include/net/ip6_fib.h

   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
  24 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
  25 #define FIB6_TABLE_HASHSZ 256
  26 #else
  27 #define FIB6_TABLE_HASHSZ 1
  28 #endif
  29
  30 #define RT6_DEBUG 2
  31
  32 #if RT6_DEBUG >= 3
  33 #define RT6_TRACE(x...) pr_debug(x)
  34 #else
  35 #define RT6_TRACE(x...) do { ; } while (0)
  36 #endif
  37
  38 struct rt6_info;
  39 struct fib6_info;
  40
  41 struct fib6_config {
  42         u32             fc_table;
  43         u32             fc_metric;
  44         int             fc_dst_len;
  45         int             fc_src_len;
  46         int             fc_ifindex;
  47         u32             fc_flags;
  48         u32             fc_protocol;
  49         u16             fc_type;        /* only 8 bits are used */
  50         u16             fc_delete_all_nh : 1,
  51                         fc_ignore_dev_down:1,
  52                         __unused : 14;
  53         u32             fc_nh_id;
  54
  55         struct in6_addr fc_dst;
  56         struct in6_addr fc_src;
  57         struct in6_addr fc_prefsrc;
  58         struct in6_addr fc_gateway;
  59
  60         unsigned long   fc_expires;
  61         struct nlattr   *fc_mx;
  62         int             fc_mx_len;
  63         int             fc_mp_len;
  64         struct nlattr   *fc_mp;
  65
  66         struct nl_info  fc_nlinfo;
  67         struct nlattr   *fc_encap;
  68         u16             fc_encap_type;
  69         bool            fc_is_fdb;
  70 };
  71
  72 struct fib6_node {
  73         struct fib6_node __rcu  *parent;
  74         struct fib6_node __rcu  *left;
  75         struct fib6_node __rcu  *right;
  76 #ifdef CONFIG_IPV6_SUBTREES
  77         struct fib6_node __rcu  *subtree;
  78 #endif
  79         struct fib6_info __rcu  *leaf;
  80
  81         __u16                   fn_bit;         /* bit key */
  82         __u16                   fn_flags;
  83         int                     fn_sernum;
  84         struct fib6_info __rcu  *rr_ptr;
  85         struct rcu_head         rcu;
  86 };
  87
  88 struct fib6_gc_args {
  89         int                     timeout;
  90         int                     more;
  91 };
  92
  93 #ifndef CONFIG_IPV6_SUBTREES
  94 #define FIB6_SUBTREE(fn)        NULL
  95
  96 static inline bool fib6_routes_require_src(const struct net *net)
  97 {
  98         return false;
  99 }
 100
 101 static inline void fib6_routes_require_src_inc(struct net *net) {}
 102 static inline void fib6_routes_require_src_dec(struct net *net) {}
 103
 104 #else
 105
 106 static inline bool fib6_routes_require_src(const struct net *net)
 107 {
 108         return net->ipv6.fib6_routes_require_src > 0;
 109 }
 110
 111 static inline void fib6_routes_require_src_inc(struct net *net)
 112 {
 113         net->ipv6.fib6_routes_require_src++;
 114 }
 115
 116 static inline void fib6_routes_require_src_dec(struct net *net)
 117 {
 118         net->ipv6.fib6_routes_require_src--;
 119 }
 120
 121 #define FIB6_SUBTREE(fn)        (rcu_dereference_protected((fn)->subtree, 1))
 122 #endif
 123
 124 /*
 125  *      routing information
 126  *
 127  */
 128
 129 struct rt6key {
 130         struct in6_addr addr;
 131         int             plen;
 132 };
 133
 134 struct fib6_table;
 135
 136 struct rt6_exception_bucket {
 137         struct hlist_head       chain;
 138         int                     depth;
 139 };
 140
 141 struct rt6_exception {
 142         struct hlist_node       hlist;
 143         struct rt6_info         *rt6i;
 144         unsigned long           stamp;
 145         struct rcu_head         rcu;
 146 };
 147
 148 #define FIB6_EXCEPTION_BUCKET_SIZE_SHIFT 10
 149 #define FIB6_EXCEPTION_BUCKET_SIZE (1 << FIB6_EXCEPTION_BUCKET_SIZE_SHIFT)
 150 #define FIB6_MAX_DEPTH 5
 151
 152 struct fib6_nh {
 153         struct fib_nh_common    nh_common;
 154
 155 #ifdef CONFIG_IPV6_ROUTER_PREF
 156         unsigned long           last_probe;
 157 #endif
 158
 159         struct rt6_info * __percpu *rt6i_pcpu;
 160         struct rt6_exception_bucket __rcu *rt6i_exception_bucket;
 161 };
 162
 163 struct fib6_info {
 164         struct fib6_table               *fib6_table;
 165         struct fib6_info __rcu          *fib6_next;
 166         struct fib6_node __rcu          *fib6_node;
 167
 168         /* Multipath routes:
 169          * siblings is a list of fib6_info that have the same metric/weight,
 170          * destination, but not the same gateway. nsiblings is just a cache
 171          * to speed up lookup.
 172          */
 173         union {
 174                 struct list_head        fib6_siblings;
 175                 struct list_head        nh_list;
 176         };
 177         unsigned int                    fib6_nsiblings;
 178
 179         refcount_t                      fib6_ref;
 180         unsigned long                   expires;
 181         struct dst_metrics              *fib6_metrics;
 182 #define fib6_pmtu               fib6_metrics->metrics[RTAX_MTU-1]
 183
 184         struct rt6key                   fib6_dst;
 185         u32                             fib6_flags;
 186         struct rt6key                   fib6_src;
 187         struct rt6key                   fib6_prefsrc;
 188
 189         u32                             fib6_metric;
 190         u8                              fib6_protocol;
 191         u8                              fib6_type;
 192         u8                              should_flush:1,
 193                                         dst_nocount:1,
 194                                         dst_nopolicy:1,
 195                                         fib6_destroying:1,
 196                                         offload:1,
 197                                         trap:1,
 198                                         offload_failed:1,
 199                                         unused:1;
 200
 201         struct rcu_head                 rcu;
 202         struct nexthop                  *nh;
 203         struct fib6_nh                  fib6_nh[];
 204 };
 205
 206 struct 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         struct list_head                rt6i_uncached;
 218         struct uncached_list            *rt6i_uncached_list;
 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 static inline struct inet6_dev *ip6_dst_idev(struct dst_entry *dst)
 241 {
 242         return ((struct rt6_info *)dst)->rt6i_idev;
 243 }
 244
 245 static inline bool fib6_requires_src(const struct fib6_info *rt)
 246 {
 247         return rt->fib6_src.plen > 0;
 248 }
 249
 250 static inline void fib6_clean_expires(struct fib6_info *f6i)
 251 {
 252         f6i->fib6_flags &= ~RTF_EXPIRES;
 253         f6i->expires = 0;
 254 }
 255
 256 static inline void fib6_set_expires(struct fib6_info *f6i,
 257                                     unsigned long expires)
 258 {
 259         f6i->expires = expires;
 260         f6i->fib6_flags |= RTF_EXPIRES;
 261 }
 262
 263 static inline bool fib6_check_expired(const struct fib6_info *f6i)
 264 {
 265         if (f6i->fib6_flags & RTF_EXPIRES)
 266                 return time_after(jiffies, f6i->expires);
 267         return false;
 268 }
 269
 270 /* Function to safely get fn->fn_sernum for passed in rt
 271  * and store result in passed in cookie.
 272  * Return true if we can get cookie safely
 273  * Return false if not
 274  */
 275 static inline bool fib6_get_cookie_safe(const struct fib6_info *f6i,
 276                                         u32 *cookie)
 277 {
 278         struct fib6_node *fn;
 279         bool status = false;
 280
 281         fn = rcu_dereference(f6i->fib6_node);
 282
 283         if (fn) {
 284                 *cookie = fn->fn_sernum;
 285                 /* pairs with smp_wmb() in __fib6_update_sernum_upto_root() */
 286                 smp_rmb();
 287                 status = true;
 288         }
 289
 290         return status;
 291 }
 292
 293 static inline u32 rt6_get_cookie(const struct rt6_info *rt)
 294 {
 295         struct fib6_info *from;
 296         u32 cookie = 0;
 297
 298         if (rt->sernum)
 299                 return rt->sernum;
 300
 301         rcu_read_lock();
 302
 303         from = rcu_dereference(rt->from);
 304         if (from)
 305                 fib6_get_cookie_safe(from, &cookie);
 306
 307         rcu_read_unlock();
 308
 309         return cookie;
 310 }
 311
 312 static inline void ip6_rt_put(struct rt6_info *rt)
 313 {
 314         /* dst_release() accepts a NULL parameter.
 315          * We rely on dst being first structure in struct rt6_info
 316          */
 317         BUILD_BUG_ON(offsetof(struct rt6_info, dst) != 0);
 318         dst_release(&rt->dst);
 319 }
 320
 321 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh);
 322 void fib6_info_destroy_rcu(struct rcu_head *head);
 323
 324 static inline void fib6_info_hold(struct fib6_info *f6i)
 325 {
 326         refcount_inc(&f6i->fib6_ref);
 327 }
 328
 329 static inline bool fib6_info_hold_safe(struct fib6_info *f6i)
 330 {
 331         return refcount_inc_not_zero(&f6i->fib6_ref);
 332 }
 333
 334 static inline void fib6_info_release(struct fib6_info *f6i)
 335 {
 336         if (f6i && refcount_dec_and_test(&f6i->fib6_ref))
 337                 call_rcu(&f6i->rcu, fib6_info_destroy_rcu);
 338 }
 339
 340 enum fib6_walk_state {
 341 #ifdef CONFIG_IPV6_SUBTREES
 342         FWS_S,
 343 #endif
 344         FWS_L,
 345         FWS_R,
 346         FWS_C,
 347         FWS_U
 348 };
 349
 350 struct fib6_walker {
 351         struct list_head lh;
 352         struct fib6_node *root, *node;
 353         struct fib6_info *leaf;
 354         enum fib6_walk_state state;
 355         unsigned int skip;
 356         unsigned int count;
 357         unsigned int skip_in_node;
 358         int (*func)(struct fib6_walker *);
 359         void *args;
 360 };
 361
 362 struct rt6_statistics {
 363         __u32           fib_nodes;              /* all fib6 nodes */
 364         __u32           fib_route_nodes;        /* intermediate nodes */
 365         __u32           fib_rt_entries;         /* rt entries in fib table */
 366         __u32           fib_rt_cache;           /* cached rt entries in exception table */
 367         __u32           fib_discarded_routes;   /* total number of routes delete */
 368
 369         /* The following stats are not protected by any lock */
 370         atomic_t        fib_rt_alloc;           /* total number of routes alloced */
 371         atomic_t        fib_rt_uncache;         /* rt entries in uncached list */
 372 };
 373
 374 #define RTN_TL_ROOT     0x0001
 375 #define RTN_ROOT        0x0002          /* tree root node               */
 376 #define RTN_RTINFO      0x0004          /* node with valid routing info */
 377
 378 /*
 379  *      priority levels (or metrics)
 380  *
 381  */
 382
 383
 384 struct fib6_table {
 385         struct hlist_node       tb6_hlist;
 386         u32                     tb6_id;
 387         spinlock_t              tb6_lock;
 388         struct fib6_node        tb6_root;
 389         struct inet_peer_base   tb6_peers;
 390         unsigned int            flags;
 391         unsigned int            fib_seq;
 392 #define RT6_TABLE_HAS_DFLT_ROUTER       BIT(0)
 393 };
 394
 395 #define RT6_TABLE_UNSPEC        RT_TABLE_UNSPEC
 396 #define RT6_TABLE_MAIN          RT_TABLE_MAIN
 397 #define RT6_TABLE_DFLT          RT6_TABLE_MAIN
 398 #define RT6_TABLE_INFO          RT6_TABLE_MAIN
 399 #define RT6_TABLE_PREFIX        RT6_TABLE_MAIN
 400
 401 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 402 #define FIB6_TABLE_MIN          1
 403 #define FIB6_TABLE_MAX          RT_TABLE_MAX
 404 #define RT6_TABLE_LOCAL         RT_TABLE_LOCAL
 405 #else
 406 #define FIB6_TABLE_MIN          RT_TABLE_MAIN
 407 #define FIB6_TABLE_MAX          FIB6_TABLE_MIN
 408 #define RT6_TABLE_LOCAL         RT6_TABLE_MAIN
 409 #endif
 410
 411 typedef struct rt6_info *(*pol_lookup_t)(struct net *,
 412                                          struct fib6_table *,
 413                                          struct flowi6 *,
 414                                          const struct sk_buff *, int);
 415
 416 struct fib6_entry_notifier_info {
 417         struct fib_notifier_info info; /* must be first */
 418         struct fib6_info *rt;
 419         unsigned int nsiblings;
 420 };
 421
 422 /*
 423  *      exported functions
 424  */
 425
 426 struct fib6_table *fib6_get_table(struct net *net, u32 id);
 427 struct fib6_table *fib6_new_table(struct net *net, u32 id);
 428 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 429                                    const struct sk_buff *skb,
 430                                    int flags, pol_lookup_t lookup);
 431
 432 /* called with rcu lock held; can return error pointer
 433  * caller needs to select path
 434  */
 435 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
 436                 struct fib6_result *res, int flags);
 437
 438 /* called with rcu lock held; caller needs to select path */
 439 int fib6_table_lookup(struct net *net, struct fib6_table *table,
 440                       int oif, struct flowi6 *fl6, struct fib6_result *res,
 441                       int strict);
 442
 443 void fib6_select_path(const struct net *net, struct fib6_result *res,
 444                       struct flowi6 *fl6, int oif, bool have_oif_match,
 445                       const struct sk_buff *skb, int strict);
 446 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
 447                                    const struct in6_addr *daddr,
 448                                    const struct in6_addr *saddr);
 449
 450 struct fib6_node *fib6_locate(struct fib6_node *root,
 451                               const struct in6_addr *daddr, int dst_len,
 452                               const struct in6_addr *saddr, int src_len,
 453                               bool exact_match);
 454
 455 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *arg),
 456                     void *arg);
 457 void fib6_clean_all_skip_notify(struct net *net,
 458                                 int (*func)(struct fib6_info *, void *arg),
 459                                 void *arg);
 460
 461 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
 462              struct nl_info *info, struct netlink_ext_ack *extack);
 463 int fib6_del(struct fib6_info *rt, struct nl_info *info);
 464
 465 static inline
 466 void rt6_get_prefsrc(const struct rt6_info *rt, struct in6_addr *addr)
 467 {
 468         const struct fib6_info *from;
 469
 470         rcu_read_lock();
 471
 472         from = rcu_dereference(rt->from);
 473         if (from) {
 474                 *addr = from->fib6_prefsrc.addr;
 475         } else {
 476                 struct in6_addr in6_zero = {};
 477
 478                 *addr = in6_zero;
 479         }
 480
 481         rcu_read_unlock();
 482 }
 483
 484 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
 485                  struct fib6_config *cfg, gfp_t gfp_flags,
 486                  struct netlink_ext_ack *extack);
 487 void fib6_nh_release(struct fib6_nh *fib6_nh);
 488 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh);
 489
 490 int call_fib6_entry_notifiers(struct net *net,
 491                               enum fib_event_type event_type,
 492                               struct fib6_info *rt,
 493                               struct netlink_ext_ack *extack);
 494 int call_fib6_multipath_entry_notifiers(struct net *net,
 495                                         enum fib_event_type event_type,
 496                                         struct fib6_info *rt,
 497                                         unsigned int nsiblings,
 498                                         struct netlink_ext_ack *extack);
 499 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt);
 500 void fib6_rt_update(struct net *net, struct fib6_info *rt,
 501                     struct nl_info *info);
 502 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
 503                      unsigned int flags);
 504
 505 void fib6_run_gc(unsigned long expires, struct net *net, bool force);
 506
 507 void fib6_gc_cleanup(void);
 508
 509 int fib6_init(void);
 510
 511 struct ipv6_route_iter {
 512         struct seq_net_private p;
 513         struct fib6_walker w;
 514         loff_t skip;
 515         struct fib6_table *tbl;
 516         int sernum;
 517 };
 518
 519 extern const struct seq_operations ipv6_route_seq_ops;
 520
 521 int call_fib6_notifier(struct notifier_block *nb,
 522                        enum fib_event_type event_type,
 523                        struct fib_notifier_info *info);
 524 int call_fib6_notifiers(struct net *net, enum fib_event_type event_type,
 525                         struct fib_notifier_info *info);
 526
 527 int __net_init fib6_notifier_init(struct net *net);
 528 void __net_exit fib6_notifier_exit(struct net *net);
 529
 530 unsigned int fib6_tables_seq_read(struct net *net);
 531 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
 532                      struct netlink_ext_ack *extack);
 533
 534 void fib6_update_sernum(struct net *net, struct fib6_info *rt);
 535 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt);
 536 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i);
 537
 538 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val);
 539 static inline bool fib6_metric_locked(struct fib6_info *f6i, int metric)
 540 {
 541         return !!(f6i->fib6_metrics->metrics[RTAX_LOCK - 1] & (1 << metric));
 542 }
 543 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
 544                             bool offload, bool trap, bool offload_failed);
 545
 546 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
 547 struct bpf_iter__ipv6_route {
 548         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 549         __bpf_md_ptr(struct fib6_info *, rt);
 550 };
 551 #endif
 552
 553 INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_output(struct net *net,
 554                                              struct fib6_table *table,
 555                                              struct flowi6 *fl6,
 556                                              const struct sk_buff *skb,
 557                                              int flags));
 558 INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_input(struct net *net,
 559                                              struct fib6_table *table,
 560                                              struct flowi6 *fl6,
 561                                              const struct sk_buff *skb,
 562                                              int flags));
 563 INDIRECT_CALLABLE_DECLARE(struct rt6_info *__ip6_route_redirect(struct net *net,
 564                                              struct fib6_table *table,
 565                                              struct flowi6 *fl6,
 566                                              const struct sk_buff *skb,
 567                                              int flags));
 568 INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_lookup(struct net *net,
 569                                              struct fib6_table *table,
 570                                              struct flowi6 *fl6,
 571                                              const struct sk_buff *skb,
 572                                              int flags));
 573 static inline struct rt6_info *pol_lookup_func(pol_lookup_t lookup,
 574                                                 struct net *net,
 575                                                 struct fib6_table *table,
 576                                                 struct flowi6 *fl6,
 577                                                 const struct sk_buff *skb,
 578                                                 int flags)
 579 {
 580         return INDIRECT_CALL_4(lookup,
 581                                ip6_pol_route_output,
 582                                ip6_pol_route_input,
 583                                ip6_pol_route_lookup,
 584                                __ip6_route_redirect,
 585                                net, table, fl6, skb, flags);
 586 }
 587
 588 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 589 static inline bool fib6_has_custom_rules(const struct net *net)
 590 {
 591         return net->ipv6.fib6_has_custom_rules;
 592 }
 593
 594 int fib6_rules_init(void);
 595 void fib6_rules_cleanup(void);
 596 bool fib6_rule_default(const struct fib_rule *rule);
 597 int fib6_rules_dump(struct net *net, struct notifier_block *nb,
 598                     struct netlink_ext_ack *extack);
 599 unsigned int fib6_rules_seq_read(struct net *net);
 600
 601 static inline bool fib6_rules_early_flow_dissect(struct net *net,
 602                                                  struct sk_buff *skb,
 603                                                  struct flowi6 *fl6,
 604                                                  struct flow_keys *flkeys)
 605 {
 606         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
 607
 608         if (!net->ipv6.fib6_rules_require_fldissect)
 609                 return false;
 610
 611         skb_flow_dissect_flow_keys(skb, flkeys, flag);
 612         fl6->fl6_sport = flkeys->ports.src;
 613         fl6->fl6_dport = flkeys->ports.dst;
 614         fl6->flowi6_proto = flkeys->basic.ip_proto;
 615
 616         return true;
 617 }
 618 #else
 619 static inline bool fib6_has_custom_rules(const struct net *net)
 620 {
 621         return false;
 622 }
 623 static inline int               fib6_rules_init(void)
 624 {
 625         return 0;
 626 }
 627 static inline void              fib6_rules_cleanup(void)
 628 {
 629         return ;
 630 }
 631 static inline bool fib6_rule_default(const struct fib_rule *rule)
 632 {
 633         return true;
 634 }
 635 static inline int fib6_rules_dump(struct net *net, struct notifier_block *nb,
 636                                   struct netlink_ext_ack *extack)
 637 {
 638         return 0;
 639 }
 640 static inline unsigned int fib6_rules_seq_read(struct net *net)
 641 {
 642         return 0;
 643 }
 644 static inline bool fib6_rules_early_flow_dissect(struct net *net,
 645                                                  struct sk_buff *skb,
 646                                                  struct flowi6 *fl6,
 647                                                  struct flow_keys *flkeys)
 648 {
 649         return false;
 650 }
 651 #endif
 652 #endif