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 33struct rt6_info; 34struct fib6_info; 35 36struct fib6_config { 37 u32 fc_table; 38 u32 fc_metric; 39 int fc_dst_len; 40 int fc_src_len; 41 int fc_ifindex; 42 u32 fc_flags; 43 u32 fc_protocol; 44 u16 fc_type; /* only 8 bits are used */ 45 u16 fc_delete_all_nh : 1, 46 fc_ignore_dev_down:1, 47 __unused : 14; 48 u32 fc_nh_id; 49 50 struct in6_addr fc_dst; 51 struct in6_addr fc_src; 52 struct in6_addr fc_prefsrc; 53 struct in6_addr fc_gateway; 54 55 unsigned long fc_expires; 56 struct nlattr *fc_mx; 57 int fc_mx_len; 58 int fc_mp_len; 59 struct nlattr *fc_mp; 60 61 struct nl_info fc_nlinfo; 62 struct nlattr *fc_encap; 63 u16 fc_encap_type; 64 bool fc_is_fdb; 65}; 66 67struct fib6_node { 68 struct fib6_node __rcu *parent; 69 struct fib6_node __rcu *left; 70 struct fib6_node __rcu *right; 71#ifdef CONFIG_IPV6_SUBTREES 72 struct fib6_node __rcu *subtree; 73#endif 74 struct fib6_info __rcu *leaf; 75 76 __u16 fn_bit; /* bit key */ 77 __u16 fn_flags; 78 int fn_sernum; 79 struct fib6_info __rcu *rr_ptr; 80 struct rcu_head rcu; 81}; 82 83struct fib6_gc_args { 84 int timeout; 85 int more; 86}; 87 88#ifndef CONFIG_IPV6_SUBTREES 89#define FIB6_SUBTREE(fn) NULL 90 91static inline bool fib6_routes_require_src(const struct net *net) 92{ 93 return false; 94} 95 96static inline void fib6_routes_require_src_inc(struct net *net) {} 97static inline void fib6_routes_require_src_dec(struct net *net) {} 98 99#else 100 101static inline bool fib6_routes_require_src(const struct net *net) 102{ 103 return net->ipv6.fib6_routes_require_src > 0; 104} 105 106static inline void fib6_routes_require_src_inc(struct net *net) 107{ 108 net->ipv6.fib6_routes_require_src++; 109} 110 111static inline void fib6_routes_require_src_dec(struct net *net) 112{ 113 net->ipv6.fib6_routes_require_src--; 114} 115 116#define FIB6_SUBTREE(fn) (rcu_dereference_protected((fn)->subtree, 1)) 117#endif 118 119/* 120 * routing information 121 * 122 */ 123 124struct rt6key { 125 struct in6_addr addr; 126 int plen; 127}; 128 129struct fib6_table; 130 131struct rt6_exception_bucket { 132 struct hlist_head chain; 133 int depth; 134}; 135 136struct rt6_exception { 137 struct hlist_node hlist; 138 struct rt6_info *rt6i; 139 unsigned long stamp; 140 struct rcu_head rcu; 141}; 142 143#define FIB6_EXCEPTION_BUCKET_SIZE_SHIFT 10 144#define FIB6_EXCEPTION_BUCKET_SIZE (1 << FIB6_EXCEPTION_BUCKET_SIZE_SHIFT) 145#define FIB6_MAX_DEPTH 5 146 147struct fib6_nh { 148 struct fib_nh_common nh_common; 149 150#ifdef CONFIG_IPV6_ROUTER_PREF 151 unsigned long last_probe; 152#endif 153 154 struct rt6_info * __percpu *rt6i_pcpu; 155 struct rt6_exception_bucket __rcu *rt6i_exception_bucket; 156}; 157 158struct fib6_info { 159 struct fib6_table *fib6_table; 160 struct fib6_info __rcu *fib6_next; 161 struct fib6_node __rcu *fib6_node; 162 163 /* Multipath routes: 164 * siblings is a list of fib6_info that have the same metric/weight, 165 * destination, but not the same gateway. nsiblings is just a cache 166 * to speed up lookup. 167 */ 168 union { 169 struct list_head fib6_siblings; 170 struct list_head nh_list; 171 }; 172 unsigned int fib6_nsiblings; 173 174 refcount_t fib6_ref; 175 unsigned long expires; 176 177 struct hlist_node gc_link; 178 179 struct dst_metrics *fib6_metrics; 180#define fib6_pmtu fib6_metrics->metrics[RTAX_MTU-1] 181 182 struct rt6key fib6_dst; 183 u32 fib6_flags; 184 struct rt6key fib6_src; 185 struct rt6key fib6_prefsrc; 186 187 u32 fib6_metric; 188 u8 fib6_protocol; 189 u8 fib6_type; 190 191 u8 offload; 192 u8 trap; 193 u8 offload_failed; 194 195 u8 should_flush:1, 196 dst_nocount:1, 197 dst_nopolicy:1, 198 fib6_destroying:1, 199 unused:4; 200 201 struct rcu_head rcu; 202 struct nexthop *nh; 203 struct fib6_nh fib6_nh[]; 204}; 205 206struct rt6_info { 207 struct dst_entry dst; 208 struct fib6_info __rcu *from; 209 int sernum; 210 211 struct rt6key rt6i_dst; 212 struct rt6key rt6i_src; 213 struct in6_addr rt6i_gateway; 214 struct inet6_dev *rt6i_idev; 215 u32 rt6i_flags; 216 217 /* more non-fragment space at head required */ 218 unsigned short rt6i_nfheader_len; 219}; 220 221struct fib6_result { 222 struct fib6_nh *nh; 223 struct fib6_info *f6i; 224 u32 fib6_flags; 225 u8 fib6_type; 226 struct rt6_info *rt6; 227}; 228 229#define for_each_fib6_node_rt_rcu(fn) \ 230 for (rt = rcu_dereference((fn)->leaf); rt; \ 231 rt = rcu_dereference(rt->fib6_next)) 232 233#define for_each_fib6_walker_rt(w) \ 234 for (rt = (w)->leaf; rt; \ 235 rt = rcu_dereference_protected(rt->fib6_next, 1)) 236 237static inline struct inet6_dev *ip6_dst_idev(struct dst_entry *dst) 238{ 239 return ((struct rt6_info *)dst)->rt6i_idev; 240} 241 242static inline bool fib6_requires_src(const struct fib6_info *rt) 243{ 244 return rt->fib6_src.plen > 0; 245} 246 247/* The callers should hold f6i->fib6_table->tb6_lock if a route has ever 248 * been added to a table before. 249 */ 250static inline void fib6_clean_expires(struct fib6_info *f6i) 251{ 252 f6i->fib6_flags &= ~RTF_EXPIRES; 253 f6i->expires = 0; 254} 255 256/* The callers should hold f6i->fib6_table->tb6_lock if a route has ever 257 * been added to a table before. 258 */ 259static inline void fib6_set_expires(struct fib6_info *f6i, 260 unsigned long expires) 261{ 262 f6i->expires = expires; 263 f6i->fib6_flags |= RTF_EXPIRES; 264} 265 266static inline bool fib6_check_expired(const struct fib6_info *f6i) 267{ 268 if (f6i->fib6_flags & RTF_EXPIRES) 269 return time_after(jiffies, f6i->expires); 270 return false; 271} 272 273/* Function to safely get fn->fn_sernum for passed in rt 274 * and store result in passed in cookie. 275 * Return true if we can get cookie safely 276 * Return false if not 277 */ 278static inline bool fib6_get_cookie_safe(const struct fib6_info *f6i, 279 u32 *cookie) 280{ 281 struct fib6_node *fn; 282 bool status = false; 283 284 fn = rcu_dereference(f6i->fib6_node); 285 286 if (fn) { 287 *cookie = READ_ONCE(fn->fn_sernum); 288 /* pairs with smp_wmb() in __fib6_update_sernum_upto_root() */ 289 smp_rmb(); 290 status = true; 291 } 292 293 return status; 294} 295 296static inline u32 rt6_get_cookie(const struct rt6_info *rt) 297{ 298 struct fib6_info *from; 299 u32 cookie = 0; 300 301 if (rt->sernum) 302 return rt->sernum; 303 304 rcu_read_lock(); 305 306 from = rcu_dereference(rt->from); 307 if (from) 308 fib6_get_cookie_safe(from, &cookie); 309 310 rcu_read_unlock(); 311 312 return cookie; 313} 314 315static inline void ip6_rt_put(struct rt6_info *rt) 316{ 317 /* dst_release() accepts a NULL parameter. 318 * We rely on dst being first structure in struct rt6_info 319 */ 320 BUILD_BUG_ON(offsetof(struct rt6_info, dst) != 0); 321 dst_release(&rt->dst); 322} 323 324struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh); 325void fib6_info_destroy_rcu(struct rcu_head *head); 326 327static inline void fib6_info_hold(struct fib6_info *f6i) 328{ 329 refcount_inc(&f6i->fib6_ref); 330} 331 332static inline bool fib6_info_hold_safe(struct fib6_info *f6i) 333{ 334 return refcount_inc_not_zero(&f6i->fib6_ref); 335} 336 337static inline void fib6_info_release(struct fib6_info *f6i) 338{ 339 if (f6i && refcount_dec_and_test(&f6i->fib6_ref)) { 340 DEBUG_NET_WARN_ON_ONCE(!hlist_unhashed(&f6i->gc_link)); 341 call_rcu(&f6i->rcu, fib6_info_destroy_rcu); 342 } 343} 344 345enum fib6_walk_state { 346#ifdef CONFIG_IPV6_SUBTREES 347 FWS_S, 348#endif 349 FWS_L, 350 FWS_R, 351 FWS_C, 352 FWS_U 353}; 354 355struct fib6_walker { 356 struct list_head lh; 357 struct fib6_node *root, *node; 358 struct fib6_info *leaf; 359 enum fib6_walk_state state; 360 unsigned int skip; 361 unsigned int count; 362 unsigned int skip_in_node; 363 int (*func)(struct fib6_walker *); 364 void *args; 365}; 366 367struct rt6_statistics { 368 __u32 fib_nodes; /* all fib6 nodes */ 369 __u32 fib_route_nodes; /* intermediate nodes */ 370 __u32 fib_rt_entries; /* rt entries in fib table */ 371 __u32 fib_rt_cache; /* cached rt entries in exception table */ 372 __u32 fib_discarded_routes; /* total number of routes delete */ 373 374 /* The following stat is not protected by any lock */ 375 atomic_t fib_rt_alloc; /* total number of routes alloced */ 376}; 377 378#define RTN_TL_ROOT 0x0001 379#define RTN_ROOT 0x0002 /* tree root node */ 380#define RTN_RTINFO 0x0004 /* node with valid routing info */ 381 382/* 383 * priority levels (or metrics) 384 * 385 */ 386 387 388struct fib6_table { 389 struct hlist_node tb6_hlist; 390 u32 tb6_id; 391 spinlock_t tb6_lock; 392 struct fib6_node tb6_root; 393 struct inet_peer_base tb6_peers; 394 unsigned int flags; 395 unsigned int fib_seq; 396 struct hlist_head tb6_gc_hlist; /* GC candidates */ 397#define RT6_TABLE_HAS_DFLT_ROUTER BIT(0) 398}; 399 400#define RT6_TABLE_UNSPEC RT_TABLE_UNSPEC 401#define RT6_TABLE_MAIN RT_TABLE_MAIN 402#define RT6_TABLE_DFLT RT6_TABLE_MAIN 403#define RT6_TABLE_INFO RT6_TABLE_MAIN 404#define RT6_TABLE_PREFIX RT6_TABLE_MAIN 405 406#ifdef CONFIG_IPV6_MULTIPLE_TABLES 407#define FIB6_TABLE_MIN 1 408#define FIB6_TABLE_MAX RT_TABLE_MAX 409#define RT6_TABLE_LOCAL RT_TABLE_LOCAL 410#else 411#define FIB6_TABLE_MIN RT_TABLE_MAIN 412#define FIB6_TABLE_MAX FIB6_TABLE_MIN 413#define RT6_TABLE_LOCAL RT6_TABLE_MAIN 414#endif 415 416typedef struct rt6_info *(*pol_lookup_t)(struct net *, 417 struct fib6_table *, 418 struct flowi6 *, 419 const struct sk_buff *, int); 420 421struct fib6_entry_notifier_info { 422 struct fib_notifier_info info; /* must be first */ 423 struct fib6_info *rt; 424 unsigned int nsiblings; 425}; 426 427/* 428 * exported functions 429 */ 430 431struct fib6_table *fib6_get_table(struct net *net, u32 id); 432struct fib6_table *fib6_new_table(struct net *net, u32 id); 433struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6, 434 const struct sk_buff *skb, 435 int flags, pol_lookup_t lookup); 436 437/* called with rcu lock held; can return error pointer 438 * caller needs to select path 439 */ 440int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6, 441 struct fib6_result *res, int flags); 442 443/* called with rcu lock held; caller needs to select path */ 444int fib6_table_lookup(struct net *net, struct fib6_table *table, 445 int oif, struct flowi6 *fl6, struct fib6_result *res, 446 int strict); 447 448void fib6_select_path(const struct net *net, struct fib6_result *res, 449 struct flowi6 *fl6, int oif, bool have_oif_match, 450 const struct sk_buff *skb, int strict); 451struct fib6_node *fib6_node_lookup(struct fib6_node *root, 452 const struct in6_addr *daddr, 453 const struct in6_addr *saddr); 454 455struct fib6_node *fib6_locate(struct fib6_node *root, 456 const struct in6_addr *daddr, int dst_len, 457 const struct in6_addr *saddr, int src_len, 458 bool exact_match); 459 460void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *arg), 461 void *arg); 462void fib6_clean_all_skip_notify(struct net *net, 463 int (*func)(struct fib6_info *, void *arg), 464 void *arg); 465 466int fib6_add(struct fib6_node *root, struct fib6_info *rt, 467 struct nl_info *info, struct netlink_ext_ack *extack); 468int fib6_del(struct fib6_info *rt, struct nl_info *info); 469 470static inline 471void rt6_get_prefsrc(const struct rt6_info *rt, struct in6_addr *addr) 472{ 473 const struct fib6_info *from; 474 475 rcu_read_lock(); 476 477 from = rcu_dereference(rt->from); 478 if (from) 479 *addr = from->fib6_prefsrc.addr; 480 else 481 *addr = in6addr_any; 482 483 rcu_read_unlock(); 484} 485 486int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh, 487 struct fib6_config *cfg, gfp_t gfp_flags, 488 struct netlink_ext_ack *extack); 489void fib6_nh_release(struct fib6_nh *fib6_nh); 490void fib6_nh_release_dsts(struct fib6_nh *fib6_nh); 491 492int call_fib6_entry_notifiers(struct net *net, 493 enum fib_event_type event_type, 494 struct fib6_info *rt, 495 struct netlink_ext_ack *extack); 496int call_fib6_multipath_entry_notifiers(struct net *net, 497 enum fib_event_type event_type, 498 struct fib6_info *rt, 499 unsigned int nsiblings, 500 struct netlink_ext_ack *extack); 501int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt); 502void fib6_rt_update(struct net *net, struct fib6_info *rt, 503 struct nl_info *info); 504void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info, 505 unsigned int flags); 506 507void fib6_run_gc(unsigned long expires, struct net *net, bool force); 508 509void fib6_gc_cleanup(void); 510 511int fib6_init(void); 512 513/* Add the route to the gc list if it is not already there 514 * 515 * The callers should hold f6i->fib6_table->tb6_lock. 516 */ 517static inline void fib6_add_gc_list(struct fib6_info *f6i) 518{ 519 /* If fib6_node is null, the f6i is not in (or removed from) the 520 * table. 521 * 522 * There is a gap between finding the f6i from the table and 523 * calling this function without the protection of the tb6_lock. 524 * This check makes sure the f6i is not added to the gc list when 525 * it is not on the table. 526 */ 527 if (!rcu_dereference_protected(f6i->fib6_node, 528 lockdep_is_held(&f6i->fib6_table->tb6_lock))) 529 return; 530 531 if (hlist_unhashed(&f6i->gc_link)) 532 hlist_add_head(&f6i->gc_link, &f6i->fib6_table->tb6_gc_hlist); 533} 534 535/* Remove the route from the gc list if it is on the list. 536 * 537 * The callers should hold f6i->fib6_table->tb6_lock. 538 */ 539static inline void fib6_remove_gc_list(struct fib6_info *f6i) 540{ 541 if (!hlist_unhashed(&f6i->gc_link)) 542 hlist_del_init(&f6i->gc_link); 543} 544 545struct ipv6_route_iter { 546 struct seq_net_private p; 547 struct fib6_walker w; 548 loff_t skip; 549 struct fib6_table *tbl; 550 int sernum; 551}; 552 553extern const struct seq_operations ipv6_route_seq_ops; 554 555int call_fib6_notifier(struct notifier_block *nb, 556 enum fib_event_type event_type, 557 struct fib_notifier_info *info); 558int call_fib6_notifiers(struct net *net, enum fib_event_type event_type, 559 struct fib_notifier_info *info); 560 561int __net_init fib6_notifier_init(struct net *net); 562void __net_exit fib6_notifier_exit(struct net *net); 563 564unsigned int fib6_tables_seq_read(struct net *net); 565int fib6_tables_dump(struct net *net, struct notifier_block *nb, 566 struct netlink_ext_ack *extack); 567 568void fib6_update_sernum(struct net *net, struct fib6_info *rt); 569void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt); 570void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i); 571 572void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val); 573static inline bool fib6_metric_locked(struct fib6_info *f6i, int metric) 574{ 575 return !!(f6i->fib6_metrics->metrics[RTAX_LOCK - 1] & (1 << metric)); 576} 577void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i, 578 bool offload, bool trap, bool offload_failed); 579 580#if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL) 581struct bpf_iter__ipv6_route { 582 __bpf_md_ptr(struct bpf_iter_meta *, meta); 583 __bpf_md_ptr(struct fib6_info *, rt); 584}; 585#endif 586 587INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_output(struct net *net, 588 struct fib6_table *table, 589 struct flowi6 *fl6, 590 const struct sk_buff *skb, 591 int flags)); 592INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_input(struct net *net, 593 struct fib6_table *table, 594 struct flowi6 *fl6, 595 const struct sk_buff *skb, 596 int flags)); 597INDIRECT_CALLABLE_DECLARE(struct rt6_info *__ip6_route_redirect(struct net *net, 598 struct fib6_table *table, 599 struct flowi6 *fl6, 600 const struct sk_buff *skb, 601 int flags)); 602INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_lookup(struct net *net, 603 struct fib6_table *table, 604 struct flowi6 *fl6, 605 const struct sk_buff *skb, 606 int flags)); 607static inline struct rt6_info *pol_lookup_func(pol_lookup_t lookup, 608 struct net *net, 609 struct fib6_table *table, 610 struct flowi6 *fl6, 611 const struct sk_buff *skb, 612 int flags) 613{ 614 return INDIRECT_CALL_4(lookup, 615 ip6_pol_route_output, 616 ip6_pol_route_input, 617 ip6_pol_route_lookup, 618 __ip6_route_redirect, 619 net, table, fl6, skb, flags); 620} 621 622#ifdef CONFIG_IPV6_MULTIPLE_TABLES 623static inline bool fib6_has_custom_rules(const struct net *net) 624{ 625 return net->ipv6.fib6_has_custom_rules; 626} 627 628int fib6_rules_init(void); 629void fib6_rules_cleanup(void); 630bool fib6_rule_default(const struct fib_rule *rule); 631int fib6_rules_dump(struct net *net, struct notifier_block *nb, 632 struct netlink_ext_ack *extack); 633unsigned int fib6_rules_seq_read(struct net *net); 634 635static inline bool fib6_rules_early_flow_dissect(struct net *net, 636 struct sk_buff *skb, 637 struct flowi6 *fl6, 638 struct flow_keys *flkeys) 639{ 640 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 641 642 if (!net->ipv6.fib6_rules_require_fldissect) 643 return false; 644 645 memset(flkeys, 0, sizeof(*flkeys)); 646 __skb_flow_dissect(net, skb, &flow_keys_dissector, 647 flkeys, NULL, 0, 0, 0, flag); 648 649 fl6->fl6_sport = flkeys->ports.src; 650 fl6->fl6_dport = flkeys->ports.dst; 651 fl6->flowi6_proto = flkeys->basic.ip_proto; 652 653 return true; 654} 655#else 656static inline bool fib6_has_custom_rules(const struct net *net) 657{ 658 return false; 659} 660static inline int fib6_rules_init(void) 661{ 662 return 0; 663} 664static inline void fib6_rules_cleanup(void) 665{ 666 return ; 667} 668static inline bool fib6_rule_default(const struct fib_rule *rule) 669{ 670 return true; 671} 672static inline int fib6_rules_dump(struct net *net, struct notifier_block *nb, 673 struct netlink_ext_ack *extack) 674{ 675 return 0; 676} 677static inline unsigned int fib6_rules_seq_read(struct net *net) 678{ 679 return 0; 680} 681static inline bool fib6_rules_early_flow_dissect(struct net *net, 682 struct sk_buff *skb, 683 struct flowi6 *fl6, 684 struct flow_keys *flkeys) 685{ 686 return false; 687} 688#endif 689#endif 690