @@ struct fib_table *__fib_new_table(int id |
| 102 | return tb; |
| 103 | } |
| 104 | |
| 105 | +#define FIB_RES_TABLE(r) (fib_result_table(r)) |
| 106 | |
| 107 | #endif /* CONFIG_IP_MULTIPLE_TABLES */ |
| 108 | |
| 109 | @@ -209,6 +212,9 @@ int fib_validate_source(u32 src, u32 dst |
| 110 | struct in_device *in_dev; |
| 111 | struct rt_key key; |
| 112 | struct fib_result res; |
| 113 | + int table; |
| 114 | + unsigned char prefixlen; |
| 115 | + unsigned char scope; |
| 116 | int no_addr, rpf; |
| 117 | int ret; |
| 118 | |
| 119 | @@ -216,6 +222,7 @@ int fib_validate_source(u32 src, u32 dst |
| 120 | key.src = dst; |
| 121 | key.tos = tos; |
| 122 | key.oif = 0; |
| 123 | + key.gw = 0; |
| 124 | key.iif = oif; |
| 125 | key.scope = RT_SCOPE_UNIVERSE; |
| 126 | |
| 127 | @@ -237,31 +244,35 @@ int fib_validate_source(u32 src, u32 dst |
| 128 | goto e_inval_res; |
| 129 | *spec_dst = FIB_RES_PREFSRC(res); |
| 130 | fib_combine_itag(itag, &res); |
| 131 | -#ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 132 | - if (FIB_RES_DEV(res) == dev || res.fi->fib_nhs > 1) |
| 133 | -#else |
| 134 | if (FIB_RES_DEV(res) == dev) |
| 135 | -#endif |
| 136 | { |
| 137 | ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; |
| 138 | fib_res_put(&res); |
| 139 | return ret; |
| 140 | } |
| 141 | + table = FIB_RES_TABLE(&res); |
| 142 | + prefixlen = res.prefixlen; |
| 143 | + scope = res.scope; |
| 144 | fib_res_put(&res); |
| 145 | if (no_addr) |
| 146 | goto last_resort; |
| 147 | - if (rpf) |
| 148 | - goto e_inval; |
| 149 | key.oif = dev->ifindex; |
| 150 | |
| 151 | ret = 0; |
| 152 | if (fib_lookup(&key, &res) == 0) { |
| 153 | - if (res.type == RTN_UNICAST) { |
| 154 | + if (res.type == RTN_UNICAST && |
| 155 | + ((table == FIB_RES_TABLE(&res) && |
| 156 | + res.prefixlen >= prefixlen && res.scope >= scope) || |
| 157 | + !rpf)) { |
| 158 | *spec_dst = FIB_RES_PREFSRC(res); |
| 159 | ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; |
| 160 | + fib_res_put(&res); |
| 161 | + return ret; |
| 162 | } |
| 163 | fib_res_put(&res); |
| 164 | } |
| 165 | + if (rpf) |
| 166 | + goto e_inval; |
| 167 | return ret; |
| 168 | |
| 169 | last_resort: |
| 170 | @@ -579,9 +590,7 @@ static int fib_inetaddr_event(struct not |
| 171 | switch (event) { |
| 172 | case NETDEV_UP: |
| 173 | fib_add_ifaddr(ifa); |
| 174 | -#ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 175 | fib_sync_up(ifa->ifa_dev->dev); |
| 176 | -#endif |
| 177 | rt_cache_flush(-1); |
| 178 | break; |
| 179 | case NETDEV_DOWN: |
| 180 | @@ -617,9 +626,7 @@ static int fib_netdev_event(struct notif |
| 181 | for_ifa(in_dev) { |
| 182 | fib_add_ifaddr(ifa); |
| 183 | } endfor_ifa(in_dev); |
| 184 | -#ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 185 | fib_sync_up(dev); |
| 186 | -#endif |
| 187 | rt_cache_flush(-1); |
| 188 | break; |
| 189 | case NETDEV_DOWN: |
| 190 | --- a/net/ipv4/fib_hash.c |
| 191 | +++ b/net/ipv4/fib_hash.c |
| 192 | @@ -71,6 +71,7 @@ struct fib_node |
| 193 | struct fib_info *fn_info; |
| 194 | #define FIB_INFO(f) ((f)->fn_info) |
| 195 | fn_key_t fn_key; |
| 196 | + int fn_last_dflt; |
| 197 | u8 fn_tos; |
| 198 | u8 fn_type; |
| 199 | u8 fn_scope; |
| 200 | @@ -336,72 +337,123 @@ out: |
| 201 | return err; |
| 202 | } |
| 203 | |
| 204 | -static int fn_hash_last_dflt=-1; |
| 205 | - |
| 206 | -static int fib_detect_death(struct fib_info *fi, int order, |
| 207 | - struct fib_info **last_resort, int *last_idx) |
| 208 | +static int fib_detect_death(struct fib_info *fi, int order, int last_dflt, |
| 209 | + struct fib_info **last_resort, int *last_idx, |
| 210 | + int *last_nhsel, const struct rt_key *key) |
| 211 | { |
| 212 | struct neighbour *n; |
| 213 | - int state = NUD_NONE; |
| 214 | + int nhsel; |
| 215 | + int state; |
| 216 | + struct fib_nh * nh; |
| 217 | + u32 dst; |
| 218 | + int flag, dead = 1; |
| 219 | + |
| 220 | + /* change_nexthops(fi) { */ |
| 221 | + for (nhsel = 0, nh = fi->fib_nh; nhsel < fi->fib_nhs; nh++, nhsel++) { |
| 222 | + if (key->oif && key->oif != nh->nh_oif) |
| 223 | + continue; |
| 224 | + if (key->gw && key->gw != nh->nh_gw && nh->nh_gw && |
| 225 | + nh->nh_scope == RT_SCOPE_LINK) |
| 226 | + continue; |
| 227 | + if (nh->nh_flags & RTNH_F_DEAD) |
| 228 | + continue; |
| 229 | |
| 230 | - n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev); |
| 231 | - if (n) { |
| 232 | - state = n->nud_state; |
| 233 | - neigh_release(n); |
| 234 | + flag = 0; |
| 235 | + if (nh->nh_dev->flags & IFF_NOARP) { |
| 236 | + dead = 0; |
| 237 | + goto setfl; |
| 238 | + } |
| 239 | + |
| 240 | + dst = nh->nh_gw; |
| 241 | + if (!nh->nh_gw || nh->nh_scope != RT_SCOPE_LINK) |
| 242 | + dst = key->dst; |
| 243 | + |
| 244 | + state = NUD_NONE; |
| 245 | + n = neigh_lookup(&arp_tbl, &dst, nh->nh_dev); |
| 246 | + if (n) { |
| 247 | + state = n->nud_state; |
| 248 | + neigh_release(n); |
| 249 | + } |
| 250 | + if (state==NUD_REACHABLE || |
| 251 | + ((state&NUD_VALID) && order != last_dflt)) { |
| 252 | + dead = 0; |
| 253 | + goto setfl; |
| 254 | + } |
| 255 | + if (!(state&NUD_VALID)) |
| 256 | + flag = 1; |
| 257 | + if (!dead) |
| 258 | + goto setfl; |
| 259 | + if ((state&NUD_VALID) || |
| 260 | + (*last_idx<0 && order >= last_dflt)) { |
| 261 | + *last_resort = fi; |
| 262 | + *last_idx = order; |
| 263 | + *last_nhsel = nhsel; |
| 264 | + } |
| 265 | + |
| 266 | + setfl: |
| 267 | + |
| 268 | + read_lock_bh(&fib_nhflags_lock); |
| 269 | + if (flag) |
| 270 | + nh->nh_flags |= RTNH_F_SUSPECT; |
| 271 | + else |
| 272 | + nh->nh_flags &= ~RTNH_F_SUSPECT; |
| 273 | + read_unlock_bh(&fib_nhflags_lock); |
| 274 | } |
| 275 | - if (state==NUD_REACHABLE) |
| 276 | - return 0; |
| 277 | - if ((state&NUD_VALID) && order != fn_hash_last_dflt) |
| 278 | - return 0; |
| 279 | - if ((state&NUD_VALID) || |
| 280 | - (*last_idx<0 && order > fn_hash_last_dflt)) { |
| 281 | - *last_resort = fi; |
| 282 | - *last_idx = order; |
| 283 | - } |
| 284 | - return 1; |
| 285 | + /* } endfor_nexthops(fi) */ |
| 286 | + |
| 287 | + return dead; |
| 288 | } |
| 289 | |
| 290 | static void |
| 291 | fn_hash_select_default(struct fib_table *tb, const struct rt_key *key, struct fib_result *res) |
| 292 | { |
| 293 | - int order, last_idx; |
| 294 | - struct fib_node *f; |
| 295 | + int order, last_idx, last_dflt, last_nhsel; |
| 296 | + struct fib_node *f, *first_node; |
| 297 | struct fib_info *fi = NULL; |
| 298 | struct fib_info *last_resort; |
| 299 | struct fn_hash *t = (struct fn_hash*)tb->tb_data; |
| 300 | - struct fn_zone *fz = t->fn_zones[0]; |
| 301 | + struct fn_zone *fz = t->fn_zones[res->prefixlen]; |
| 302 | + fn_key_t k; |
| 303 | |
| 304 | if (fz == NULL) |
| 305 | return; |
| 306 | |
| 307 | + k = fz_key(key->dst, fz); |
| 308 | + last_dflt = -2; |
| 309 | + first_node = NULL; |
| 310 | last_idx = -1; |
| 311 | last_resort = NULL; |
| 312 | + last_nhsel = 0; |
| 313 | order = -1; |
| 314 | |
| 315 | read_lock(&fib_hash_lock); |
| 316 | - for (f = fz->fz_hash[0]; f; f = f->fn_next) { |
| 317 | + for (f = fz_chain(k, fz); f; f = f->fn_next) { |
| 318 | struct fib_info *next_fi = FIB_INFO(f); |
| 319 | |
| 320 | - if ((f->fn_state&FN_S_ZOMBIE) || |
| 321 | + if (!fn_key_eq(k, f->fn_key) || |
| 322 | + (f->fn_state&FN_S_ZOMBIE) || |
| 323 | f->fn_scope != res->scope || |
| 324 | +#ifdef CONFIG_IP_ROUTE_TOS |
| 325 | + (f->fn_tos && f->fn_tos != key->tos) || |
| 326 | +#endif |
| 327 | f->fn_type != RTN_UNICAST) |
| 328 | continue; |
| 329 | |
| 330 | if (next_fi->fib_priority > res->fi->fib_priority) |
| 331 | break; |
| 332 | - if (!next_fi->fib_nh[0].nh_gw || next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK) |
| 333 | - continue; |
| 334 | f->fn_state |= FN_S_ACCESSED; |
| 335 | |
| 336 | - if (fi == NULL) { |
| 337 | - if (next_fi != res->fi) |
| 338 | - break; |
| 339 | - } else if (!fib_detect_death(fi, order, &last_resort, &last_idx)) { |
| 340 | + if (!first_node) { |
| 341 | + last_dflt = f->fn_last_dflt; |
| 342 | + first_node = f; |
| 343 | + } |
| 344 | + if (fi && !fib_detect_death(fi, order, last_dflt, |
| 345 | + &last_resort, &last_idx, &last_nhsel, key)) { |
| 346 | if (res->fi) |
| 347 | fib_info_put(res->fi); |
| 348 | res->fi = fi; |
| 349 | atomic_inc(&fi->fib_clntref); |
| 350 | - fn_hash_last_dflt = order; |
| 351 | + first_node->fn_last_dflt = order; |
| 352 | goto out; |
| 353 | } |
| 354 | fi = next_fi; |
| 355 | @@ -409,16 +461,25 @@ fn_hash_select_default(struct fib_table |
| 356 | } |
| 357 | |
| 358 | if (order<=0 || fi==NULL) { |
| 359 | - fn_hash_last_dflt = -1; |
| 360 | + if (fi && fi->fib_nhs > 1 && |
| 361 | + fib_detect_death(fi, order, last_dflt, |
| 362 | + &last_resort, &last_idx, &last_nhsel, key) && |
| 363 | + last_resort == fi) { |
| 364 | + read_lock_bh(&fib_nhflags_lock); |
| 365 | + fi->fib_nh[last_nhsel].nh_flags &= ~RTNH_F_SUSPECT; |
| 366 | + read_unlock_bh(&fib_nhflags_lock); |
| 367 | + } |
| 368 | + if (first_node) first_node->fn_last_dflt = -1; |
| 369 | goto out; |
| 370 | } |
| 371 | |
| 372 | - if (!fib_detect_death(fi, order, &last_resort, &last_idx)) { |
| 373 | + if (!fib_detect_death(fi, order, last_dflt, &last_resort, &last_idx, |
| 374 | + &last_nhsel, key)) { |
| 375 | if (res->fi) |
| 376 | fib_info_put(res->fi); |
| 377 | res->fi = fi; |
| 378 | atomic_inc(&fi->fib_clntref); |
| 379 | - fn_hash_last_dflt = order; |
| 380 | + first_node->fn_last_dflt = order; |
| 381 | goto out; |
| 382 | } |
| 383 | |
| 384 | @@ -428,8 +489,11 @@ fn_hash_select_default(struct fib_table |
| 385 | res->fi = last_resort; |
| 386 | if (last_resort) |
| 387 | atomic_inc(&last_resort->fib_clntref); |
| 388 | + read_lock_bh(&fib_nhflags_lock); |
| 389 | + last_resort->fib_nh[last_nhsel].nh_flags &= ~RTNH_F_SUSPECT; |
| 390 | + read_unlock_bh(&fib_nhflags_lock); |
| 391 | + first_node->fn_last_dflt = last_idx; |
| 392 | } |
| 393 | - fn_hash_last_dflt = last_idx; |
| 394 | out: |
| 395 | read_unlock(&fib_hash_lock); |
| 396 | } |
| 397 | @@ -589,6 +653,7 @@ replace: |
| 398 | |
| 399 | memset(new_f, 0, sizeof(struct fib_node)); |
| 400 | |
| 401 | + new_f->fn_last_dflt = -1; |
| 402 | new_f->fn_key = key; |
| 403 | #ifdef CONFIG_IP_ROUTE_TOS |
| 404 | new_f->fn_tos = tos; |
| 405 | --- a/net/ipv4/fib_rules.c |
| 406 | +++ b/net/ipv4/fib_rules.c |
| 407 | @@ -307,6 +307,11 @@ static void fib_rules_attach(struct net_ |
| 408 | } |
| 409 | } |
| 410 | |
| 411 | +int fib_result_table(struct fib_result *res) |
| 412 | +{ |
| 413 | + return res->r->r_table; |
| 414 | +} |
| 415 | + |
| 416 | int fib_lookup(const struct rt_key *key, struct fib_result *res) |
| 417 | { |
| 418 | int err; |
| 419 | @@ -371,8 +376,10 @@ FRprintk("FAILURE\n"); |
| 420 | |
| 421 | void fib_select_default(const struct rt_key *key, struct fib_result *res) |
| 422 | { |
| 423 | - if (res->r && res->r->r_action == RTN_UNICAST && |
| 424 | - FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) { |
| 425 | + if (res->r && |
| 426 | + (res->r->r_action == RTN_UNICAST || res->r->r_action == RTN_NAT) && |
| 427 | + ((FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) || |
| 428 | + FIB_RES_NH(*res).nh_scope == RT_SCOPE_HOST)) { |
| 429 | struct fib_table *tb; |
| 430 | if ((tb = fib_get_table(res->r->r_table)) != NULL) |
| 431 | tb->tb_select_default(tb, key, res); |
| 432 | --- a/net/ipv4/fib_semantics.c |
| 433 | +++ b/net/ipv4/fib_semantics.c |
| 434 | @@ -48,6 +48,7 @@ |
| 435 | static struct fib_info *fib_info_list; |
| 436 | static rwlock_t fib_info_lock = RW_LOCK_UNLOCKED; |
| 437 | int fib_info_cnt; |
| 438 | +rwlock_t fib_nhflags_lock = RW_LOCK_UNLOCKED; |
| 439 | |
| 440 | #define for_fib_info() { struct fib_info *fi; \ |
| 441 | for (fi = fib_info_list; fi; fi = fi->fib_next) |
| 442 | @@ -150,7 +151,7 @@ static __inline__ int nh_comp(const stru |
| 443 | #ifdef CONFIG_NET_CLS_ROUTE |
| 444 | nh->nh_tclassid != onh->nh_tclassid || |
| 445 | #endif |
| 446 | - ((nh->nh_flags^onh->nh_flags)&~RTNH_F_DEAD)) |
| 447 | + ((nh->nh_flags^onh->nh_flags)&~RTNH_F_BADSTATE)) |
| 448 | return -1; |
| 449 | onh++; |
| 450 | } endfor_nexthops(fi); |
| 451 | @@ -166,7 +167,7 @@ static __inline__ struct fib_info * fib_ |
| 452 | nfi->fib_prefsrc == fi->fib_prefsrc && |
| 453 | nfi->fib_priority == fi->fib_priority && |
| 454 | memcmp(nfi->fib_metrics, fi->fib_metrics, sizeof(fi->fib_metrics)) == 0 && |
| 455 | - ((nfi->fib_flags^fi->fib_flags)&~RTNH_F_DEAD) == 0 && |
| 456 | + ((nfi->fib_flags^fi->fib_flags)&~RTNH_F_BADSTATE) == 0 && |
| 457 | (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0)) |
| 458 | return fi; |
| 459 | } endfor_fib_info(); |
| 460 | @@ -365,8 +366,11 @@ static int fib_check_nh(const struct rtm |
| 461 | return -EINVAL; |
| 462 | if ((dev = __dev_get_by_index(nh->nh_oif)) == NULL) |
| 463 | return -ENODEV; |
| 464 | - if (!(dev->flags&IFF_UP)) |
| 465 | - return -ENETDOWN; |
| 466 | + if (!(dev->flags&IFF_UP)) { |
| 467 | + if (fi->fib_protocol != RTPROT_STATIC) |
| 468 | + return -ENETDOWN; |
| 469 | + nh->nh_flags |= RTNH_F_DEAD; |
| 470 | + } |
| 471 | nh->nh_dev = dev; |
| 472 | dev_hold(dev); |
| 473 | nh->nh_scope = RT_SCOPE_LINK; |
| 474 | @@ -380,23 +384,48 @@ static int fib_check_nh(const struct rtm |
| 475 | /* It is not necessary, but requires a bit of thinking */ |
| 476 | if (key.scope < RT_SCOPE_LINK) |
| 477 | key.scope = RT_SCOPE_LINK; |
| 478 | - if ((err = fib_lookup(&key, &res)) != 0) |
| 479 | - return err; |
| 480 | - err = -EINVAL; |
| 481 | - if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) |
| 482 | - goto out; |
| 483 | - nh->nh_scope = res.scope; |
| 484 | - nh->nh_oif = FIB_RES_OIF(res); |
| 485 | - if ((nh->nh_dev = FIB_RES_DEV(res)) == NULL) |
| 486 | - goto out; |
| 487 | - dev_hold(nh->nh_dev); |
| 488 | - err = -ENETDOWN; |
| 489 | - if (!(nh->nh_dev->flags & IFF_UP)) |
| 490 | - goto out; |
| 491 | - err = 0; |
| 492 | + |
| 493 | + err = fib_lookup(&key, &res); |
| 494 | + if (err) { |
| 495 | + struct in_device *in_dev; |
| 496 | + |
| 497 | + if (err != -ENETUNREACH || |
| 498 | + fi->fib_protocol != RTPROT_STATIC) |
| 499 | + return err; |
| 500 | + |
| 501 | + in_dev = inetdev_by_index(nh->nh_oif); |
| 502 | + if (in_dev == NULL || |
| 503 | + in_dev->dev->flags & IFF_UP) { |
| 504 | + if (in_dev) |
| 505 | + in_dev_put(in_dev); |
| 506 | + return err; |
| 507 | + } |
| 508 | + nh->nh_flags |= RTNH_F_DEAD; |
| 509 | + nh->nh_scope = RT_SCOPE_LINK; |
| 510 | + nh->nh_dev = in_dev->dev; |
| 511 | + dev_hold(nh->nh_dev); |
| 512 | + in_dev_put(in_dev); |
| 513 | + } else { |
| 514 | + err = -EINVAL; |
| 515 | + if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) |
| 516 | + goto out; |
| 517 | + nh->nh_scope = res.scope; |
| 518 | + nh->nh_oif = FIB_RES_OIF(res); |
| 519 | + if ((nh->nh_dev = FIB_RES_DEV(res)) == NULL) |
| 520 | + goto out; |
| 521 | + dev_hold(nh->nh_dev); |
| 522 | + if (!(nh->nh_dev->flags & IFF_UP)) { |
| 523 | + if (fi->fib_protocol != RTPROT_STATIC) { |
| 524 | + err = -ENETDOWN; |
| 525 | + goto out; |
| 526 | + } |
| 527 | + nh->nh_flags |= RTNH_F_DEAD; |
| 528 | + } |
| 529 | + err = 0; |
| 530 | out: |
| 531 | - fib_res_put(&res); |
| 532 | - return err; |
| 533 | + fib_res_put(&res); |
| 534 | + return err; |
| 535 | + } |
| 536 | } else { |
| 537 | struct in_device *in_dev; |
| 538 | |
| 539 | @@ -407,8 +436,11 @@ out: |
| 540 | if (in_dev == NULL) |
| 541 | return -ENODEV; |
| 542 | if (!(in_dev->dev->flags&IFF_UP)) { |
| 543 | - in_dev_put(in_dev); |
| 544 | - return -ENETDOWN; |
| 545 | + if (fi->fib_protocol != RTPROT_STATIC) { |
| 546 | + in_dev_put(in_dev); |
| 547 | + return -ENETDOWN; |
| 548 | + } |
| 549 | + nh->nh_flags |= RTNH_F_DEAD; |
| 550 | } |
| 551 | nh->nh_dev = in_dev->dev; |
| 552 | dev_hold(nh->nh_dev); |
| 553 | @@ -606,8 +638,12 @@ fib_semantic_match(int type, struct fib_ |
| 554 | for_nexthops(fi) { |
| 555 | if (nh->nh_flags&RTNH_F_DEAD) |
| 556 | continue; |
| 557 | - if (!key->oif || key->oif == nh->nh_oif) |
| 558 | - break; |
| 559 | + if (key->oif && key->oif != nh->nh_oif) |
| 560 | + continue; |
| 561 | + if (key->gw && key->gw != nh->nh_gw && |
| 562 | + nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) |
| 563 | + continue; |
| 564 | + break; |
| 565 | } |
| 566 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 567 | if (nhsel < fi->fib_nhs) { |
| 568 | @@ -873,22 +909,35 @@ int fib_sync_down(u32 local, struct net_ |
| 569 | if (local && fi->fib_prefsrc == local) { |
| 570 | fi->fib_flags |= RTNH_F_DEAD; |
| 571 | ret++; |
| 572 | - } else if (dev && fi->fib_nhs) { |
| 573 | + } else if (fi->fib_nhs) { |
| 574 | int dead = 0; |
| 575 | |
| 576 | change_nexthops(fi) { |
| 577 | - if (nh->nh_flags&RTNH_F_DEAD) |
| 578 | - dead++; |
| 579 | - else if (nh->nh_dev == dev && |
| 580 | - nh->nh_scope != scope) { |
| 581 | - nh->nh_flags |= RTNH_F_DEAD; |
| 582 | + if (nh->nh_flags&RTNH_F_DEAD) { |
| 583 | + if (fi->fib_protocol!=RTPROT_STATIC || |
| 584 | + nh->nh_dev == NULL || |
| 585 | + !__in_dev_get(nh->nh_dev) || |
| 586 | + nh->nh_dev->flags&IFF_UP) |
| 587 | + dead++; |
| 588 | + } else if ((nh->nh_dev == dev && dev && |
| 589 | + nh->nh_scope != scope) || |
| 590 | + (local == nh->nh_gw && local && |
| 591 | + nh->nh_oif)) { |
| 592 | + write_lock_bh(&fib_nhflags_lock); |
| 593 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 594 | - spin_lock_bh(&fib_multipath_lock); |
| 595 | + spin_lock(&fib_multipath_lock); |
| 596 | + nh->nh_flags |= RTNH_F_DEAD; |
| 597 | fi->fib_power -= nh->nh_power; |
| 598 | nh->nh_power = 0; |
| 599 | - spin_unlock_bh(&fib_multipath_lock); |
| 600 | + spin_unlock(&fib_multipath_lock); |
| 601 | +#else |
| 602 | + nh->nh_flags |= RTNH_F_DEAD; |
| 603 | #endif |
| 604 | - dead++; |
| 605 | + write_unlock_bh(&fib_nhflags_lock); |
| 606 | + if (fi->fib_protocol!=RTPROT_STATIC || |
| 607 | + force || |
| 608 | + (dev && __in_dev_get(dev) == NULL)) |
| 609 | + dead++; |
| 610 | } |
| 611 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 612 | if (force > 1 && nh->nh_dev == dev) { |
| 613 | @@ -906,37 +955,55 @@ int fib_sync_down(u32 local, struct net_ |
| 614 | return ret; |
| 615 | } |
| 616 | |
| 617 | -#ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 618 | - |
| 619 | /* |
| 620 | - Dead device goes up. We wake up dead nexthops. |
| 621 | - It takes sense only on multipath routes. |
| 622 | + Dead device goes up or new address is added. We wake up dead nexthops. |
| 623 | */ |
| 624 | |
| 625 | int fib_sync_up(struct net_device *dev) |
| 626 | { |
| 627 | - int ret = 0; |
| 628 | + struct rt_key key; |
| 629 | + struct fib_result res; |
| 630 | + int ret, rep; |
| 631 | |
| 632 | +repeat: |
| 633 | if (!(dev->flags&IFF_UP)) |
| 634 | return 0; |
| 635 | |
| 636 | + ret = 0; |
| 637 | + rep = 0; |
| 638 | for_fib_info() { |
| 639 | int alive = 0; |
| 640 | |
| 641 | change_nexthops(fi) { |
| 642 | - if (!(nh->nh_flags&RTNH_F_DEAD)) { |
| 643 | - alive++; |
| 644 | + if (!(nh->nh_flags&RTNH_F_DEAD)) |
| 645 | continue; |
| 646 | - } |
| 647 | if (nh->nh_dev == NULL || !(nh->nh_dev->flags&IFF_UP)) |
| 648 | continue; |
| 649 | if (nh->nh_dev != dev || __in_dev_get(dev) == NULL) |
| 650 | continue; |
| 651 | + if (nh->nh_gw && fi->fib_protocol == RTPROT_STATIC) { |
| 652 | + memset(&key, 0, sizeof(key)); |
| 653 | + key.dst = nh->nh_gw; |
| 654 | + key.oif = nh->nh_oif; |
| 655 | + key.scope = nh->nh_scope; |
| 656 | + if (fib_lookup(&key, &res) != 0) |
| 657 | + continue; |
| 658 | + if (res.type != RTN_UNICAST && |
| 659 | + res.type != RTN_LOCAL) { |
| 660 | + fib_res_put(&res); |
| 661 | + continue; |
| 662 | + } |
| 663 | + nh->nh_scope = res.scope; |
| 664 | + fib_res_put(&res); |
| 665 | + rep = 1; |
| 666 | + } |
| 667 | alive++; |
| 668 | +#ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 669 | spin_lock_bh(&fib_multipath_lock); |
| 670 | nh->nh_power = 0; |
| 671 | nh->nh_flags &= ~RTNH_F_DEAD; |
| 672 | spin_unlock_bh(&fib_multipath_lock); |
| 673 | +#endif |
| 674 | } endfor_nexthops(fi) |
| 675 | |
| 676 | if (alive > 0) { |
| 677 | @@ -944,9 +1011,13 @@ int fib_sync_up(struct net_device *dev) |
| 678 | ret++; |
| 679 | } |
| 680 | } endfor_fib_info(); |
| 681 | + if (rep) |
| 682 | + goto repeat; |
| 683 | return ret; |
| 684 | } |
| 685 | |
| 686 | +#ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 687 | + |
| 688 | /* |
| 689 | The algorithm is suboptimal, but it provides really |
| 690 | fair weighted route distribution. |
| 691 | @@ -955,24 +1026,45 @@ int fib_sync_up(struct net_device *dev) |
| 692 | void fib_select_multipath(const struct rt_key *key, struct fib_result *res) |
| 693 | { |
| 694 | struct fib_info *fi = res->fi; |
| 695 | - int w; |
| 696 | + int w, alive; |
| 697 | |
| 698 | spin_lock_bh(&fib_multipath_lock); |
| 699 | + if (key->oif) { |
| 700 | + int sel = -1; |
| 701 | + w = -1; |
| 702 | + change_nexthops(fi) { |
| 703 | + if (key->oif != nh->nh_oif) |
| 704 | + continue; |
| 705 | + if (key->gw && key->gw != nh->nh_gw && |
| 706 | + nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) |
| 707 | + continue; |
| 708 | + if (!(nh->nh_flags&RTNH_F_BADSTATE)) { |
| 709 | + if (nh->nh_power > w) { |
| 710 | + w = nh->nh_power; |
| 711 | + sel = nhsel; |
| 712 | + } |
| 713 | + } |
| 714 | + } endfor_nexthops(fi); |
| 715 | + if (sel >= 0) { |
| 716 | + spin_unlock_bh(&fib_multipath_lock); |
| 717 | + res->nh_sel = sel; |
| 718 | + return; |
| 719 | + } |
| 720 | + goto last_resort; |
| 721 | + } |
| 722 | + |
| 723 | +repeat: |
| 724 | if (fi->fib_power <= 0) { |
| 725 | int power = 0; |
| 726 | change_nexthops(fi) { |
| 727 | - if (!(nh->nh_flags&RTNH_F_DEAD)) { |
| 728 | + if (!(nh->nh_flags&RTNH_F_BADSTATE)) { |
| 729 | power += nh->nh_weight; |
| 730 | nh->nh_power = nh->nh_weight; |
| 731 | } |
| 732 | } endfor_nexthops(fi); |
| 733 | fi->fib_power = power; |
| 734 | - if (power <= 0) { |
| 735 | - spin_unlock_bh(&fib_multipath_lock); |
| 736 | - /* Race condition: route has just become dead. */ |
| 737 | - res->nh_sel = 0; |
| 738 | - return; |
| 739 | - } |
| 740 | + if (power <= 0) |
| 741 | + goto last_resort; |
| 742 | } |
| 743 | |
| 744 | |
| 745 | @@ -982,20 +1074,40 @@ void fib_select_multipath(const struct r |
| 746 | |
| 747 | w = jiffies % fi->fib_power; |
| 748 | |
| 749 | + alive = 0; |
| 750 | change_nexthops(fi) { |
| 751 | - if (!(nh->nh_flags&RTNH_F_DEAD) && nh->nh_power) { |
| 752 | + if (!(nh->nh_flags&RTNH_F_BADSTATE) && nh->nh_power) { |
| 753 | if ((w -= nh->nh_power) <= 0) { |
| 754 | nh->nh_power--; |
| 755 | fi->fib_power--; |
| 756 | - res->nh_sel = nhsel; |
| 757 | spin_unlock_bh(&fib_multipath_lock); |
| 758 | + res->nh_sel = nhsel; |
| 759 | return; |
| 760 | } |
| 761 | + alive = 1; |
| 762 | + } |
| 763 | + } endfor_nexthops(fi); |
| 764 | + if (alive) { |
| 765 | + fi->fib_power = 0; |
| 766 | + goto repeat; |
| 767 | + } |
| 768 | + |
| 769 | +last_resort: |
| 770 | + |
| 771 | + for_nexthops(fi) { |
| 772 | + if (!(nh->nh_flags&RTNH_F_DEAD)) { |
| 773 | + if (key->oif && key->oif != nh->nh_oif) |
| 774 | + continue; |
| 775 | + if (key->gw && key->gw != nh->nh_gw && |
| 776 | + nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) |
| 777 | + continue; |
| 778 | + spin_unlock_bh(&fib_multipath_lock); |
| 779 | + res->nh_sel = nhsel; |
| 780 | + return; |
| 781 | } |
| 782 | } endfor_nexthops(fi); |
| 783 | |
| 784 | /* Race condition: route has just become dead. */ |
| 785 | - res->nh_sel = 0; |
| 786 | spin_unlock_bh(&fib_multipath_lock); |
| 787 | } |
| 788 | #endif |
| 789 | --- a/net/ipv4/ip_nat_dumb.c |
| 790 | +++ b/net/ipv4/ip_nat_dumb.c |
| 791 | @@ -124,6 +124,7 @@ ip_do_nat(struct sk_buff *skb) |
| 792 | key.dst = ciph->saddr; |
| 793 | key.iif = skb->dev->ifindex; |
| 794 | key.oif = 0; |
| 795 | + key.gw = 0; |
| 796 | #ifdef CONFIG_IP_ROUTE_TOS |
| 797 | key.tos = RT_TOS(ciph->tos); |
| 798 | #endif |
| 799 | --- a/net/ipv4/netfilter/ip_fw_compat_masq.c |
| 800 | +++ b/net/ipv4/netfilter/ip_fw_compat_masq.c |
| 801 | @@ -41,6 +41,10 @@ do_masquerade(struct sk_buff **pskb, con |
| 802 | enum ip_conntrack_info ctinfo; |
| 803 | struct ip_conntrack *ct; |
| 804 | unsigned int ret; |
| 805 | + struct rtable *rt, *skb_rt; |
| 806 | + struct net_device *skb_dev; |
| 807 | + __u32 saddr; |
| 808 | + int new; |
| 809 | |
| 810 | /* Sorry, only ICMP, TCP and UDP. */ |
| 811 | if (iph->protocol != IPPROTO_ICMP |
| 812 | @@ -64,22 +68,28 @@ do_masquerade(struct sk_buff **pskb, con |
| 813 | } |
| 814 | |
| 815 | info = &ct->nat.info; |
| 816 | + iph = (*pskb)->nh.iph; |
| 817 | + saddr = iph->saddr; |
| 818 | + new = 0; |
| 819 | |
| 820 | WRITE_LOCK(&ip_nat_lock); |
| 821 | /* Setup the masquerade, if not already */ |
| 822 | if (!info->initialized) { |
| 823 | u_int32_t newsrc; |
| 824 | - struct rtable *rt; |
| 825 | struct ip_nat_multi_range range; |
| 826 | |
| 827 | + skb_rt = (struct rtable *) (*pskb)->dst; |
| 828 | + skb_dev = skb_rt->u.dst.dev; |
| 829 | /* Pass 0 instead of saddr, since it's going to be changed |
| 830 | anyway. */ |
| 831 | - if (ip_route_output(&rt, iph->daddr, 0, 0, 0) != 0) { |
| 832 | + if (ip_route_output_lookup(&rt, iph->daddr, 0, RT_TOS(iph->tos), |
| 833 | + skb_dev? skb_dev->ifindex : 0, |
| 834 | + skb_dev? skb_rt->rt_gateway : 0) != 0) { |
| 835 | + WRITE_UNLOCK(&ip_nat_lock); |
| 836 | DEBUGP("ipnat_rule_masquerade: Can't reroute.\n"); |
| 837 | return NF_DROP; |
| 838 | } |
| 839 | - newsrc = inet_select_addr(rt->u.dst.dev, rt->rt_gateway, |
| 840 | - RT_SCOPE_UNIVERSE); |
| 841 | + newsrc = rt->rt_src; |
| 842 | ip_rt_put(rt); |
| 843 | range = ((struct ip_nat_multi_range) |
| 844 | { 1, |
| 845 | @@ -92,11 +102,31 @@ do_masquerade(struct sk_buff **pskb, con |
| 846 | WRITE_UNLOCK(&ip_nat_lock); |
| 847 | return ret; |
| 848 | } |
| 849 | + new = 1; |
| 850 | } else |
| 851 | DEBUGP("Masquerading already done on this conn.\n"); |
| 852 | WRITE_UNLOCK(&ip_nat_lock); |
| 853 | |
| 854 | - return do_bindings(ct, ctinfo, info, NF_IP_POST_ROUTING, pskb); |
| 855 | + ret = do_bindings(ct, ctinfo, info, NF_IP_POST_ROUTING, pskb); |
| 856 | + if (ret != NF_ACCEPT || saddr == (*pskb)->nh.iph->saddr || new) |
| 857 | + return ret; |
| 858 | + |
| 859 | + iph = (*pskb)->nh.iph; |
| 860 | + if (ip_route_output(&rt, iph->daddr, iph->saddr, RT_TOS(iph->tos), 0) != 0) |
| 861 | + return NF_DROP; |
| 862 | + |
| 863 | + skb_rt = (struct rtable *) (*pskb)->dst; |
| 864 | + skb_dev = skb_rt->u.dst.dev; |
| 865 | + if (skb_dev != rt->u.dst.dev || rt->rt_gateway != skb_rt->rt_gateway) { |
| 866 | + if (skb_dev != rt->u.dst.dev) { |
| 867 | + /* TODO: check the new mtu and reply FRAG_NEEDED */ |
| 868 | + } |
| 869 | + dst_release((*pskb)->dst); |
| 870 | + (*pskb)->dst = &rt->u.dst; |
| 871 | + } else { |
| 872 | + ip_rt_put(rt); |
| 873 | + } |
| 874 | + return NF_ACCEPT; |
| 875 | } |
| 876 | |
| 877 | void |
| 878 | --- a/net/ipv4/netfilter/ip_nat_core.c |
| 879 | +++ b/net/ipv4/netfilter/ip_nat_core.c |
| 880 | @@ -994,6 +994,60 @@ icmp_reply_translation(struct sk_buff *s |
| 881 | return NF_ACCEPT; |
| 882 | } |
| 883 | |
| 884 | +unsigned int |
| 885 | +ip_nat_route_input(unsigned int hooknum, |
| 886 | + struct sk_buff **pskb, |
| 887 | + const struct net_device *in, |
| 888 | + const struct net_device *out, |
| 889 | + int (*okfn)(struct sk_buff *)) |
| 890 | +{ |
| 891 | + struct sk_buff *skb = *pskb; |
| 892 | + struct iphdr *iph; |
| 893 | + struct ip_conntrack *ct; |
| 894 | + enum ip_conntrack_info ctinfo; |
| 895 | + struct ip_nat_info *info; |
| 896 | + enum ip_conntrack_dir dir; |
| 897 | + __u32 saddr; |
| 898 | + int i; |
| 899 | + |
| 900 | + if (!(ct = ip_conntrack_get(skb, &ctinfo))) |
| 901 | + return NF_ACCEPT; |
| 902 | + |
| 903 | + info = &ct->nat.info; |
| 904 | + if (!info->initialized) |
| 905 | + return NF_ACCEPT; |
| 906 | + |
| 907 | + if (skb->dst) |
| 908 | + return NF_ACCEPT; |
| 909 | + |
| 910 | + if (skb->len < sizeof(struct iphdr)) |
| 911 | + return NF_ACCEPT; |
| 912 | + |
| 913 | + iph = skb->nh.iph; |
| 914 | + saddr = iph->saddr; |
| 915 | + hooknum = NF_IP_POST_ROUTING; |
| 916 | + dir = CTINFO2DIR(ctinfo); |
| 917 | + |
| 918 | + READ_LOCK(&ip_nat_lock); |
| 919 | + for (i = 0; i < info->num_manips; i++) { |
| 920 | + if (info->manips[i].direction == dir |
| 921 | + && info->manips[i].hooknum == hooknum |
| 922 | + && info->manips[i].maniptype == IP_NAT_MANIP_SRC) { |
| 923 | + saddr = info->manips[i].manip.ip; |
| 924 | + } |
| 925 | + } |
| 926 | + READ_UNLOCK(&ip_nat_lock); |
| 927 | + |
| 928 | + if (saddr == iph->saddr) |
| 929 | + return NF_ACCEPT; |
| 930 | + |
| 931 | + if (ip_route_input_lookup(skb, iph->daddr, iph->saddr, iph->tos, |
| 932 | + skb->dev, saddr)) |
| 933 | + return NF_DROP; |
| 934 | + |
| 935 | + return NF_ACCEPT; |
| 936 | +} |
| 937 | + |
| 938 | int __init ip_nat_init(void) |
| 939 | { |
| 940 | size_t i; |
| 941 | --- a/net/ipv4/netfilter/ip_nat_standalone.c |
| 942 | +++ b/net/ipv4/netfilter/ip_nat_standalone.c |
| 943 | @@ -245,6 +245,9 @@ ip_nat_local_fn(unsigned int hooknum, |
| 944 | /* Before packet filtering, change destination */ |
| 945 | static struct nf_hook_ops ip_nat_in_ops |
| 946 | = { { NULL, NULL }, ip_nat_in, PF_INET, NF_IP_PRE_ROUTING, NF_IP_PRI_NAT_DST }; |
| 947 | +/* Before routing, route before mangling */ |
| 948 | +static struct nf_hook_ops ip_nat_inr_ops |
| 949 | += { { NULL, NULL }, ip_nat_route_input, PF_INET, NF_IP_PRE_ROUTING, NF_IP_PRI_LAST-1 }; |
| 950 | /* After packet filtering, change source */ |
| 951 | static struct nf_hook_ops ip_nat_out_ops |
| 952 | = { { NULL, NULL }, ip_nat_out, PF_INET, NF_IP_POST_ROUTING, NF_IP_PRI_NAT_SRC}; |
| 953 | @@ -313,10 +316,15 @@ static int init_or_cleanup(int init) |
| 954 | printk("ip_nat_init: can't register in hook.\n"); |
| 955 | goto cleanup_nat; |
| 956 | } |
| 957 | + ret = nf_register_hook(&ip_nat_inr_ops); |
| 958 | + if (ret < 0) { |
| 959 | + printk("ip_nat_init: can't register inr hook.\n"); |
| 960 | + goto cleanup_inops; |
| 961 | + } |
| 962 | ret = nf_register_hook(&ip_nat_out_ops); |
| 963 | if (ret < 0) { |
| 964 | printk("ip_nat_init: can't register out hook.\n"); |
| 965 | - goto cleanup_inops; |
| 966 | + goto cleanup_inrops; |
| 967 | } |
| 968 | ret = nf_register_hook(&ip_nat_local_out_ops); |
| 969 | if (ret < 0) { |
| 970 | @@ -336,6 +344,8 @@ static int init_or_cleanup(int init) |
| 971 | nf_unregister_hook(&ip_nat_local_out_ops); |
| 972 | cleanup_outops: |
| 973 | nf_unregister_hook(&ip_nat_out_ops); |
| 974 | + cleanup_inrops: |
| 975 | + nf_unregister_hook(&ip_nat_inr_ops); |
| 976 | cleanup_inops: |
| 977 | nf_unregister_hook(&ip_nat_in_ops); |
| 978 | cleanup_nat: |
| 979 | --- a/net/ipv4/netfilter/ipt_MASQUERADE.c |
| 980 | +++ b/net/ipv4/netfilter/ipt_MASQUERADE.c |
| 981 | @@ -87,7 +87,8 @@ masquerade_target(struct sk_buff **pskb, |
| 982 | key.dst = (*pskb)->nh.iph->daddr; |
| 983 | key.src = 0; /* Unknown: that's what we're trying to establish */ |
| 984 | key.tos = RT_TOS((*pskb)->nh.iph->tos)|RTO_CONN; |
| 985 | - key.oif = 0; |
| 986 | + key.oif = out->ifindex; |
| 987 | + key.gw = ((struct rtable *) (*pskb)->dst)->rt_gateway; |
| 988 | #ifdef CONFIG_IP_ROUTE_FWMARK |
| 989 | key.fwmark = (*pskb)->nfmark; |
| 990 | #endif |
| 991 | @@ -98,13 +99,6 @@ masquerade_target(struct sk_buff **pskb, |
| 992 | " No route: Rusty's brain broke!\n"); |
| 993 | return NF_DROP; |
| 994 | } |
| 995 | - if (rt->u.dst.dev != out) { |
| 996 | - if (net_ratelimit()) |
| 997 | - printk("MASQUERADE:" |
| 998 | - " Route sent us somewhere else.\n"); |
| 999 | - ip_rt_put(rt); |
| 1000 | - return NF_DROP; |
| 1001 | - } |
| 1002 | |
| 1003 | newsrc = rt->rt_src; |
| 1004 | DEBUGP("newsrc = %u.%u.%u.%u\n", NIPQUAD(newsrc)); |
| 1005 | --- a/net/ipv4/route.c |
| 1006 | +++ b/net/ipv4/route.c |
| 1007 | @@ -919,6 +919,7 @@ void ip_rt_redirect(u32 old_gw, u32 dadd |
| 1008 | |
| 1009 | /* Gateway is different ... */ |
| 1010 | rt->rt_gateway = new_gw; |
| 1011 | + if (rt->key.gw) rt->key.gw = new_gw; |
| 1012 | |
| 1013 | /* Redirect received -> path was valid */ |
| 1014 | dst_confirm(&rth->u.dst); |
| 1015 | @@ -1343,6 +1344,7 @@ static int ip_route_input_mc(struct sk_b |
| 1016 | rth->key.fwmark = skb->nfmark; |
| 1017 | #endif |
| 1018 | rth->key.src = saddr; |
| 1019 | + rth->key.lsrc = 0; |
| 1020 | rth->rt_src = saddr; |
| 1021 | #ifdef CONFIG_IP_ROUTE_NAT |
| 1022 | rth->rt_dst_map = daddr; |
| 1023 | @@ -1356,6 +1358,7 @@ static int ip_route_input_mc(struct sk_b |
| 1024 | rth->u.dst.dev = &loopback_dev; |
| 1025 | dev_hold(rth->u.dst.dev); |
| 1026 | rth->key.oif = 0; |
| 1027 | + rth->key.gw = 0; |
| 1028 | rth->rt_gateway = daddr; |
| 1029 | rth->rt_spec_dst= spec_dst; |
| 1030 | rth->rt_type = RTN_MULTICAST; |
| 1031 | @@ -1395,7 +1398,7 @@ e_inval: |
| 1032 | */ |
| 1033 | |
| 1034 | int ip_route_input_slow(struct sk_buff *skb, u32 daddr, u32 saddr, |
| 1035 | - u8 tos, struct net_device *dev) |
| 1036 | + u8 tos, struct net_device *dev, u32 lsrc) |
| 1037 | { |
| 1038 | struct rt_key key; |
| 1039 | struct fib_result res; |
| 1040 | @@ -1415,16 +1418,17 @@ int ip_route_input_slow(struct sk_buff * |
| 1041 | goto out; |
| 1042 | |
| 1043 | key.dst = daddr; |
| 1044 | - key.src = saddr; |
| 1045 | + key.src = lsrc? : saddr; |
| 1046 | key.tos = tos; |
| 1047 | #ifdef CONFIG_IP_ROUTE_FWMARK |
| 1048 | key.fwmark = skb->nfmark; |
| 1049 | #endif |
| 1050 | - key.iif = dev->ifindex; |
| 1051 | + key.iif = lsrc? loopback_dev.ifindex : dev->ifindex; |
| 1052 | key.oif = 0; |
| 1053 | + key.gw = 0; |
| 1054 | key.scope = RT_SCOPE_UNIVERSE; |
| 1055 | |
| 1056 | - hash = rt_hash_code(daddr, saddr ^ (key.iif << 5), tos); |
| 1057 | + hash = rt_hash_code(daddr, saddr ^ (dev->ifindex << 5), tos); |
| 1058 | |
| 1059 | /* Check for the most weird martians, which can be not detected |
| 1060 | by fib_lookup. |
| 1061 | @@ -1445,6 +1449,12 @@ int ip_route_input_slow(struct sk_buff * |
| 1062 | if (BADCLASS(daddr) || ZERONET(daddr) || LOOPBACK(daddr)) |
| 1063 | goto martian_destination; |
| 1064 | |
| 1065 | + if (lsrc) { |
| 1066 | + if (MULTICAST(lsrc) || BADCLASS(lsrc) || |
| 1067 | + ZERONET(lsrc) || LOOPBACK(lsrc)) |
| 1068 | + goto e_inval; |
| 1069 | + } |
| 1070 | + |
| 1071 | /* |
| 1072 | * Now we are ready to route packet. |
| 1073 | */ |
| 1074 | @@ -1454,6 +1464,10 @@ int ip_route_input_slow(struct sk_buff * |
| 1075 | goto no_route; |
| 1076 | } |
| 1077 | free_res = 1; |
| 1078 | + if (lsrc && res.type != RTN_UNICAST && res.type != RTN_NAT) |
| 1079 | + goto e_inval; |
| 1080 | + key.iif = dev->ifindex; |
| 1081 | + key.src = saddr; |
| 1082 | |
| 1083 | rt_cache_stat[smp_processor_id()].in_slow_tot++; |
| 1084 | |
| 1085 | @@ -1464,7 +1478,7 @@ int ip_route_input_slow(struct sk_buff * |
| 1086 | |
| 1087 | if (1) { |
| 1088 | u32 src_map = saddr; |
| 1089 | - if (res.r) |
| 1090 | + if (res.r && !lsrc) |
| 1091 | src_map = fib_rules_policy(saddr, &res, &flags); |
| 1092 | |
| 1093 | if (res.type == RTN_NAT) { |
| 1094 | @@ -1503,8 +1517,9 @@ int ip_route_input_slow(struct sk_buff * |
| 1095 | if (res.type != RTN_UNICAST) |
| 1096 | goto martian_destination; |
| 1097 | |
| 1098 | + fib_select_default(&key, &res); |
| 1099 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 1100 | - if (res.fi->fib_nhs > 1 && key.oif == 0) |
| 1101 | + if (res.fi->fib_nhs > 1) |
| 1102 | fib_select_multipath(&key, &res); |
| 1103 | #endif |
| 1104 | out_dev = in_dev_get(FIB_RES_DEV(res)); |
| 1105 | @@ -1524,6 +1539,7 @@ int ip_route_input_slow(struct sk_buff * |
| 1106 | flags |= RTCF_DIRECTSRC; |
| 1107 | |
| 1108 | if (out_dev == in_dev && err && !(flags & (RTCF_NAT | RTCF_MASQ)) && |
| 1109 | + !lsrc && |
| 1110 | (IN_DEV_SHARED_MEDIA(out_dev) || |
| 1111 | inet_addr_onlink(out_dev, saddr, FIB_RES_GW(res)))) |
| 1112 | flags |= RTCF_DOREDIRECT; |
| 1113 | @@ -1550,6 +1566,7 @@ int ip_route_input_slow(struct sk_buff * |
| 1114 | #endif |
| 1115 | rth->key.src = saddr; |
| 1116 | rth->rt_src = saddr; |
| 1117 | + rth->key.lsrc = lsrc; |
| 1118 | rth->rt_gateway = daddr; |
| 1119 | #ifdef CONFIG_IP_ROUTE_NAT |
| 1120 | rth->rt_src_map = key.src; |
| 1121 | @@ -1562,6 +1579,7 @@ int ip_route_input_slow(struct sk_buff * |
| 1122 | rth->u.dst.dev = out_dev->dev; |
| 1123 | dev_hold(rth->u.dst.dev); |
| 1124 | rth->key.oif = 0; |
| 1125 | + rth->key.gw = 0; |
| 1126 | rth->rt_spec_dst= spec_dst; |
| 1127 | |
| 1128 | rth->u.dst.input = ip_forward; |
| 1129 | @@ -1572,7 +1590,8 @@ int ip_route_input_slow(struct sk_buff * |
| 1130 | rth->rt_flags = flags; |
| 1131 | |
| 1132 | #ifdef CONFIG_NET_FASTROUTE |
| 1133 | - if (netdev_fastroute && !(flags&(RTCF_NAT|RTCF_MASQ|RTCF_DOREDIRECT))) { |
| 1134 | + if (netdev_fastroute && !(flags&(RTCF_NAT|RTCF_MASQ|RTCF_DOREDIRECT)) && |
| 1135 | + !lsrc) { |
| 1136 | struct net_device *odev = rth->u.dst.dev; |
| 1137 | if (odev != dev && |
| 1138 | dev->accept_fastpath && |
| 1139 | @@ -1595,6 +1614,8 @@ out: return err; |
| 1140 | brd_input: |
| 1141 | if (skb->protocol != htons(ETH_P_IP)) |
| 1142 | goto e_inval; |
| 1143 | + if (lsrc) |
| 1144 | + goto e_inval; |
| 1145 | |
| 1146 | if (ZERONET(saddr)) |
| 1147 | spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK); |
| 1148 | @@ -1627,6 +1648,7 @@ local_input: |
| 1149 | #endif |
| 1150 | rth->key.src = saddr; |
| 1151 | rth->rt_src = saddr; |
| 1152 | + rth->key.lsrc = 0; |
| 1153 | #ifdef CONFIG_IP_ROUTE_NAT |
| 1154 | rth->rt_dst_map = key.dst; |
| 1155 | rth->rt_src_map = key.src; |
| 1156 | @@ -1639,6 +1661,7 @@ local_input: |
| 1157 | rth->u.dst.dev = &loopback_dev; |
| 1158 | dev_hold(rth->u.dst.dev); |
| 1159 | rth->key.oif = 0; |
| 1160 | + rth->key.gw = 0; |
| 1161 | rth->rt_gateway = daddr; |
| 1162 | rth->rt_spec_dst= spec_dst; |
| 1163 | rth->u.dst.input= ip_local_deliver; |
| 1164 | @@ -1704,8 +1727,9 @@ martian_source: |
| 1165 | goto e_inval; |
| 1166 | } |
| 1167 | |
| 1168 | -int ip_route_input(struct sk_buff *skb, u32 daddr, u32 saddr, |
| 1169 | - u8 tos, struct net_device *dev) |
| 1170 | +static inline int |
| 1171 | +ip_route_input_cached(struct sk_buff *skb, u32 daddr, u32 saddr, |
| 1172 | + u8 tos, struct net_device *dev, u32 lsrc) |
| 1173 | { |
| 1174 | struct rtable * rth; |
| 1175 | unsigned hash; |
| 1176 | @@ -1719,6 +1743,7 @@ int ip_route_input(struct sk_buff *skb, |
| 1177 | if (rth->key.dst == daddr && |
| 1178 | rth->key.src == saddr && |
| 1179 | rth->key.iif == iif && |
| 1180 | + rth->key.lsrc == lsrc && |
| 1181 | rth->key.oif == 0 && |
| 1182 | #ifdef CONFIG_IP_ROUTE_FWMARK |
| 1183 | rth->key.fwmark == skb->nfmark && |
| 1184 | @@ -1766,9 +1791,21 @@ int ip_route_input(struct sk_buff *skb, |
| 1185 | read_unlock(&inetdev_lock); |
| 1186 | return -EINVAL; |
| 1187 | } |
| 1188 | - return ip_route_input_slow(skb, daddr, saddr, tos, dev); |
| 1189 | + return ip_route_input_slow(skb, daddr, saddr, tos, dev, lsrc); |
| 1190 | +} |
| 1191 | + |
| 1192 | +int ip_route_input(struct sk_buff *skb, u32 daddr, u32 saddr, |
| 1193 | + u8 tos, struct net_device *dev) |
| 1194 | +{ |
| 1195 | + return ip_route_input_cached(skb, daddr, saddr, tos, dev, 0); |
| 1196 | } |
| 1197 | |
| 1198 | +int ip_route_input_lookup(struct sk_buff *skb, u32 daddr, u32 saddr, |
| 1199 | + u8 tos, struct net_device *dev, u32 lsrc) |
| 1200 | +{ |
| 1201 | + return ip_route_input_cached(skb, daddr, saddr, tos, dev, lsrc); |
| 1202 | +} |
| 1203 | + |
| 1204 | /* |
| 1205 | * Major route resolver routine. |
| 1206 | */ |
| 1207 | @@ -1791,6 +1828,7 @@ int ip_route_output_slow(struct rtable * |
| 1208 | key.tos = tos & IPTOS_RT_MASK; |
| 1209 | key.iif = loopback_dev.ifindex; |
| 1210 | key.oif = oldkey->oif; |
| 1211 | + key.gw = oldkey->gw; |
| 1212 | #ifdef CONFIG_IP_ROUTE_FWMARK |
| 1213 | key.fwmark = oldkey->fwmark; |
| 1214 | #endif |
| 1215 | @@ -1880,6 +1918,7 @@ int ip_route_output_slow(struct rtable * |
| 1216 | dev_out = &loopback_dev; |
| 1217 | dev_hold(dev_out); |
| 1218 | key.oif = loopback_dev.ifindex; |
| 1219 | + key.gw = 0; |
| 1220 | res.type = RTN_LOCAL; |
| 1221 | flags |= RTCF_LOCAL; |
| 1222 | goto make_route; |
| 1223 | @@ -1887,7 +1926,7 @@ int ip_route_output_slow(struct rtable * |
| 1224 | |
| 1225 | if (fib_lookup(&key, &res)) { |
| 1226 | res.fi = NULL; |
| 1227 | - if (oldkey->oif) { |
| 1228 | + if (oldkey->oif && dev_out->flags&IFF_UP) { |
| 1229 | /* Apparently, routing tables are wrong. Assume, |
| 1230 | that the destination is on link. |
| 1231 | |
| 1232 | @@ -1930,6 +1969,7 @@ int ip_route_output_slow(struct rtable * |
| 1233 | dev_out = &loopback_dev; |
| 1234 | dev_hold(dev_out); |
| 1235 | key.oif = dev_out->ifindex; |
| 1236 | + key.gw = 0; |
| 1237 | if (res.fi) |
| 1238 | fib_info_put(res.fi); |
| 1239 | res.fi = NULL; |
| 1240 | @@ -1937,13 +1977,12 @@ int ip_route_output_slow(struct rtable * |
| 1241 | goto make_route; |
| 1242 | } |
| 1243 | |
| 1244 | + if (res.type == RTN_UNICAST) |
| 1245 | + fib_select_default(&key, &res); |
| 1246 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
| 1247 | - if (res.fi->fib_nhs > 1 && key.oif == 0) |
| 1248 | + if (res.fi->fib_nhs > 1) |
| 1249 | fib_select_multipath(&key, &res); |
| 1250 | - else |
| 1251 | #endif |
| 1252 | - if (!res.prefixlen && res.type == RTN_UNICAST && !key.oif) |
| 1253 | - fib_select_default(&key, &res); |
| 1254 | |
| 1255 | if (!key.src) |
| 1256 | key.src = FIB_RES_PREFSRC(res); |
| 1257 | @@ -2001,7 +2040,9 @@ make_route: |
| 1258 | rth->key.tos = tos; |
| 1259 | rth->key.src = oldkey->src; |
| 1260 | rth->key.iif = 0; |
| 1261 | + rth->key.lsrc = 0; |
| 1262 | rth->key.oif = oldkey->oif; |
| 1263 | + rth->key.gw = oldkey->gw; |
| 1264 | #ifdef CONFIG_IP_ROUTE_FWMARK |
| 1265 | rth->key.fwmark = oldkey->fwmark; |
| 1266 | #endif |
| 1267 | @@ -2080,6 +2121,7 @@ int ip_route_output_key(struct rtable ** |
| 1268 | rth->key.src == key->src && |
| 1269 | rth->key.iif == 0 && |
| 1270 | rth->key.oif == key->oif && |
| 1271 | + rth->key.gw == key->gw && |
| 1272 | #ifdef CONFIG_IP_ROUTE_FWMARK |
| 1273 | rth->key.fwmark == key->fwmark && |
| 1274 | #endif |
| 1275 | --- a/net/netsyms.c |
| 1276 | +++ b/net/netsyms.c |
| 1277 | @@ -260,6 +260,7 @@ EXPORT_SYMBOL(inet_register_protosw); |
| 1278 | EXPORT_SYMBOL(inet_unregister_protosw); |
| 1279 | EXPORT_SYMBOL(ip_route_output_key); |
| 1280 | EXPORT_SYMBOL(ip_route_input); |
| 1281 | +EXPORT_SYMBOL(ip_route_input_lookup); |
| 1282 | EXPORT_SYMBOL(icmp_send); |
| 1283 | EXPORT_SYMBOL(icmp_statistics); |
| 1284 | EXPORT_SYMBOL(icmp_err_convert); |
| 1285 | |
