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         bool                            fnhe_mtu_locked;
  61         __be32                          fnhe_gw;
  62         unsigned long                   fnhe_expires;
  63         struct rtable __rcu             *fnhe_rth_input;
  64         struct rtable __rcu             *fnhe_rth_output;
  65         unsigned long                   fnhe_stamp;
  66         struct rcu_head                 rcu;
  67 };
  68
  69 struct fnhe_hash_bucket {
  70         struct fib_nh_exception __rcu   *chain;
  71 };
  72
  73 #define FNHE_HASH_SHIFT         11
  74 #define FNHE_HASH_SIZE          (1 << FNHE_HASH_SHIFT)
  75 #define FNHE_RECLAIM_DEPTH      5
  76
  77 struct fib_nh {
  78         struct net_device       *nh_dev;
  79         struct hlist_node       nh_hash;
  80         struct fib_info         *nh_parent;
  81         unsigned int            nh_flags;
  82         unsigned char           nh_scope;
  83 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  84         int                     nh_weight;
  85         atomic_t                nh_upper_bound;
  86 #endif
  87 #ifdef CONFIG_IP_ROUTE_CLASSID
  88         __u32                   nh_tclassid;
  89 #endif
  90         int                     nh_oif;
  91         __be32                  nh_gw;
  92         __be32                  nh_saddr;
  93         int                     nh_saddr_genid;
  94         struct rtable __rcu * __percpu *nh_pcpu_rth_output;
  95         struct rtable __rcu     *nh_rth_input;
  96         struct fnhe_hash_bucket __rcu *nh_exceptions;
  97         struct lwtunnel_state   *nh_lwtstate;
  98 };
  99
 100 /*
 101  * This structure contains data shared by many of routes.
 102  */
 103
 104 struct fib_info {
 105         struct hlist_node       fib_hash;
 106         struct hlist_node       fib_lhash;
 107         struct net              *fib_net;
 108         int                     fib_treeref;
 109         atomic_t                fib_clntref;
 110         unsigned int            fib_flags;
 111         unsigned char           fib_dead;
 112         unsigned char           fib_protocol;
 113         unsigned char           fib_scope;
 114         unsigned char           fib_type;
 115         __be32                  fib_prefsrc;
 116         u32                     fib_tb_id;
 117         u32                     fib_priority;
 118         struct dst_metrics      *fib_metrics;
 119 #define fib_mtu fib_metrics->metrics[RTAX_MTU-1]
 120 #define fib_window fib_metrics->metrics[RTAX_WINDOW-1]
 121 #define fib_rtt fib_metrics->metrics[RTAX_RTT-1]
 122 #define fib_advmss fib_metrics->metrics[RTAX_ADVMSS-1]
 123         int                     fib_nhs;
 124 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 125         int                     fib_weight;
 126 #endif
 127         unsigned int            fib_offload_cnt;
 128         struct rcu_head         rcu;
 129         struct fib_nh           fib_nh[0];
 130 #define fib_dev         fib_nh[0].nh_dev
 131 };
 132
 133
 134 #ifdef CONFIG_IP_MULTIPLE_TABLES
 135 struct fib_rule;
 136 #endif
 137
 138 struct fib_table;
 139 struct fib_result {
 140         unsigned char   prefixlen;
 141         unsigned char   nh_sel;
 142         unsigned char   type;
 143         unsigned char   scope;
 144         u32             tclassid;
 145         struct fib_info *fi;
 146         struct fib_table *table;
 147         struct hlist_head *fa_head;
 148 };
 149
 150 struct fib_result_nl {
 151         __be32          fl_addr;   /* To be looked up*/
 152         u32             fl_mark;
 153         unsigned char   fl_tos;
 154         unsigned char   fl_scope;
 155         unsigned char   tb_id_in;
 156
 157         unsigned char   tb_id;      /* Results */
 158         unsigned char   prefixlen;
 159         unsigned char   nh_sel;
 160         unsigned char   type;
 161         unsigned char   scope;
 162         int             err;
 163 };
 164
 165 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 166 #define FIB_RES_NH(res)         ((res).fi->fib_nh[(res).nh_sel])
 167 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 168 #define FIB_RES_NH(res)         ((res).fi->fib_nh[0])
 169 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 170
 171 #ifdef CONFIG_IP_MULTIPLE_TABLES
 172 #define FIB_TABLE_HASHSZ 256
 173 #else
 174 #define FIB_TABLE_HASHSZ 2
 175 #endif
 176
 177 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh);
 178
 179 static inline void fib_info_offload_inc(struct fib_info *fi)
 180 {
 181         fi->fib_offload_cnt++;
 182         fi->fib_flags |= RTNH_F_OFFLOAD;
 183 }
 184
 185 static inline void fib_info_offload_dec(struct fib_info *fi)
 186 {
 187         if (--fi->fib_offload_cnt == 0)
 188                 fi->fib_flags &= ~RTNH_F_OFFLOAD;
 189 }
 190
 191 #define FIB_RES_SADDR(net, res)                         \
 192         ((FIB_RES_NH(res).nh_saddr_genid ==             \
 193           atomic_read(&(net)->ipv4.dev_addr_genid)) ?   \
 194          FIB_RES_NH(res).nh_saddr :                     \
 195          fib_info_update_nh_saddr((net), &FIB_RES_NH(res)))
 196 #define FIB_RES_GW(res)                 (FIB_RES_NH(res).nh_gw)
 197 #define FIB_RES_DEV(res)                (FIB_RES_NH(res).nh_dev)
 198 #define FIB_RES_OIF(res)                (FIB_RES_NH(res).nh_oif)
 199
 200 #define FIB_RES_PREFSRC(net, res)       ((res).fi->fib_prefsrc ? : \
 201                                          FIB_RES_SADDR(net, res))
 202
 203 struct fib_notifier_info {
 204         struct net *net;
 205 };
 206
 207 struct fib_entry_notifier_info {
 208         struct fib_notifier_info info; /* must be first */
 209         u32 dst;
 210         int dst_len;
 211         struct fib_info *fi;
 212         u8 tos;
 213         u8 type;
 214         u32 tb_id;
 215         u32 nlflags;
 216 };
 217
 218 enum fib_event_type {
 219         FIB_EVENT_ENTRY_ADD,
 220         FIB_EVENT_ENTRY_DEL,
 221         FIB_EVENT_RULE_ADD,
 222         FIB_EVENT_RULE_DEL,
 223 };
 224
 225 int register_fib_notifier(struct notifier_block *nb);
 226 int unregister_fib_notifier(struct notifier_block *nb);
 227 int call_fib_notifiers(struct net *net, enum fib_event_type event_type,
 228                        struct fib_notifier_info *info);
 229
 230 struct fib_table {
 231         struct hlist_node       tb_hlist;
 232         u32                     tb_id;
 233         int                     tb_num_default;
 234         struct rcu_head         rcu;
 235         unsigned long           *tb_data;
 236         unsigned long           __data[0];
 237 };
 238
 239 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
 240                      struct fib_result *res, int fib_flags);
 241 int fib_table_insert(struct net *, struct fib_table *, struct fib_config *);
 242 int fib_table_delete(struct net *, struct fib_table *, struct fib_config *);
 243 int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
 244                    struct netlink_callback *cb);
 245 int fib_table_flush(struct net *net, struct fib_table *table, bool flush_all);
 246 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
 247 void fib_table_flush_external(struct fib_table *table);
 248 void fib_free_table(struct fib_table *tb);
 249
 250 #ifndef CONFIG_IP_MULTIPLE_TABLES
 251
 252 #define TABLE_LOCAL_INDEX       (RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
 253 #define TABLE_MAIN_INDEX        (RT_TABLE_MAIN  & (FIB_TABLE_HASHSZ - 1))
 254
 255 static inline struct fib_table *fib_get_table(struct net *net, u32 id)
 256 {
 257         struct hlist_node *tb_hlist;
 258         struct hlist_head *ptr;
 259
 260         ptr = id == RT_TABLE_LOCAL ?
 261                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
 262                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
 263
 264         tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
 265
 266         return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
 267 }
 268
 269 static inline struct fib_table *fib_new_table(struct net *net, u32 id)
 270 {
 271         return fib_get_table(net, id);
 272 }
 273
 274 static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
 275                              struct fib_result *res, unsigned int flags)
 276 {
 277         struct fib_table *tb;
 278         int err = -ENETUNREACH;
 279
 280         rcu_read_lock();
 281
 282         tb = fib_get_table(net, RT_TABLE_MAIN);
 283         if (tb)
 284                 err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
 285
 286         if (err == -EAGAIN)
 287                 err = -ENETUNREACH;
 288
 289         rcu_read_unlock();
 290
 291         return err;
 292 }
 293
 294 #else /* CONFIG_IP_MULTIPLE_TABLES */
 295 int __net_init fib4_rules_init(struct net *net);
 296 void __net_exit fib4_rules_exit(struct net *net);
 297
 298 struct fib_table *fib_new_table(struct net *net, u32 id);
 299 struct fib_table *fib_get_table(struct net *net, u32 id);
 300
 301 int __fib_lookup(struct net *net, struct flowi4 *flp,
 302                  struct fib_result *res, unsigned int flags);
 303
 304 static inline int fib_lookup(struct net *net, struct flowi4 *flp,
 305                              struct fib_result *res, unsigned int flags)
 306 {
 307         struct fib_table *tb;
 308         int err = -ENETUNREACH;
 309
 310         flags |= FIB_LOOKUP_NOREF;
 311         if (net->ipv4.fib_has_custom_rules)
 312                 return __fib_lookup(net, flp, res, flags);
 313
 314         rcu_read_lock();
 315
 316         res->tclassid = 0;
 317
 318         tb = rcu_dereference_rtnl(net->ipv4.fib_main);
 319         if (tb)
 320                 err = fib_table_lookup(tb, flp, res, flags);
 321
 322         if (!err)
 323                 goto out;
 324
 325         tb = rcu_dereference_rtnl(net->ipv4.fib_default);
 326         if (tb)
 327                 err = fib_table_lookup(tb, flp, res, flags);
 328
 329 out:
 330         if (err == -EAGAIN)
 331                 err = -ENETUNREACH;
 332
 333         rcu_read_unlock();
 334
 335         return err;
 336 }
 337
 338 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 339
 340 /* Exported by fib_frontend.c */
 341 extern const struct nla_policy rtm_ipv4_policy[];
 342 void ip_fib_init(void);
 343 __be32 fib_compute_spec_dst(struct sk_buff *skb);
 344 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 345                         u8 tos, int oif, struct net_device *dev,
 346                         struct in_device *idev, u32 *itag);
 347 void fib_select_default(const struct flowi4 *flp, struct fib_result *res);
 348 #ifdef CONFIG_IP_ROUTE_CLASSID
 349 static inline int fib_num_tclassid_users(struct net *net)
 350 {
 351         return net->ipv4.fib_num_tclassid_users;
 352 }
 353 #else
 354 static inline int fib_num_tclassid_users(struct net *net)
 355 {
 356         return 0;
 357 }
 358 #endif
 359 int fib_unmerge(struct net *net);
 360
 361 /* Exported by fib_semantics.c */
 362 int ip_fib_check_default(__be32 gw, struct net_device *dev);
 363 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
 364 int fib_sync_down_addr(struct net_device *dev, __be32 local);
 365 int fib_sync_up(struct net_device *dev, unsigned int nh_flags);
 366 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu);
 367
 368 extern u32 fib_multipath_secret __read_mostly;
 369
 370 static inline int fib_multipath_hash(__be32 saddr, __be32 daddr)
 371 {
 372         return jhash_2words((__force u32)saddr, (__force u32)daddr,
 373                             fib_multipath_secret) >> 1;
 374 }
 375
 376 void fib_select_multipath(struct fib_result *res, int hash);
 377 void fib_select_path(struct net *net, struct fib_result *res,
 378                      struct flowi4 *fl4, int mp_hash);
 379
 380 /* Exported by fib_trie.c */
 381 void fib_trie_init(void);
 382 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
 383
 384 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
 385 {
 386 #ifdef CONFIG_IP_ROUTE_CLASSID
 387 #ifdef CONFIG_IP_MULTIPLE_TABLES
 388         u32 rtag;
 389 #endif
 390         *itag = FIB_RES_NH(*res).nh_tclassid<<16;
 391 #ifdef CONFIG_IP_MULTIPLE_TABLES
 392         rtag = res->tclassid;
 393         if (*itag == 0)
 394                 *itag = (rtag<<16);
 395         *itag |= (rtag>>16);
 396 #endif
 397 #endif
 398 }
 399
 400 void free_fib_info(struct fib_info *fi);
 401
 402 static inline void fib_info_put(struct fib_info *fi)
 403 {
 404         if (atomic_dec_and_test(&fi->fib_clntref))
 405                 free_fib_info(fi);
 406 }
 407
 408 #ifdef CONFIG_PROC_FS
 409 int __net_init fib_proc_init(struct net *net);
 410 void __net_exit fib_proc_exit(struct net *net);
 411 #else
 412 static inline int fib_proc_init(struct net *net)
 413 {
 414         return 0;
 415 }
 416 static inline void fib_proc_exit(struct net *net)
 417 {
 418 }
 419 #endif
 420
 421 #endif  /* _NET_FIB_H */