1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the Forwarding Information Base. 8 * 9 * Authors: A.N.Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 */ 11 12 #ifndef _NET_IP_FIB_H 13 #define _NET_IP_FIB_H 14 15 #include <net/flow.h> 16 #include <linux/seq_file.h> 17 #include <linux/rcupdate.h> 18 #include <net/fib_notifier.h> 19 #include <net/fib_rules.h> 20 #include <net/inet_dscp.h> 21 #include <net/inetpeer.h> 22 #include <linux/percpu.h> 23 #include <linux/notifier.h> 24 #include <linux/refcount.h> 25 26 struct fib_config { 27 u8 fc_dst_len; 28 dscp_t fc_dscp; 29 u8 fc_protocol; 30 u8 fc_scope; 31 u8 fc_type; 32 u8 fc_gw_family; 33 /* 2 bytes unused */ 34 u32 fc_table; 35 __be32 fc_dst; 36 union { 37 __be32 fc_gw4; 38 struct in6_addr fc_gw6; 39 }; 40 int fc_oif; 41 u32 fc_flags; 42 u32 fc_priority; 43 __be32 fc_prefsrc; 44 u32 fc_nh_id; 45 struct nlattr *fc_mx; 46 struct rtnexthop *fc_mp; 47 int fc_mx_len; 48 int fc_mp_len; 49 u32 fc_flow; 50 u32 fc_nlflags; 51 struct nl_info fc_nlinfo; 52 struct nlattr *fc_encap; 53 u16 fc_encap_type; 54 }; 55 56 struct fib_info; 57 struct rtable; 58 59 struct fib_nh_exception { 60 struct fib_nh_exception __rcu *fnhe_next; 61 int fnhe_genid; 62 __be32 fnhe_daddr; 63 u32 fnhe_pmtu; 64 bool fnhe_mtu_locked; 65 __be32 fnhe_gw; 66 unsigned long fnhe_expires; 67 struct rtable __rcu *fnhe_rth_input; 68 struct rtable __rcu *fnhe_rth_output; 69 unsigned long fnhe_stamp; 70 struct rcu_head rcu; 71 }; 72 73 struct fnhe_hash_bucket { 74 struct fib_nh_exception __rcu *chain; 75 }; 76 77 #define FNHE_HASH_SHIFT 11 78 #define FNHE_HASH_SIZE (1 << FNHE_HASH_SHIFT) 79 #define FNHE_RECLAIM_DEPTH 5 80 81 struct fib_nh_common { 82 struct net_device *nhc_dev; 83 netdevice_tracker nhc_dev_tracker; 84 int nhc_oif; 85 unsigned char nhc_scope; 86 u8 nhc_family; 87 u8 nhc_gw_family; 88 unsigned char nhc_flags; 89 struct lwtunnel_state *nhc_lwtstate; 90 91 union { 92 __be32 ipv4; 93 struct in6_addr ipv6; 94 } nhc_gw; 95 96 int nhc_weight; 97 atomic_t nhc_upper_bound; 98 99 /* v4 specific, but allows fib6_nh with v4 routes */ 100 struct rtable __rcu * __percpu *nhc_pcpu_rth_output; 101 struct rtable __rcu *nhc_rth_input; 102 struct fnhe_hash_bucket __rcu *nhc_exceptions; 103 }; 104 105 struct fib_nh { 106 struct fib_nh_common nh_common; 107 struct hlist_node nh_hash; 108 struct fib_info *nh_parent; 109 #ifdef CONFIG_IP_ROUTE_CLASSID 110 __u32 nh_tclassid; 111 #endif 112 __be32 nh_saddr; 113 int nh_saddr_genid; 114 #define fib_nh_family nh_common.nhc_family 115 #define fib_nh_dev nh_common.nhc_dev 116 #define fib_nh_dev_tracker nh_common.nhc_dev_tracker 117 #define fib_nh_oif nh_common.nhc_oif 118 #define fib_nh_flags nh_common.nhc_flags 119 #define fib_nh_lws nh_common.nhc_lwtstate 120 #define fib_nh_scope nh_common.nhc_scope 121 #define fib_nh_gw_family nh_common.nhc_gw_family 122 #define fib_nh_gw4 nh_common.nhc_gw.ipv4 123 #define fib_nh_gw6 nh_common.nhc_gw.ipv6 124 #define fib_nh_weight nh_common.nhc_weight 125 #define fib_nh_upper_bound nh_common.nhc_upper_bound 126 }; 127 128 /* 129 * This structure contains data shared by many of routes. 130 */ 131 132 struct nexthop; 133 134 struct fib_info { 135 struct hlist_node fib_hash; 136 struct hlist_node fib_lhash; 137 struct list_head nh_list; 138 struct net *fib_net; 139 refcount_t fib_treeref; 140 refcount_t fib_clntref; 141 unsigned int fib_flags; 142 unsigned char fib_dead; 143 unsigned char fib_protocol; 144 unsigned char fib_scope; 145 unsigned char fib_type; 146 __be32 fib_prefsrc; 147 u32 fib_tb_id; 148 u32 fib_priority; 149 struct dst_metrics *fib_metrics; 150 #define fib_mtu fib_metrics->metrics[RTAX_MTU-1] 151 #define fib_window fib_metrics->metrics[RTAX_WINDOW-1] 152 #define fib_rtt fib_metrics->metrics[RTAX_RTT-1] 153 #define fib_advmss fib_metrics->metrics[RTAX_ADVMSS-1] 154 int fib_nhs; 155 bool fib_nh_is_v6; 156 bool nh_updated; 157 bool pfsrc_removed; 158 struct nexthop *nh; 159 struct rcu_head rcu; 160 struct fib_nh fib_nh[] __counted_by(fib_nhs); 161 }; 162 163 164 #ifdef CONFIG_IP_MULTIPLE_TABLES 165 struct fib_rule; 166 #endif 167 168 struct fib_table; 169 struct fib_result { 170 __be32 prefix; 171 unsigned char prefixlen; 172 unsigned char nh_sel; 173 unsigned char type; 174 unsigned char scope; 175 u32 tclassid; 176 struct fib_nh_common *nhc; 177 struct fib_info *fi; 178 struct fib_table *table; 179 struct hlist_head *fa_head; 180 }; 181 182 struct fib_result_nl { 183 __be32 fl_addr; /* To be looked up*/ 184 u32 fl_mark; 185 unsigned char fl_tos; 186 unsigned char fl_scope; 187 unsigned char tb_id_in; 188 189 unsigned char tb_id; /* Results */ 190 unsigned char prefixlen; 191 unsigned char nh_sel; 192 unsigned char type; 193 unsigned char scope; 194 int err; 195 }; 196 197 #ifdef CONFIG_IP_MULTIPLE_TABLES 198 #define FIB_TABLE_HASHSZ 256 199 #else 200 #define FIB_TABLE_HASHSZ 2 201 #endif 202 203 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc, 204 unsigned char scope); 205 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res); 206 207 #define FIB_RES_NHC(res) ((res).nhc) 208 #define FIB_RES_DEV(res) (FIB_RES_NHC(res)->nhc_dev) 209 #define FIB_RES_OIF(res) (FIB_RES_NHC(res)->nhc_oif) 210 211 struct fib_rt_info { 212 struct fib_info *fi; 213 u32 tb_id; 214 __be32 dst; 215 int dst_len; 216 dscp_t dscp; 217 u8 type; 218 u8 offload:1, 219 trap:1, 220 offload_failed:1, 221 unused:5; 222 }; 223 224 struct fib_entry_notifier_info { 225 struct fib_notifier_info info; /* must be first */ 226 u32 dst; 227 int dst_len; 228 struct fib_info *fi; 229 dscp_t dscp; 230 u8 type; 231 u32 tb_id; 232 }; 233 234 struct fib_nh_notifier_info { 235 struct fib_notifier_info info; /* must be first */ 236 struct fib_nh *fib_nh; 237 }; 238 239 int call_fib4_notifier(struct notifier_block *nb, 240 enum fib_event_type event_type, 241 struct fib_notifier_info *info); 242 int call_fib4_notifiers(struct net *net, enum fib_event_type event_type, 243 struct fib_notifier_info *info); 244 245 int __net_init fib4_notifier_init(struct net *net); 246 void __net_exit fib4_notifier_exit(struct net *net); 247 248 void fib_info_notify_update(struct net *net, struct nl_info *info); 249 int fib_notify(struct net *net, struct notifier_block *nb, 250 struct netlink_ext_ack *extack); 251 252 struct fib_table { 253 struct hlist_node tb_hlist; 254 u32 tb_id; 255 int tb_num_default; 256 struct rcu_head rcu; 257 unsigned long *tb_data; 258 unsigned long __data[]; 259 }; 260 261 struct fib_dump_filter { 262 u32 table_id; 263 /* filter_set is an optimization that an entry is set */ 264 bool filter_set; 265 bool dump_routes; 266 bool dump_exceptions; 267 unsigned char protocol; 268 unsigned char rt_type; 269 unsigned int flags; 270 struct net_device *dev; 271 }; 272 273 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp, 274 struct fib_result *res, int fib_flags); 275 int fib_table_insert(struct net *, struct fib_table *, struct fib_config *, 276 struct netlink_ext_ack *extack); 277 int fib_table_delete(struct net *, struct fib_table *, struct fib_config *, 278 struct netlink_ext_ack *extack); 279 int fib_table_dump(struct fib_table *table, struct sk_buff *skb, 280 struct netlink_callback *cb, struct fib_dump_filter *filter); 281 int fib_table_flush(struct net *net, struct fib_table *table, bool flush_all); 282 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb); 283 void fib_table_flush_external(struct fib_table *table); 284 void fib_free_table(struct fib_table *tb); 285 286 #ifndef CONFIG_IP_MULTIPLE_TABLES 287 288 #define TABLE_LOCAL_INDEX (RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1)) 289 #define TABLE_MAIN_INDEX (RT_TABLE_MAIN & (FIB_TABLE_HASHSZ - 1)) 290 291 static inline struct fib_table *fib_get_table(struct net *net, u32 id) 292 { 293 struct hlist_node *tb_hlist; 294 struct hlist_head *ptr; 295 296 ptr = id == RT_TABLE_LOCAL ? 297 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] : 298 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]; 299 300 tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr)); 301 302 return hlist_entry(tb_hlist, struct fib_table, tb_hlist); 303 } 304 305 static inline struct fib_table *fib_new_table(struct net *net, u32 id) 306 { 307 return fib_get_table(net, id); 308 } 309 310 static inline int fib_lookup(struct net *net, const struct flowi4 *flp, 311 struct fib_result *res, unsigned int flags) 312 { 313 struct fib_table *tb; 314 int err = -ENETUNREACH; 315 316 rcu_read_lock(); 317 318 tb = fib_get_table(net, RT_TABLE_MAIN); 319 if (tb) 320 err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF); 321 322 if (err == -EAGAIN) 323 err = -ENETUNREACH; 324 325 rcu_read_unlock(); 326 327 return err; 328 } 329 330 static inline bool fib4_has_custom_rules(const struct net *net) 331 { 332 return false; 333 } 334 335 static inline bool fib4_rule_default(const struct fib_rule *rule) 336 { 337 return true; 338 } 339 340 static inline int fib4_rules_dump(struct net *net, struct notifier_block *nb, 341 struct netlink_ext_ack *extack) 342 { 343 return 0; 344 } 345 346 static inline unsigned int fib4_rules_seq_read(struct net *net) 347 { 348 return 0; 349 } 350 351 static inline bool fib4_rules_early_flow_dissect(struct net *net, 352 struct sk_buff *skb, 353 struct flowi4 *fl4, 354 struct flow_keys *flkeys) 355 { 356 return false; 357 } 358 #else /* CONFIG_IP_MULTIPLE_TABLES */ 359 int __net_init fib4_rules_init(struct net *net); 360 void __net_exit fib4_rules_exit(struct net *net); 361 362 struct fib_table *fib_new_table(struct net *net, u32 id); 363 struct fib_table *fib_get_table(struct net *net, u32 id); 364 365 int __fib_lookup(struct net *net, struct flowi4 *flp, 366 struct fib_result *res, unsigned int flags); 367 368 static inline int fib_lookup(struct net *net, struct flowi4 *flp, 369 struct fib_result *res, unsigned int flags) 370 { 371 struct fib_table *tb; 372 int err = -ENETUNREACH; 373 374 flags |= FIB_LOOKUP_NOREF; 375 if (net->ipv4.fib_has_custom_rules) 376 return __fib_lookup(net, flp, res, flags); 377 378 rcu_read_lock(); 379 380 res->tclassid = 0; 381 382 tb = rcu_dereference_rtnl(net->ipv4.fib_main); 383 if (tb) 384 err = fib_table_lookup(tb, flp, res, flags); 385 386 if (!err) 387 goto out; 388 389 tb = rcu_dereference_rtnl(net->ipv4.fib_default); 390 if (tb) 391 err = fib_table_lookup(tb, flp, res, flags); 392 393 out: 394 if (err == -EAGAIN) 395 err = -ENETUNREACH; 396 397 rcu_read_unlock(); 398 399 return err; 400 } 401 402 static inline bool fib4_has_custom_rules(const struct net *net) 403 { 404 return net->ipv4.fib_has_custom_rules; 405 } 406 407 bool fib4_rule_default(const struct fib_rule *rule); 408 int fib4_rules_dump(struct net *net, struct notifier_block *nb, 409 struct netlink_ext_ack *extack); 410 unsigned int fib4_rules_seq_read(struct net *net); 411 412 static inline bool fib4_rules_early_flow_dissect(struct net *net, 413 struct sk_buff *skb, 414 struct flowi4 *fl4, 415 struct flow_keys *flkeys) 416 { 417 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 418 419 if (!net->ipv4.fib_rules_require_fldissect) 420 return false; 421 422 memset(flkeys, 0, sizeof(*flkeys)); 423 __skb_flow_dissect(net, skb, &flow_keys_dissector, 424 flkeys, NULL, 0, 0, 0, flag); 425 426 fl4->fl4_sport = flkeys->ports.src; 427 fl4->fl4_dport = flkeys->ports.dst; 428 fl4->flowi4_proto = flkeys->basic.ip_proto; 429 430 return true; 431 } 432 433 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 434 435 /* Exported by fib_frontend.c */ 436 extern const struct nla_policy rtm_ipv4_policy[]; 437 void ip_fib_init(void); 438 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla, 439 struct netlink_ext_ack *extack); 440 __be32 fib_compute_spec_dst(struct sk_buff *skb); 441 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev); 442 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 443 u8 tos, int oif, struct net_device *dev, 444 struct in_device *idev, u32 *itag); 445 #ifdef CONFIG_IP_ROUTE_CLASSID 446 static inline int fib_num_tclassid_users(struct net *net) 447 { 448 return atomic_read(&net->ipv4.fib_num_tclassid_users); 449 } 450 #else 451 static inline int fib_num_tclassid_users(struct net *net) 452 { 453 return 0; 454 } 455 #endif 456 int fib_unmerge(struct net *net); 457 458 static inline bool nhc_l3mdev_matches_dev(const struct fib_nh_common *nhc, 459 const struct net_device *dev) 460 { 461 if (nhc->nhc_dev == dev || 462 l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex) 463 return true; 464 465 return false; 466 } 467 468 /* Exported by fib_semantics.c */ 469 int ip_fib_check_default(__be32 gw, struct net_device *dev); 470 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force); 471 int fib_sync_down_addr(struct net_device *dev, __be32 local); 472 int fib_sync_up(struct net_device *dev, unsigned char nh_flags); 473 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu); 474 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig); 475 476 /* Fields used for sysctl_fib_multipath_hash_fields. 477 * Common to IPv4 and IPv6. 478 * 479 * Add new fields at the end. This is user API. 480 */ 481 #define FIB_MULTIPATH_HASH_FIELD_SRC_IP BIT(0) 482 #define FIB_MULTIPATH_HASH_FIELD_DST_IP BIT(1) 483 #define FIB_MULTIPATH_HASH_FIELD_IP_PROTO BIT(2) 484 #define FIB_MULTIPATH_HASH_FIELD_FLOWLABEL BIT(3) 485 #define FIB_MULTIPATH_HASH_FIELD_SRC_PORT BIT(4) 486 #define FIB_MULTIPATH_HASH_FIELD_DST_PORT BIT(5) 487 #define FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP BIT(6) 488 #define FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP BIT(7) 489 #define FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO BIT(8) 490 #define FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL BIT(9) 491 #define FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT BIT(10) 492 #define FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT BIT(11) 493 494 #define FIB_MULTIPATH_HASH_FIELD_OUTER_MASK \ 495 (FIB_MULTIPATH_HASH_FIELD_SRC_IP | \ 496 FIB_MULTIPATH_HASH_FIELD_DST_IP | \ 497 FIB_MULTIPATH_HASH_FIELD_IP_PROTO | \ 498 FIB_MULTIPATH_HASH_FIELD_FLOWLABEL | \ 499 FIB_MULTIPATH_HASH_FIELD_SRC_PORT | \ 500 FIB_MULTIPATH_HASH_FIELD_DST_PORT) 501 502 #define FIB_MULTIPATH_HASH_FIELD_INNER_MASK \ 503 (FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP | \ 504 FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP | \ 505 FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO | \ 506 FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL | \ 507 FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT | \ 508 FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 509 510 #define FIB_MULTIPATH_HASH_FIELD_ALL_MASK \ 511 (FIB_MULTIPATH_HASH_FIELD_OUTER_MASK | \ 512 FIB_MULTIPATH_HASH_FIELD_INNER_MASK) 513 514 #define FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK \ 515 (FIB_MULTIPATH_HASH_FIELD_SRC_IP | \ 516 FIB_MULTIPATH_HASH_FIELD_DST_IP | \ 517 FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 518 519 #ifdef CONFIG_IP_ROUTE_MULTIPATH 520 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 521 const struct sk_buff *skb, struct flow_keys *flkeys); 522 #endif 523 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope, 524 struct netlink_ext_ack *extack); 525 void fib_select_multipath(struct fib_result *res, int hash); 526 void fib_select_path(struct net *net, struct fib_result *res, 527 struct flowi4 *fl4, const struct sk_buff *skb); 528 529 int fib_nh_init(struct net *net, struct fib_nh *fib_nh, 530 struct fib_config *cfg, int nh_weight, 531 struct netlink_ext_ack *extack); 532 void fib_nh_release(struct net *net, struct fib_nh *fib_nh); 533 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc, 534 struct nlattr *fc_encap, u16 fc_encap_type, 535 void *cfg, gfp_t gfp_flags, 536 struct netlink_ext_ack *extack); 537 void fib_nh_common_release(struct fib_nh_common *nhc); 538 539 /* Exported by fib_trie.c */ 540 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri); 541 void fib_trie_init(void); 542 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias); 543 bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags, 544 const struct flowi4 *flp); 545 546 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res) 547 { 548 #ifdef CONFIG_IP_ROUTE_CLASSID 549 struct fib_nh_common *nhc = res->nhc; 550 #ifdef CONFIG_IP_MULTIPLE_TABLES 551 u32 rtag; 552 #endif 553 if (nhc->nhc_family == AF_INET) { 554 struct fib_nh *nh; 555 556 nh = container_of(nhc, struct fib_nh, nh_common); 557 *itag = nh->nh_tclassid << 16; 558 } else { 559 *itag = 0; 560 } 561 562 #ifdef CONFIG_IP_MULTIPLE_TABLES 563 rtag = res->tclassid; 564 if (*itag == 0) 565 *itag = (rtag<<16); 566 *itag |= (rtag>>16); 567 #endif 568 #endif 569 } 570 571 void fib_flush(struct net *net); 572 void free_fib_info(struct fib_info *fi); 573 574 static inline void fib_info_hold(struct fib_info *fi) 575 { 576 refcount_inc(&fi->fib_clntref); 577 } 578 579 static inline void fib_info_put(struct fib_info *fi) 580 { 581 if (refcount_dec_and_test(&fi->fib_clntref)) 582 free_fib_info(fi); 583 } 584 585 #ifdef CONFIG_PROC_FS 586 int __net_init fib_proc_init(struct net *net); 587 void __net_exit fib_proc_exit(struct net *net); 588 #else 589 static inline int fib_proc_init(struct net *net) 590 { 591 return 0; 592 } 593 static inline void fib_proc_exit(struct net *net) 594 { 595 } 596 #endif 597 598 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr); 599 600 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh, 601 struct fib_dump_filter *filter, 602 struct netlink_callback *cb); 603 604 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nh, 605 u8 rt_family, unsigned char *flags, bool skip_oif); 606 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nh, 607 int nh_weight, u8 rt_family, u32 nh_tclassid); 608 #endif /* _NET_FIB_H */ 609