include/net/ip_fib.h

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET  is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Definitions for the Forwarding Information Base.
   7  *
   8  * Authors:     A.N.Kuznetsov, <kuznet@ms2.inr.ac.ru>
   9  *
  10  *              This program is free software; you can redistribute it and/or
  11  *              modify it under the terms of the GNU General Public License
  12  *              as published by the Free Software Foundation; either version
  13  *              2 of the License, or (at your option) any later version.
  14  */
  15
  16 #ifndef _NET_IP_FIB_H
  17 #define _NET_IP_FIB_H
  18
  19 #include <net/flow.h>
  20 #include <linux/seq_file.h>
  21 #include <linux/rcupdate.h>
  22 #include <net/fib_notifier.h>
  23 #include <net/fib_rules.h>
  24 #include <net/inetpeer.h>
  25 #include <linux/percpu.h>
  26 #include <linux/notifier.h>
  27 #include <linux/refcount.h>
  28
  29 struct fib_config {
  30         u8                      fc_dst_len;
  31         u8                      fc_tos;
  32         u8                      fc_protocol;
  33         u8                      fc_scope;
  34         u8                      fc_type;
  35         /* 3 bytes unused */
  36         u32                     fc_table;
  37         __be32                  fc_dst;
  38         __be32                  fc_gw;
  39         int                     fc_oif;
  40         u32                     fc_flags;
  41         u32                     fc_priority;
  42         __be32                  fc_prefsrc;
  43         struct nlattr           *fc_mx;
  44         struct rtnexthop        *fc_mp;
  45         int                     fc_mx_len;
  46         int                     fc_mp_len;
  47         u32                     fc_flow;
  48         u32                     fc_nlflags;
  49         struct nl_info          fc_nlinfo;
  50         struct nlattr           *fc_encap;
  51         u16                     fc_encap_type;
  52 };
  53
  54 struct fib_info;
  55 struct rtable;
  56
  57 struct fib_nh_exception {
  58         struct fib_nh_exception __rcu   *fnhe_next;
  59         int                             fnhe_genid;
  60         __be32                          fnhe_daddr;
  61         u32                             fnhe_pmtu;
  62         bool                            fnhe_mtu_locked;
  63         __be32                          fnhe_gw;
  64         unsigned long                   fnhe_expires;
  65         struct rtable __rcu             *fnhe_rth_input;
  66         struct rtable __rcu             *fnhe_rth_output;
  67         unsigned long                   fnhe_stamp;
  68         struct rcu_head                 rcu;
  69 };
  70
  71 struct fnhe_hash_bucket {
  72         struct fib_nh_exception __rcu   *chain;
  73 };
  74
  75 #define FNHE_HASH_SHIFT         11
  76 #define FNHE_HASH_SIZE          (1 << FNHE_HASH_SHIFT)
  77 #define FNHE_RECLAIM_DEPTH      5
  78
  79 struct fib_nh {
  80         struct net_device       *nh_dev;
  81         struct hlist_node       nh_hash;
  82         struct fib_info         *nh_parent;
  83         unsigned int            nh_flags;
  84         unsigned char           nh_scope;
  85 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  86         int                     nh_weight;
  87         atomic_t                nh_upper_bound;
  88 #endif
  89 #ifdef CONFIG_IP_ROUTE_CLASSID
  90         __u32                   nh_tclassid;
  91 #endif
  92         int                     nh_oif;
  93         __be32                  nh_gw;
  94         __be32                  nh_saddr;
  95         int                     nh_saddr_genid;
  96         struct rtable __rcu * __percpu *nh_pcpu_rth_output;
  97         struct rtable __rcu     *nh_rth_input;
  98         struct fnhe_hash_bucket __rcu *nh_exceptions;
  99         struct lwtunnel_state   *nh_lwtstate;
 100 };
 101
 102 /*
 103  * This structure contains data shared by many of routes.
 104  */
 105
 106 struct fib_info {
 107         struct hlist_node       fib_hash;
 108         struct hlist_node       fib_lhash;
 109         struct net              *fib_net;
 110         int                     fib_treeref;
 111         refcount_t              fib_clntref;
 112         unsigned int            fib_flags;
 113         unsigned char           fib_dead;
 114         unsigned char           fib_protocol;
 115         unsigned char           fib_scope;
 116         unsigned char           fib_type;
 117         __be32                  fib_prefsrc;
 118         u32                     fib_tb_id;
 119         u32                     fib_priority;
 120         struct dst_metrics      *fib_metrics;
 121 #define fib_mtu fib_metrics->metrics[RTAX_MTU-1]
 122 #define fib_window fib_metrics->metrics[RTAX_WINDOW-1]
 123 #define fib_rtt fib_metrics->metrics[RTAX_RTT-1]
 124 #define fib_advmss fib_metrics->metrics[RTAX_ADVMSS-1]
 125         int                     fib_nhs;
 126         struct rcu_head         rcu;
 127         struct fib_nh           fib_nh[0];
 128 #define fib_dev         fib_nh[0].nh_dev
 129 };
 130
 131
 132 #ifdef CONFIG_IP_MULTIPLE_TABLES
 133 struct fib_rule;
 134 #endif
 135
 136 struct fib_table;
 137 struct fib_result {
 138         __be32          prefix;
 139         unsigned char   prefixlen;
 140         unsigned char   nh_sel;
 141         unsigned char   type;
 142         unsigned char   scope;
 143         u32             tclassid;
 144         struct fib_info *fi;
 145         struct fib_table *table;
 146         struct hlist_head *fa_head;
 147 };
 148
 149 struct fib_result_nl {
 150         __be32          fl_addr;   /* To be looked up*/
 151         u32             fl_mark;
 152         unsigned char   fl_tos;
 153         unsigned char   fl_scope;
 154         unsigned char   tb_id_in;
 155
 156         unsigned char   tb_id;      /* Results */
 157         unsigned char   prefixlen;
 158         unsigned char   nh_sel;
 159         unsigned char   type;
 160         unsigned char   scope;
 161         int             err;
 162 };
 163
 164 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 165 #define FIB_RES_NH(res)         ((res).fi->fib_nh[(res).nh_sel])
 166 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 167 #define FIB_RES_NH(res)         ((res).fi->fib_nh[0])
 168 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 169
 170 #ifdef CONFIG_IP_MULTIPLE_TABLES
 171 #define FIB_TABLE_HASHSZ 256
 172 #else
 173 #define FIB_TABLE_HASHSZ 2
 174 #endif
 175
 176 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh);
 177
 178 #define FIB_RES_SADDR(net, res)                         \
 179         ((FIB_RES_NH(res).nh_saddr_genid ==             \
 180           atomic_read(&(net)->ipv4.dev_addr_genid)) ?   \
 181          FIB_RES_NH(res).nh_saddr :                     \
 182          fib_info_update_nh_saddr((net), &FIB_RES_NH(res)))
 183 #define FIB_RES_GW(res)                 (FIB_RES_NH(res).nh_gw)
 184 #define FIB_RES_DEV(res)                (FIB_RES_NH(res).nh_dev)
 185 #define FIB_RES_OIF(res)                (FIB_RES_NH(res).nh_oif)
 186
 187 #define FIB_RES_PREFSRC(net, res)       ((res).fi->fib_prefsrc ? : \
 188                                          FIB_RES_SADDR(net, res))
 189
 190 struct fib_entry_notifier_info {
 191         struct fib_notifier_info info; /* must be first */
 192         u32 dst;
 193         int dst_len;
 194         struct fib_info *fi;
 195         u8 tos;
 196         u8 type;
 197         u32 tb_id;
 198 };
 199
 200 struct fib_nh_notifier_info {
 201         struct fib_notifier_info info; /* must be first */
 202         struct fib_nh *fib_nh;
 203 };
 204
 205 int call_fib4_notifier(struct notifier_block *nb, struct net *net,
 206                        enum fib_event_type event_type,
 207                        struct fib_notifier_info *info);
 208 int call_fib4_notifiers(struct net *net, enum fib_event_type event_type,
 209                         struct fib_notifier_info *info);
 210
 211 int __net_init fib4_notifier_init(struct net *net);
 212 void __net_exit fib4_notifier_exit(struct net *net);
 213
 214 void fib_notify(struct net *net, struct notifier_block *nb);
 215
 216 struct fib_table {
 217         struct hlist_node       tb_hlist;
 218         u32                     tb_id;
 219         int                     tb_num_default;
 220         struct rcu_head         rcu;
 221         unsigned long           *tb_data;
 222         unsigned long           __data[0];
 223 };
 224
 225 struct fib_dump_filter {
 226         u32                     table_id;
 227         /* filter_set is an optimization that an entry is set */
 228         bool                    filter_set;
 229         bool                    dump_all_families;
 230         unsigned char           protocol;
 231         unsigned char           rt_type;
 232         unsigned int            flags;
 233         struct net_device       *dev;
 234 };
 235
 236 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
 237                      struct fib_result *res, int fib_flags);
 238 int fib_table_insert(struct net *, struct fib_table *, struct fib_config *,
 239                      struct netlink_ext_ack *extack);
 240 int fib_table_delete(struct net *, struct fib_table *, struct fib_config *,
 241                      struct netlink_ext_ack *extack);
 242 int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
 243                    struct netlink_callback *cb, struct fib_dump_filter *filter);
 244 int fib_table_flush(struct net *net, struct fib_table *table, bool flush_all);
 245 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
 246 void fib_table_flush_external(struct fib_table *table);
 247 void fib_free_table(struct fib_table *tb);
 248
 249 #ifndef CONFIG_IP_MULTIPLE_TABLES
 250
 251 #define TABLE_LOCAL_INDEX       (RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
 252 #define TABLE_MAIN_INDEX        (RT_TABLE_MAIN  & (FIB_TABLE_HASHSZ - 1))
 253
 254 static inline struct fib_table *fib_get_table(struct net *net, u32 id)
 255 {
 256         struct hlist_node *tb_hlist;
 257         struct hlist_head *ptr;
 258
 259         ptr = id == RT_TABLE_LOCAL ?
 260                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
 261                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
 262
 263         tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
 264
 265         return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
 266 }
 267
 268 static inline struct fib_table *fib_new_table(struct net *net, u32 id)
 269 {
 270         return fib_get_table(net, id);
 271 }
 272
 273 static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
 274                              struct fib_result *res, unsigned int flags)
 275 {
 276         struct fib_table *tb;
 277         int err = -ENETUNREACH;
 278
 279         rcu_read_lock();
 280
 281         tb = fib_get_table(net, RT_TABLE_MAIN);
 282         if (tb)
 283                 err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
 284
 285         if (err == -EAGAIN)
 286                 err = -ENETUNREACH;
 287
 288         rcu_read_unlock();
 289
 290         return err;
 291 }
 292
 293 static inline bool fib4_rule_default(const struct fib_rule *rule)
 294 {
 295         return true;
 296 }
 297
 298 static inline int fib4_rules_dump(struct net *net, struct notifier_block *nb)
 299 {
 300         return 0;
 301 }
 302
 303 static inline unsigned int fib4_rules_seq_read(struct net *net)
 304 {
 305         return 0;
 306 }
 307
 308 static inline bool fib4_rules_early_flow_dissect(struct net *net,
 309                                                  struct sk_buff *skb,
 310                                                  struct flowi4 *fl4,
 311                                                  struct flow_keys *flkeys)
 312 {
 313         return false;
 314 }
 315 #else /* CONFIG_IP_MULTIPLE_TABLES */
 316 int __net_init fib4_rules_init(struct net *net);
 317 void __net_exit fib4_rules_exit(struct net *net);
 318
 319 struct fib_table *fib_new_table(struct net *net, u32 id);
 320 struct fib_table *fib_get_table(struct net *net, u32 id);
 321
 322 int __fib_lookup(struct net *net, struct flowi4 *flp,
 323                  struct fib_result *res, unsigned int flags);
 324
 325 static inline int fib_lookup(struct net *net, struct flowi4 *flp,
 326                              struct fib_result *res, unsigned int flags)
 327 {
 328         struct fib_table *tb;
 329         int err = -ENETUNREACH;
 330
 331         flags |= FIB_LOOKUP_NOREF;
 332         if (net->ipv4.fib_has_custom_rules)
 333                 return __fib_lookup(net, flp, res, flags);
 334
 335         rcu_read_lock();
 336
 337         res->tclassid = 0;
 338
 339         tb = rcu_dereference_rtnl(net->ipv4.fib_main);
 340         if (tb)
 341                 err = fib_table_lookup(tb, flp, res, flags);
 342
 343         if (!err)
 344                 goto out;
 345
 346         tb = rcu_dereference_rtnl(net->ipv4.fib_default);
 347         if (tb)
 348                 err = fib_table_lookup(tb, flp, res, flags);
 349
 350 out:
 351         if (err == -EAGAIN)
 352                 err = -ENETUNREACH;
 353
 354         rcu_read_unlock();
 355
 356         return err;
 357 }
 358
 359 bool fib4_rule_default(const struct fib_rule *rule);
 360 int fib4_rules_dump(struct net *net, struct notifier_block *nb);
 361 unsigned int fib4_rules_seq_read(struct net *net);
 362
 363 static inline bool fib4_rules_early_flow_dissect(struct net *net,
 364                                                  struct sk_buff *skb,
 365                                                  struct flowi4 *fl4,
 366                                                  struct flow_keys *flkeys)
 367 {
 368         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
 369
 370         if (!net->ipv4.fib_rules_require_fldissect)
 371                 return false;
 372
 373         skb_flow_dissect_flow_keys(skb, flkeys, flag);
 374         fl4->fl4_sport = flkeys->ports.src;
 375         fl4->fl4_dport = flkeys->ports.dst;
 376         fl4->flowi4_proto = flkeys->basic.ip_proto;
 377
 378         return true;
 379 }
 380
 381 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 382
 383 /* Exported by fib_frontend.c */
 384 extern const struct nla_policy rtm_ipv4_policy[];
 385 void ip_fib_init(void);
 386 __be32 fib_compute_spec_dst(struct sk_buff *skb);
 387 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev);
 388 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 389                         u8 tos, int oif, struct net_device *dev,
 390                         struct in_device *idev, u32 *itag);
 391 #ifdef CONFIG_IP_ROUTE_CLASSID
 392 static inline int fib_num_tclassid_users(struct net *net)
 393 {
 394         return net->ipv4.fib_num_tclassid_users;
 395 }
 396 #else
 397 static inline int fib_num_tclassid_users(struct net *net)
 398 {
 399         return 0;
 400 }
 401 #endif
 402 int fib_unmerge(struct net *net);
 403
 404 /* Exported by fib_semantics.c */
 405 int ip_fib_check_default(__be32 gw, struct net_device *dev);
 406 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
 407 int fib_sync_down_addr(struct net_device *dev, __be32 local);
 408 int fib_sync_up(struct net_device *dev, unsigned int nh_flags);
 409 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu);
 410
 411 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 412 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
 413                        const struct sk_buff *skb, struct flow_keys *flkeys);
 414 #endif
 415 void fib_select_multipath(struct fib_result *res, int hash);
 416 void fib_select_path(struct net *net, struct fib_result *res,
 417                      struct flowi4 *fl4, const struct sk_buff *skb);
 418
 419 /* Exported by fib_trie.c */
 420 void fib_trie_init(void);
 421 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
 422
 423 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
 424 {
 425 #ifdef CONFIG_IP_ROUTE_CLASSID
 426 #ifdef CONFIG_IP_MULTIPLE_TABLES
 427         u32 rtag;
 428 #endif
 429         *itag = FIB_RES_NH(*res).nh_tclassid<<16;
 430 #ifdef CONFIG_IP_MULTIPLE_TABLES
 431         rtag = res->tclassid;
 432         if (*itag == 0)
 433                 *itag = (rtag<<16);
 434         *itag |= (rtag>>16);
 435 #endif
 436 #endif
 437 }
 438
 439 void free_fib_info(struct fib_info *fi);
 440
 441 static inline void fib_info_hold(struct fib_info *fi)
 442 {
 443         refcount_inc(&fi->fib_clntref);
 444 }
 445
 446 static inline void fib_info_put(struct fib_info *fi)
 447 {
 448         if (refcount_dec_and_test(&fi->fib_clntref))
 449                 free_fib_info(fi);
 450 }
 451
 452 #ifdef CONFIG_PROC_FS
 453 int __net_init fib_proc_init(struct net *net);
 454 void __net_exit fib_proc_exit(struct net *net);
 455 #else
 456 static inline int fib_proc_init(struct net *net)
 457 {
 458         return 0;
 459 }
 460 static inline void fib_proc_exit(struct net *net)
 461 {
 462 }
 463 #endif
 464
 465 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr);
 466
 467 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
 468                           struct fib_dump_filter *filter,
 469                           struct netlink_callback *cb);
 470 #endif  /* _NET_FIB_H */