1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Linux INET6 implementation 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 */ 8 9 #ifndef _IP6_FIB_H 10 #define _IP6_FIB_H 11 12 #include <linux/ipv6_route.h> 13 #include <linux/rtnetlink.h> 14 #include <linux/spinlock.h> 15 #include <linux/notifier.h> 16 #include <net/dst.h> 17 #include <net/flow.h> 18 #include <net/ip_fib.h> 19 #include <net/netlink.h> 20 #include <net/inetpeer.h> 21 #include <net/fib_notifier.h> 22 #include <linux/indirect_call_wrapper.h> 23 #include <uapi/linux/bpf.h> 24 25 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 26 #define FIB6_TABLE_HASHSZ 256 27 #else 28 #define FIB6_TABLE_HASHSZ 1 29 #endif 30 31 #define RT6_DEBUG 2 32 33 #if RT6_DEBUG >= 3 34 #define RT6_TRACE(x...) pr_debug(x) 35 #else 36 #define RT6_TRACE(x...) do { ; } while (0) 37 #endif 38 39 struct rt6_info; 40 struct fib6_info; 41 42 struct fib6_config { 43 u32 fc_table; 44 u32 fc_metric; 45 int fc_dst_len; 46 int fc_src_len; 47 int fc_ifindex; 48 u32 fc_flags; 49 u32 fc_protocol; 50 u16 fc_type; /* only 8 bits are used */ 51 u16 fc_delete_all_nh : 1, 52 fc_ignore_dev_down:1, 53 __unused : 14; 54 u32 fc_nh_id; 55 56 struct in6_addr fc_dst; 57 struct in6_addr fc_src; 58 struct in6_addr fc_prefsrc; 59 struct in6_addr fc_gateway; 60 61 unsigned long fc_expires; 62 struct nlattr *fc_mx; 63 int fc_mx_len; 64 int fc_mp_len; 65 struct nlattr *fc_mp; 66 67 struct nl_info fc_nlinfo; 68 struct nlattr *fc_encap; 69 u16 fc_encap_type; 70 bool fc_is_fdb; 71 }; 72 73 struct fib6_node { 74 struct fib6_node __rcu *parent; 75 struct fib6_node __rcu *left; 76 struct fib6_node __rcu *right; 77 #ifdef CONFIG_IPV6_SUBTREES 78 struct fib6_node __rcu *subtree; 79 #endif 80 struct fib6_info __rcu *leaf; 81 82 __u16 fn_bit; /* bit key */ 83 __u16 fn_flags; 84 int fn_sernum; 85 struct fib6_info __rcu *rr_ptr; 86 struct rcu_head rcu; 87 }; 88 89 struct fib6_gc_args { 90 int timeout; 91 int more; 92 }; 93 94 #ifndef CONFIG_IPV6_SUBTREES 95 #define FIB6_SUBTREE(fn) NULL 96 fib6_routes_require_src(const struct net * net)97 static inline bool fib6_routes_require_src(const struct net *net) 98 { 99 return false; 100 } 101 fib6_routes_require_src_inc(struct net * net)102 static inline void fib6_routes_require_src_inc(struct net *net) {} fib6_routes_require_src_dec(struct net * net)103 static inline void fib6_routes_require_src_dec(struct net *net) {} 104 105 #else 106 fib6_routes_require_src(const struct net * net)107 static inline bool fib6_routes_require_src(const struct net *net) 108 { 109 return net->ipv6.fib6_routes_require_src > 0; 110 } 111 fib6_routes_require_src_inc(struct net * net)112 static inline void fib6_routes_require_src_inc(struct net *net) 113 { 114 net->ipv6.fib6_routes_require_src++; 115 } 116 fib6_routes_require_src_dec(struct net * net)117 static inline void fib6_routes_require_src_dec(struct net *net) 118 { 119 net->ipv6.fib6_routes_require_src--; 120 } 121 122 #define FIB6_SUBTREE(fn) (rcu_dereference_protected((fn)->subtree, 1)) 123 #endif 124 125 /* 126 * routing information 127 * 128 */ 129 130 struct rt6key { 131 struct in6_addr addr; 132 int plen; 133 }; 134 135 struct fib6_table; 136 137 struct rt6_exception_bucket { 138 struct hlist_head chain; 139 int depth; 140 }; 141 142 struct rt6_exception { 143 struct hlist_node hlist; 144 struct rt6_info *rt6i; 145 unsigned long stamp; 146 struct rcu_head rcu; 147 }; 148 149 #define FIB6_EXCEPTION_BUCKET_SIZE_SHIFT 10 150 #define FIB6_EXCEPTION_BUCKET_SIZE (1 << FIB6_EXCEPTION_BUCKET_SIZE_SHIFT) 151 #define FIB6_MAX_DEPTH 5 152 153 struct fib6_nh { 154 struct fib_nh_common nh_common; 155 156 #ifdef CONFIG_IPV6_ROUTER_PREF 157 unsigned long last_probe; 158 #endif 159 160 struct rt6_info * __percpu *rt6i_pcpu; 161 struct rt6_exception_bucket __rcu *rt6i_exception_bucket; 162 }; 163 164 struct fib6_info { 165 struct fib6_table *fib6_table; 166 struct fib6_info __rcu *fib6_next; 167 struct fib6_node __rcu *fib6_node; 168 169 /* Multipath routes: 170 * siblings is a list of fib6_info that have the same metric/weight, 171 * destination, but not the same gateway. nsiblings is just a cache 172 * to speed up lookup. 173 */ 174 union { 175 struct list_head fib6_siblings; 176 struct list_head nh_list; 177 }; 178 unsigned int fib6_nsiblings; 179 180 refcount_t fib6_ref; 181 unsigned long expires; 182 struct dst_metrics *fib6_metrics; 183 #define fib6_pmtu fib6_metrics->metrics[RTAX_MTU-1] 184 185 struct rt6key fib6_dst; 186 u32 fib6_flags; 187 struct rt6key fib6_src; 188 struct rt6key fib6_prefsrc; 189 190 u32 fib6_metric; 191 u8 fib6_protocol; 192 u8 fib6_type; 193 194 u8 offload; 195 u8 trap; 196 u8 offload_failed; 197 198 u8 should_flush:1, 199 dst_nocount:1, 200 dst_nopolicy:1, 201 fib6_destroying:1, 202 unused:4; 203 204 struct rcu_head rcu; 205 struct nexthop *nh; 206 struct fib6_nh fib6_nh[]; 207 }; 208 209 struct rt6_info { 210 struct dst_entry dst; 211 struct fib6_info __rcu *from; 212 int sernum; 213 214 struct rt6key rt6i_dst; 215 struct rt6key rt6i_src; 216 struct in6_addr rt6i_gateway; 217 struct inet6_dev *rt6i_idev; 218 u32 rt6i_flags; 219 220 /* more non-fragment space at head required */ 221 unsigned short rt6i_nfheader_len; 222 }; 223 224 struct fib6_result { 225 struct fib6_nh *nh; 226 struct fib6_info *f6i; 227 u32 fib6_flags; 228 u8 fib6_type; 229 struct rt6_info *rt6; 230 }; 231 232 #define for_each_fib6_node_rt_rcu(fn) \ 233 for (rt = rcu_dereference((fn)->leaf); rt; \ 234 rt = rcu_dereference(rt->fib6_next)) 235 236 #define for_each_fib6_walker_rt(w) \ 237 for (rt = (w)->leaf; rt; \ 238 rt = rcu_dereference_protected(rt->fib6_next, 1)) 239 240 #define dst_rt6_info(_ptr) container_of_const(_ptr, struct rt6_info, dst) 241 ip6_dst_idev(const struct dst_entry * dst)242 static inline struct inet6_dev *ip6_dst_idev(const struct dst_entry *dst) 243 { 244 return dst_rt6_info(dst)->rt6i_idev; 245 } 246 fib6_requires_src(const struct fib6_info * rt)247 static inline bool fib6_requires_src(const struct fib6_info *rt) 248 { 249 return rt->fib6_src.plen > 0; 250 } 251 fib6_clean_expires(struct fib6_info * f6i)252 static inline void fib6_clean_expires(struct fib6_info *f6i) 253 { 254 f6i->fib6_flags &= ~RTF_EXPIRES; 255 f6i->expires = 0; 256 } 257 fib6_set_expires(struct fib6_info * f6i,unsigned long expires)258 static inline void fib6_set_expires(struct fib6_info *f6i, 259 unsigned long expires) 260 { 261 f6i->expires = expires; 262 f6i->fib6_flags |= RTF_EXPIRES; 263 } 264 fib6_check_expired(const struct fib6_info * f6i)265 static inline bool fib6_check_expired(const struct fib6_info *f6i) 266 { 267 if (f6i->fib6_flags & RTF_EXPIRES) 268 return time_after(jiffies, f6i->expires); 269 return false; 270 } 271 272 /* Function to safely get fn->fn_sernum for passed in rt 273 * and store result in passed in cookie. 274 * Return true if we can get cookie safely 275 * Return false if not 276 */ fib6_get_cookie_safe(const struct fib6_info * f6i,u32 * cookie)277 static inline bool fib6_get_cookie_safe(const struct fib6_info *f6i, 278 u32 *cookie) 279 { 280 struct fib6_node *fn; 281 bool status = false; 282 283 fn = rcu_dereference(f6i->fib6_node); 284 285 if (fn) { 286 *cookie = READ_ONCE(fn->fn_sernum); 287 /* pairs with smp_wmb() in __fib6_update_sernum_upto_root() */ 288 smp_rmb(); 289 status = true; 290 } 291 292 return status; 293 } 294 rt6_get_cookie(const struct rt6_info * rt)295 static inline u32 rt6_get_cookie(const struct rt6_info *rt) 296 { 297 struct fib6_info *from; 298 u32 cookie = 0; 299 300 if (rt->sernum) 301 return rt->sernum; 302 303 rcu_read_lock(); 304 305 from = rcu_dereference(rt->from); 306 if (from) 307 fib6_get_cookie_safe(from, &cookie); 308 309 rcu_read_unlock(); 310 311 return cookie; 312 } 313 ip6_rt_put(struct rt6_info * rt)314 static inline void ip6_rt_put(struct rt6_info *rt) 315 { 316 /* dst_release() accepts a NULL parameter. 317 * We rely on dst being first structure in struct rt6_info 318 */ 319 BUILD_BUG_ON(offsetof(struct rt6_info, dst) != 0); 320 dst_release(&rt->dst); 321 } 322 323 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh); 324 void fib6_info_destroy_rcu(struct rcu_head *head); 325 fib6_info_hold(struct fib6_info * f6i)326 static inline void fib6_info_hold(struct fib6_info *f6i) 327 { 328 refcount_inc(&f6i->fib6_ref); 329 } 330 fib6_info_hold_safe(struct fib6_info * f6i)331 static inline bool fib6_info_hold_safe(struct fib6_info *f6i) 332 { 333 return refcount_inc_not_zero(&f6i->fib6_ref); 334 } 335 fib6_info_release(struct fib6_info * f6i)336 static inline void fib6_info_release(struct fib6_info *f6i) 337 { 338 if (f6i && refcount_dec_and_test(&f6i->fib6_ref)) 339 call_rcu(&f6i->rcu, fib6_info_destroy_rcu); 340 } 341 342 enum fib6_walk_state { 343 #ifdef CONFIG_IPV6_SUBTREES 344 FWS_S, 345 #endif 346 FWS_L, 347 FWS_R, 348 FWS_C, 349 FWS_U 350 }; 351 352 struct fib6_walker { 353 struct list_head lh; 354 struct fib6_node *root, *node; 355 struct fib6_info *leaf; 356 enum fib6_walk_state state; 357 unsigned int skip; 358 unsigned int count; 359 unsigned int skip_in_node; 360 int (*func)(struct fib6_walker *); 361 void *args; 362 }; 363 364 struct rt6_statistics { 365 __u32 fib_nodes; /* all fib6 nodes */ 366 __u32 fib_route_nodes; /* intermediate nodes */ 367 __u32 fib_rt_entries; /* rt entries in fib table */ 368 __u32 fib_rt_cache; /* cached rt entries in exception table */ 369 __u32 fib_discarded_routes; /* total number of routes delete */ 370 371 /* The following stat is not protected by any lock */ 372 atomic_t fib_rt_alloc; /* total number of routes alloced */ 373 }; 374 375 #define RTN_TL_ROOT 0x0001 376 #define RTN_ROOT 0x0002 /* tree root node */ 377 #define RTN_RTINFO 0x0004 /* node with valid routing info */ 378 379 /* 380 * priority levels (or metrics) 381 * 382 */ 383 384 385 struct fib6_table { 386 struct hlist_node tb6_hlist; 387 u32 tb6_id; 388 spinlock_t tb6_lock; 389 struct fib6_node tb6_root; 390 struct inet_peer_base tb6_peers; 391 unsigned int flags; 392 unsigned int fib_seq; 393 #define RT6_TABLE_HAS_DFLT_ROUTER BIT(0) 394 }; 395 396 #define RT6_TABLE_UNSPEC RT_TABLE_UNSPEC 397 #define RT6_TABLE_MAIN RT_TABLE_MAIN 398 #define RT6_TABLE_DFLT RT6_TABLE_MAIN 399 #define RT6_TABLE_INFO RT6_TABLE_MAIN 400 #define RT6_TABLE_PREFIX RT6_TABLE_MAIN 401 402 #ifdef CONFIG_IPV6_MULTIPLE_TABLES 403 #define FIB6_TABLE_MIN 1 404 #define FIB6_TABLE_MAX RT_TABLE_MAX 405 #define RT6_TABLE_LOCAL RT_TABLE_LOCAL 406 #else 407 #define FIB6_TABLE_MIN RT_TABLE_MAIN 408 #define FIB6_TABLE_MAX FIB6_TABLE_MIN 409 #define RT6_TABLE_LOCAL RT6_TABLE_MAIN 410 #endif 411 412 typedef struct rt6_info *(*pol_lookup_t)(struct net *, 413 struct fib6_table *, 414 struct flowi6 *, 415 const struct sk_buff *, int); 416 417 struct fib6_entry_notifier_info { 418 struct fib_notifier_info info; /* must be first */ 419 struct fib6_info *rt; 420 unsigned int nsiblings; 421 }; 422 423 /* 424 * exported functions 425 */ 426 427 struct fib6_table *fib6_get_table(struct net *net, u32 id); 428 struct fib6_table *fib6_new_table(struct net *net, u32 id); 429 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 430 const struct sk_buff *skb, 431 int flags, pol_lookup_t lookup); 432 433 /* called with rcu lock held; can return error pointer 434 * caller needs to select path 435 */ 436 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6, 437 struct fib6_result *res, int flags); 438 439 /* called with rcu lock held; caller needs to select path */ 440 int fib6_table_lookup(struct net *net, struct fib6_table *table, 441 int oif, struct flowi6 *fl6, struct fib6_result *res, 442 int strict); 443 444 void fib6_select_path(const struct net *net, struct fib6_result *res, 445 struct flowi6 *fl6, int oif, bool have_oif_match, 446 const struct sk_buff *skb, int strict); 447 struct fib6_node *fib6_node_lookup(struct fib6_node *root, 448 const struct in6_addr *daddr, 449 const struct in6_addr *saddr); 450 451 struct fib6_node *fib6_locate(struct fib6_node *root, 452 const struct in6_addr *daddr, int dst_len, 453 const struct in6_addr *saddr, int src_len, 454 bool exact_match); 455 456 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *arg), 457 void *arg); 458 void fib6_clean_all_skip_notify(struct net *net, 459 int (*func)(struct fib6_info *, void *arg), 460 void *arg); 461 462 int fib6_add(struct fib6_node *root, struct fib6_info *rt, 463 struct nl_info *info, struct netlink_ext_ack *extack); 464 int fib6_del(struct fib6_info *rt, struct nl_info *info); 465 466 static inline rt6_get_prefsrc(const struct rt6_info * rt,struct in6_addr * addr)467 void rt6_get_prefsrc(const struct rt6_info *rt, struct in6_addr *addr) 468 { 469 const struct fib6_info *from; 470 471 rcu_read_lock(); 472 473 from = rcu_dereference(rt->from); 474 if (from) 475 *addr = from->fib6_prefsrc.addr; 476 else 477 *addr = in6addr_any; 478 479 rcu_read_unlock(); 480 } 481 482 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh, 483 struct fib6_config *cfg, gfp_t gfp_flags, 484 struct netlink_ext_ack *extack); 485 void fib6_nh_release(struct fib6_nh *fib6_nh); 486 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh); 487 488 int call_fib6_entry_notifiers(struct net *net, 489 enum fib_event_type event_type, 490 struct fib6_info *rt, 491 struct netlink_ext_ack *extack); 492 int call_fib6_multipath_entry_notifiers(struct net *net, 493 enum fib_event_type event_type, 494 struct fib6_info *rt, 495 unsigned int nsiblings, 496 struct netlink_ext_ack *extack); 497 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt); 498 void fib6_rt_update(struct net *net, struct fib6_info *rt, 499 struct nl_info *info); 500 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info, 501 unsigned int flags); 502 503 void fib6_run_gc(unsigned long expires, struct net *net, bool force); 504 505 void fib6_gc_cleanup(void); 506 507 int fib6_init(void); 508 509 struct ipv6_route_iter { 510 struct seq_net_private p; 511 struct fib6_walker w; 512 loff_t skip; 513 struct fib6_table *tbl; 514 int sernum; 515 }; 516 517 extern const struct seq_operations ipv6_route_seq_ops; 518 519 int call_fib6_notifier(struct notifier_block *nb, 520 enum fib_event_type event_type, 521 struct fib_notifier_info *info); 522 int call_fib6_notifiers(struct net *net, enum fib_event_type event_type, 523 struct fib_notifier_info *info); 524 525 int __net_init fib6_notifier_init(struct net *net); 526 void __net_exit fib6_notifier_exit(struct net *net); 527 528 unsigned int fib6_tables_seq_read(struct net *net); 529 int fib6_tables_dump(struct net *net, struct notifier_block *nb, 530 struct netlink_ext_ack *extack); 531 532 void fib6_update_sernum(struct net *net, struct fib6_info *rt); 533 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt); 534 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i); 535 536 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val); fib6_metric_locked(struct fib6_info * f6i,int metric)537 static inline bool fib6_metric_locked(struct fib6_info *f6i, int metric) 538 { 539 return !!(f6i->fib6_metrics->metrics[RTAX_LOCK - 1] & (1 << metric)); 540 } 541 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i, 542 bool offload, bool trap, bool offload_failed); 543 544 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL) 545 struct bpf_iter__ipv6_route { 546 __bpf_md_ptr(struct bpf_iter_meta *, meta); 547 __bpf_md_ptr(struct fib6_info *, rt); 548 }; 549 #endif 550 551 INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_output(struct net *net, 552 struct fib6_table *table, 553 struct flowi6 *fl6, 554 const struct sk_buff *skb, 555 int flags)); 556 INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_input(struct net *net, 557 struct fib6_table *table, 558 struct flowi6 *fl6, 559 const struct sk_buff *skb, 560 int flags)); 561 INDIRECT_CALLABLE_DECLARE(struct rt6_info *__ip6_route_redirect(struct net *net, 562 struct fib6_table *table, 563 struct flowi6 *fl6, 564 const struct sk_buff *skb, 565 int flags)); 566 INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_lookup(struct net *net, 567 struct fib6_table *table, 568 struct flowi6 *fl6, 569 const struct sk_buff *skb, 570 int flags)); pol_lookup_func(pol_lookup_t lookup,struct net * net,struct fib6_table * table,struct flowi6 * fl6,const struct sk_buff * skb,int flags)571 static inline struct rt6_info *pol_lookup_func(pol_lookup_t lookup, 572 struct net *net, 573 struct fib6_table *table, 574 struct flowi6 *fl6, 575 const struct sk_buff *skb, 576 int flags) 577 { 578 return INDIRECT_CALL_4(lookup, 579 ip6_pol_route_output, 580 ip6_pol_route_input, 581 ip6_pol_route_lookup, 582 __ip6_route_redirect, 583 net, table, fl6, skb, flags); 584 } 585 586 #ifdef CONFIG_IPV6_MULTIPLE_TABLES fib6_has_custom_rules(const struct net * net)587 static inline bool fib6_has_custom_rules(const struct net *net) 588 { 589 return net->ipv6.fib6_has_custom_rules; 590 } 591 592 int fib6_rules_init(void); 593 void fib6_rules_cleanup(void); 594 bool fib6_rule_default(const struct fib_rule *rule); 595 int fib6_rules_dump(struct net *net, struct notifier_block *nb, 596 struct netlink_ext_ack *extack); 597 unsigned int fib6_rules_seq_read(struct net *net); 598 fib6_rules_early_flow_dissect(struct net * net,struct sk_buff * skb,struct flowi6 * fl6,struct flow_keys * flkeys)599 static inline bool fib6_rules_early_flow_dissect(struct net *net, 600 struct sk_buff *skb, 601 struct flowi6 *fl6, 602 struct flow_keys *flkeys) 603 { 604 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 605 606 if (!net->ipv6.fib6_rules_require_fldissect) 607 return false; 608 609 memset(flkeys, 0, sizeof(*flkeys)); 610 __skb_flow_dissect(net, skb, &flow_keys_dissector, 611 flkeys, NULL, 0, 0, 0, flag); 612 613 fl6->fl6_sport = flkeys->ports.src; 614 fl6->fl6_dport = flkeys->ports.dst; 615 fl6->flowi6_proto = flkeys->basic.ip_proto; 616 617 return true; 618 } 619 #else fib6_has_custom_rules(const struct net * net)620 static inline bool fib6_has_custom_rules(const struct net *net) 621 { 622 return false; 623 } fib6_rules_init(void)624 static inline int fib6_rules_init(void) 625 { 626 return 0; 627 } fib6_rules_cleanup(void)628 static inline void fib6_rules_cleanup(void) 629 { 630 return ; 631 } fib6_rule_default(const struct fib_rule * rule)632 static inline bool fib6_rule_default(const struct fib_rule *rule) 633 { 634 return true; 635 } fib6_rules_dump(struct net * net,struct notifier_block * nb,struct netlink_ext_ack * extack)636 static inline int fib6_rules_dump(struct net *net, struct notifier_block *nb, 637 struct netlink_ext_ack *extack) 638 { 639 return 0; 640 } fib6_rules_seq_read(struct net * net)641 static inline unsigned int fib6_rules_seq_read(struct net *net) 642 { 643 return 0; 644 } fib6_rules_early_flow_dissect(struct net * net,struct sk_buff * skb,struct flowi6 * fl6,struct flow_keys * flkeys)645 static inline bool fib6_rules_early_flow_dissect(struct net *net, 646 struct sk_buff *skb, 647 struct flowi6 *fl6, 648 struct flow_keys *flkeys) 649 { 650 return false; 651 } 652 #endif 653 #endif 654