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_rules.h>
  23 #include <net/inetpeer.h>
  24 #include <linux/percpu.h>
  25 #include <linux/notifier.h>
  26
  27 struct fib_config {
  28         u8                      fc_dst_len;
  29         u8                      fc_tos;
  30         u8                      fc_protocol;
  31         u8                      fc_scope;
  32         u8                      fc_type;
  33         /* 3 bytes unused */
  34         u32                     fc_table;
  35         __be32                  fc_dst;
  36         __be32                  fc_gw;
  37         int                     fc_oif;
  38         u32                     fc_flags;
  39         u32                     fc_priority;
  40         __be32                  fc_prefsrc;
  41         struct nlattr           *fc_mx;
  42         struct rtnexthop        *fc_mp;
  43         int                     fc_mx_len;
  44         int                     fc_mp_len;
  45         u32                     fc_flow;
  46         u32                     fc_nlflags;
  47         struct nl_info          fc_nlinfo;
  48         struct nlattr           *fc_encap;
  49         u16                     fc_encap_type;
  50 };
  51
  52 struct fib_info;
  53 struct rtable;
  54
  55 struct fib_nh_exception {
  56         struct fib_nh_exception __rcu   *fnhe_next;
  57         int                             fnhe_genid;
  58         __be32                          fnhe_daddr;
  59         u32                             fnhe_pmtu;
  60         __be32                          fnhe_gw;
  61         unsigned long                   fnhe_expires;
  62         struct rtable __rcu             *fnhe_rth_input;
  63         struct rtable __rcu             *fnhe_rth_output;
  64         unsigned long                   fnhe_stamp;
  65         struct rcu_head                 rcu;
  66 };
  67
  68 struct fnhe_hash_bucket {
  69         struct fib_nh_exception __rcu   *chain;
  70 };
  71
  72 #define FNHE_HASH_SHIFT         11
  73 #define FNHE_HASH_SIZE          (1 << FNHE_HASH_SHIFT)
  74 #define FNHE_RECLAIM_DEPTH      5
  75
  76 struct fib_nh {
  77         struct net_device       *nh_dev;
  78         struct hlist_node       nh_hash;
  79         struct fib_info         *nh_parent;
  80         unsigned int            nh_flags;
  81         unsigned char           nh_scope;
  82 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  83         int                     nh_weight;
  84         atomic_t                nh_upper_bound;
  85 #endif
  86 #ifdef CONFIG_IP_ROUTE_CLASSID
  87         __u32                   nh_tclassid;
  88 #endif
  89         int                     nh_oif;
  90         __be32                  nh_gw;
  91         __be32                  nh_saddr;
  92         int                     nh_saddr_genid;
  93         struct rtable __rcu * __percpu *nh_pcpu_rth_output;
  94         struct rtable __rcu     *nh_rth_input;
  95         struct fnhe_hash_bucket __rcu *nh_exceptions;
  96         struct lwtunnel_state   *nh_lwtstate;
  97 };
  98
  99 /*
 100  * This structure contains data shared by many of routes.
 101  */
 102
 103 struct fib_info {
 104         struct hlist_node       fib_hash;
 105         struct hlist_node       fib_lhash;
 106         struct net              *fib_net;
 107         int                     fib_treeref;
 108         atomic_t                fib_clntref;
 109         unsigned int            fib_flags;
 110         unsigned char           fib_dead;
 111         unsigned char           fib_protocol;
 112         unsigned char           fib_scope;
 113         unsigned char           fib_type;
 114         __be32                  fib_prefsrc;
 115         u32                     fib_tb_id;
 116         u32                     fib_priority;
 117         u32                     *fib_metrics;
 118 #define fib_mtu fib_metrics[RTAX_MTU-1]
 119 #define fib_window fib_metrics[RTAX_WINDOW-1]
 120 #define fib_rtt fib_metrics[RTAX_RTT-1]
 121 #define fib_advmss fib_metrics[RTAX_ADVMSS-1]
 122         int                     fib_nhs;
 123 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 124         int                     fib_weight;
 125 #endif
 126         unsigned int            fib_offload_cnt;
 127         struct rcu_head         rcu;
 128         struct fib_nh           fib_nh[0];
 129 #define fib_dev         fib_nh[0].nh_dev
 130 };
 131
 132
 133 #ifdef CONFIG_IP_MULTIPLE_TABLES
 134 struct fib_rule;
 135 #endif
 136
 137 struct fib_table;
 138 struct fib_result {
 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 static inline void fib_info_offload_inc(struct fib_info *fi)
 179 {
 180         fi->fib_offload_cnt++;
 181         fi->fib_flags |= RTNH_F_OFFLOAD;
 182 }
 183
 184 static inline void fib_info_offload_dec(struct fib_info *fi)
 185 {
 186         if (--fi->fib_offload_cnt == 0)
 187                 fi->fib_flags &= ~RTNH_F_OFFLOAD;
 188 }
 189
 190 #define FIB_RES_SADDR(net, res)                         \
 191         ((FIB_RES_NH(res).nh_saddr_genid ==             \
 192           atomic_read(&(net)->ipv4.dev_addr_genid)) ?   \
 193          FIB_RES_NH(res).nh_saddr :                     \
 194          fib_info_update_nh_saddr((net), &FIB_RES_NH(res)))
 195 #define FIB_RES_GW(res)                 (FIB_RES_NH(res).nh_gw)
 196 #define FIB_RES_DEV(res)                (FIB_RES_NH(res).nh_dev)
 197 #define FIB_RES_OIF(res)                (FIB_RES_NH(res).nh_oif)
 198
 199 #define FIB_RES_PREFSRC(net, res)       ((res).fi->fib_prefsrc ? : \
 200                                          FIB_RES_SADDR(net, res))
 201
 202 struct fib_notifier_info {
 203         struct net *net;
 204 };
 205
 206 struct fib_entry_notifier_info {
 207         struct fib_notifier_info info; /* must be first */
 208         u32 dst;
 209         int dst_len;
 210         struct fib_info *fi;
 211         u8 tos;
 212         u8 type;
 213         u32 tb_id;
 214 };
 215
 216 struct fib_rule_notifier_info {
 217         struct fib_notifier_info info; /* must be first */
 218         struct fib_rule *rule;
 219 };
 220
 221 struct fib_nh_notifier_info {
 222         struct fib_notifier_info info; /* must be first */
 223         struct fib_nh *fib_nh;
 224 };
 225
 226 enum fib_event_type {
 227         FIB_EVENT_ENTRY_REPLACE,
 228         FIB_EVENT_ENTRY_APPEND,
 229         FIB_EVENT_ENTRY_ADD,
 230         FIB_EVENT_ENTRY_DEL,
 231         FIB_EVENT_RULE_ADD,
 232         FIB_EVENT_RULE_DEL,
 233         FIB_EVENT_NH_ADD,
 234         FIB_EVENT_NH_DEL,
 235 };
 236
 237 int register_fib_notifier(struct notifier_block *nb,
 238                           void (*cb)(struct notifier_block *nb));
 239 int unregister_fib_notifier(struct notifier_block *nb);
 240 int call_fib_notifier(struct notifier_block *nb, struct net *net,
 241                       enum fib_event_type event_type,
 242                       struct fib_notifier_info *info);
 243 int call_fib_notifiers(struct net *net, enum fib_event_type event_type,
 244                        struct fib_notifier_info *info);
 245
 246 void fib_notify(struct net *net, struct notifier_block *nb);
 247 #ifdef CONFIG_IP_MULTIPLE_TABLES
 248 void fib_rules_notify(struct net *net, struct notifier_block *nb);
 249 #else
 250 static inline void fib_rules_notify(struct net *net, struct notifier_block *nb)
 251 {
 252 }
 253 #endif
 254
 255 struct fib_table {
 256         struct hlist_node       tb_hlist;
 257         u32                     tb_id;
 258         int                     tb_num_default;
 259         struct rcu_head         rcu;
 260         unsigned long           *tb_data;
 261         unsigned long           __data[0];
 262 };
 263
 264 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
 265                      struct fib_result *res, int fib_flags);
 266 int fib_table_insert(struct net *, struct fib_table *, struct fib_config *);
 267 int fib_table_delete(struct net *, struct fib_table *, struct fib_config *);
 268 int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
 269                    struct netlink_callback *cb);
 270 int fib_table_flush(struct net *net, struct fib_table *table);
 271 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
 272 void fib_table_flush_external(struct fib_table *table);
 273 void fib_free_table(struct fib_table *tb);
 274
 275 #ifndef CONFIG_IP_MULTIPLE_TABLES
 276
 277 #define TABLE_LOCAL_INDEX       (RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
 278 #define TABLE_MAIN_INDEX        (RT_TABLE_MAIN  & (FIB_TABLE_HASHSZ - 1))
 279
 280 static inline struct fib_table *fib_get_table(struct net *net, u32 id)
 281 {
 282         struct hlist_node *tb_hlist;
 283         struct hlist_head *ptr;
 284
 285         ptr = id == RT_TABLE_LOCAL ?
 286                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
 287                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
 288
 289         tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
 290
 291         return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
 292 }
 293
 294 static inline struct fib_table *fib_new_table(struct net *net, u32 id)
 295 {
 296         return fib_get_table(net, id);
 297 }
 298
 299 static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
 300                              struct fib_result *res, unsigned int flags)
 301 {
 302         struct fib_table *tb;
 303         int err = -ENETUNREACH;
 304
 305         rcu_read_lock();
 306
 307         tb = fib_get_table(net, RT_TABLE_MAIN);
 308         if (tb)
 309                 err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
 310
 311         if (err == -EAGAIN)
 312                 err = -ENETUNREACH;
 313
 314         rcu_read_unlock();
 315
 316         return err;
 317 }
 318
 319 static inline bool fib4_rule_default(const struct fib_rule *rule)
 320 {
 321         return true;
 322 }
 323
 324 #else /* CONFIG_IP_MULTIPLE_TABLES */
 325 int __net_init fib4_rules_init(struct net *net);
 326 void __net_exit fib4_rules_exit(struct net *net);
 327
 328 struct fib_table *fib_new_table(struct net *net, u32 id);
 329 struct fib_table *fib_get_table(struct net *net, u32 id);
 330
 331 int __fib_lookup(struct net *net, struct flowi4 *flp,
 332                  struct fib_result *res, unsigned int flags);
 333
 334 static inline int fib_lookup(struct net *net, struct flowi4 *flp,
 335                              struct fib_result *res, unsigned int flags)
 336 {
 337         struct fib_table *tb;
 338         int err = -ENETUNREACH;
 339
 340         flags |= FIB_LOOKUP_NOREF;
 341         if (net->ipv4.fib_has_custom_rules)
 342                 return __fib_lookup(net, flp, res, flags);
 343
 344         rcu_read_lock();
 345
 346         res->tclassid = 0;
 347
 348         tb = rcu_dereference_rtnl(net->ipv4.fib_main);
 349         if (tb)
 350                 err = fib_table_lookup(tb, flp, res, flags);
 351
 352         if (!err)
 353                 goto out;
 354
 355         tb = rcu_dereference_rtnl(net->ipv4.fib_default);
 356         if (tb)
 357                 err = fib_table_lookup(tb, flp, res, flags);
 358
 359 out:
 360         if (err == -EAGAIN)
 361                 err = -ENETUNREACH;
 362
 363         rcu_read_unlock();
 364
 365         return err;
 366 }
 367
 368 bool fib4_rule_default(const struct fib_rule *rule);
 369
 370 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 371
 372 /* Exported by fib_frontend.c */
 373 extern const struct nla_policy rtm_ipv4_policy[];
 374 void ip_fib_init(void);
 375 __be32 fib_compute_spec_dst(struct sk_buff *skb);
 376 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 377                         u8 tos, int oif, struct net_device *dev,
 378                         struct in_device *idev, u32 *itag);
 379 #ifdef CONFIG_IP_ROUTE_CLASSID
 380 static inline int fib_num_tclassid_users(struct net *net)
 381 {
 382         return net->ipv4.fib_num_tclassid_users;
 383 }
 384 #else
 385 static inline int fib_num_tclassid_users(struct net *net)
 386 {
 387         return 0;
 388 }
 389 #endif
 390 int fib_unmerge(struct net *net);
 391
 392 /* Exported by fib_semantics.c */
 393 int ip_fib_check_default(__be32 gw, struct net_device *dev);
 394 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
 395 int fib_sync_down_addr(struct net_device *dev, __be32 local);
 396 int fib_sync_up(struct net_device *dev, unsigned int nh_flags);
 397
 398 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 399 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
 400                        const struct sk_buff *skb);
 401 #endif
 402 void fib_select_multipath(struct fib_result *res, int hash);
 403 void fib_select_path(struct net *net, struct fib_result *res,
 404                      struct flowi4 *fl4, const struct sk_buff *skb);
 405
 406 /* Exported by fib_trie.c */
 407 void fib_trie_init(void);
 408 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
 409
 410 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
 411 {
 412 #ifdef CONFIG_IP_ROUTE_CLASSID
 413 #ifdef CONFIG_IP_MULTIPLE_TABLES
 414         u32 rtag;
 415 #endif
 416         *itag = FIB_RES_NH(*res).nh_tclassid<<16;
 417 #ifdef CONFIG_IP_MULTIPLE_TABLES
 418         rtag = res->tclassid;
 419         if (*itag == 0)
 420                 *itag = (rtag<<16);
 421         *itag |= (rtag>>16);
 422 #endif
 423 #endif
 424 }
 425
 426 void free_fib_info(struct fib_info *fi);
 427
 428 static inline void fib_info_hold(struct fib_info *fi)
 429 {
 430         atomic_inc(&fi->fib_clntref);
 431 }
 432
 433 static inline void fib_info_put(struct fib_info *fi)
 434 {
 435         if (atomic_dec_and_test(&fi->fib_clntref))
 436                 free_fib_info(fi);
 437 }
 438
 439 #ifdef CONFIG_PROC_FS
 440 int __net_init fib_proc_init(struct net *net);
 441 void __net_exit fib_proc_exit(struct net *net);
 442 #else
 443 static inline int fib_proc_init(struct net *net)
 444 {
 445         return 0;
 446 }
 447 static inline void fib_proc_exit(struct net *net)
 448 {
 449 }
 450 #endif
 451
 452 #endif  /* _NET_FIB_H */