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
  26 struct fib_config {
  27         u8                      fc_dst_len;
  28         u8                      fc_tos;
  29         u8                      fc_protocol;
  30         u8                      fc_scope;
  31         u8                      fc_type;
  32         /* 3 bytes unused */
  33         u32                     fc_table;
  34         __be32                  fc_dst;
  35         __be32                  fc_gw;
  36         int                     fc_oif;
  37         u32                     fc_flags;
  38         u32                     fc_priority;
  39         __be32                  fc_prefsrc;
  40         struct nlattr           *fc_mx;
  41         struct rtnexthop        *fc_mp;
  42         int                     fc_mx_len;
  43         int                     fc_mp_len;
  44         u32                     fc_flow;
  45         u32                     fc_nlflags;
  46         struct nl_info          fc_nlinfo;
  47         struct nlattr           *fc_encap;
  48         u16                     fc_encap_type;
  49 };
  50
  51 struct fib_info;
  52 struct rtable;
  53
  54 struct fib_nh_exception {
  55         struct fib_nh_exception __rcu   *fnhe_next;
  56         int                             fnhe_genid;
  57         __be32                          fnhe_daddr;
  58         u32                             fnhe_pmtu;
  59         __be32                          fnhe_gw;
  60         unsigned long                   fnhe_expires;
  61         struct rtable __rcu             *fnhe_rth_input;
  62         struct rtable __rcu             *fnhe_rth_output;
  63         unsigned long                   fnhe_stamp;
  64         struct rcu_head                 rcu;
  65 };
  66
  67 struct fnhe_hash_bucket {
  68         struct fib_nh_exception __rcu   *chain;
  69 };
  70
  71 #define FNHE_HASH_SHIFT         11
  72 #define FNHE_HASH_SIZE          (1 << FNHE_HASH_SHIFT)
  73 #define FNHE_RECLAIM_DEPTH      5
  74
  75 struct fib_nh {
  76         struct net_device       *nh_dev;
  77         struct hlist_node       nh_hash;
  78         struct fib_info         *nh_parent;
  79         unsigned int            nh_flags;
  80         unsigned char           nh_scope;
  81 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  82         int                     nh_weight;
  83         atomic_t                nh_upper_bound;
  84 #endif
  85 #ifdef CONFIG_IP_ROUTE_CLASSID
  86         __u32                   nh_tclassid;
  87 #endif
  88         int                     nh_oif;
  89         __be32                  nh_gw;
  90         __be32                  nh_saddr;
  91         int                     nh_saddr_genid;
  92         struct rtable __rcu * __percpu *nh_pcpu_rth_output;
  93         struct rtable __rcu     *nh_rth_input;
  94         struct fnhe_hash_bucket __rcu *nh_exceptions;
  95         struct lwtunnel_state   *nh_lwtstate;
  96 };
  97
  98 /*
  99  * This structure contains data shared by many of routes.
 100  */
 101
 102 struct fib_info {
 103         struct hlist_node       fib_hash;
 104         struct hlist_node       fib_lhash;
 105         struct net              *fib_net;
 106         int                     fib_treeref;
 107         atomic_t                fib_clntref;
 108         unsigned int            fib_flags;
 109         unsigned char           fib_dead;
 110         unsigned char           fib_protocol;
 111         unsigned char           fib_scope;
 112         unsigned char           fib_type;
 113         __be32                  fib_prefsrc;
 114         u32                     fib_priority;
 115         u32                     *fib_metrics;
 116 #define fib_mtu fib_metrics[RTAX_MTU-1]
 117 #define fib_window fib_metrics[RTAX_WINDOW-1]
 118 #define fib_rtt fib_metrics[RTAX_RTT-1]
 119 #define fib_advmss fib_metrics[RTAX_ADVMSS-1]
 120         int                     fib_nhs;
 121 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 122         int                     fib_weight;
 123 #endif
 124         struct rcu_head         rcu;
 125         struct fib_nh           fib_nh[0];
 126 #define fib_dev         fib_nh[0].nh_dev
 127 };
 128
 129
 130 #ifdef CONFIG_IP_MULTIPLE_TABLES
 131 struct fib_rule;
 132 #endif
 133
 134 struct fib_table;
 135 struct fib_result {
 136         unsigned char   prefixlen;
 137         unsigned char   nh_sel;
 138         unsigned char   type;
 139         unsigned char   scope;
 140         u32             tclassid;
 141         struct fib_info *fi;
 142         struct fib_table *table;
 143         struct hlist_head *fa_head;
 144 };
 145
 146 struct fib_result_nl {
 147         __be32          fl_addr;   /* To be looked up*/
 148         u32             fl_mark;
 149         unsigned char   fl_tos;
 150         unsigned char   fl_scope;
 151         unsigned char   tb_id_in;
 152
 153         unsigned char   tb_id;      /* Results */
 154         unsigned char   prefixlen;
 155         unsigned char   nh_sel;
 156         unsigned char   type;
 157         unsigned char   scope;
 158         int             err;
 159 };
 160
 161 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 162 #define FIB_RES_NH(res)         ((res).fi->fib_nh[(res).nh_sel])
 163 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 164 #define FIB_RES_NH(res)         ((res).fi->fib_nh[0])
 165 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 166
 167 #ifdef CONFIG_IP_MULTIPLE_TABLES
 168 #define FIB_TABLE_HASHSZ 256
 169 #else
 170 #define FIB_TABLE_HASHSZ 2
 171 #endif
 172
 173 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh);
 174
 175 #define FIB_RES_SADDR(net, res)                         \
 176         ((FIB_RES_NH(res).nh_saddr_genid ==             \
 177           atomic_read(&(net)->ipv4.dev_addr_genid)) ?   \
 178          FIB_RES_NH(res).nh_saddr :                     \
 179          fib_info_update_nh_saddr((net), &FIB_RES_NH(res)))
 180 #define FIB_RES_GW(res)                 (FIB_RES_NH(res).nh_gw)
 181 #define FIB_RES_DEV(res)                (FIB_RES_NH(res).nh_dev)
 182 #define FIB_RES_OIF(res)                (FIB_RES_NH(res).nh_oif)
 183
 184 #define FIB_RES_PREFSRC(net, res)       ((res).fi->fib_prefsrc ? : \
 185                                          FIB_RES_SADDR(net, res))
 186
 187 struct fib_table {
 188         struct hlist_node       tb_hlist;
 189         u32                     tb_id;
 190         int                     tb_num_default;
 191         struct rcu_head         rcu;
 192         unsigned long           *tb_data;
 193         unsigned long           __data[0];
 194 };
 195
 196 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
 197                      struct fib_result *res, int fib_flags);
 198 int fib_table_insert(struct fib_table *, struct fib_config *);
 199 int fib_table_delete(struct fib_table *, struct fib_config *);
 200 int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
 201                    struct netlink_callback *cb);
 202 int fib_table_flush(struct fib_table *table);
 203 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
 204 void fib_table_flush_external(struct fib_table *table);
 205 void fib_free_table(struct fib_table *tb);
 206
 207 #ifndef CONFIG_IP_MULTIPLE_TABLES
 208
 209 #define TABLE_LOCAL_INDEX       (RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
 210 #define TABLE_MAIN_INDEX        (RT_TABLE_MAIN  & (FIB_TABLE_HASHSZ - 1))
 211
 212 static inline struct fib_table *fib_get_table(struct net *net, u32 id)
 213 {
 214         struct hlist_node *tb_hlist;
 215         struct hlist_head *ptr;
 216
 217         ptr = id == RT_TABLE_LOCAL ?
 218                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
 219                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
 220
 221         tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
 222
 223         return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
 224 }
 225
 226 static inline struct fib_table *fib_new_table(struct net *net, u32 id)
 227 {
 228         return fib_get_table(net, id);
 229 }
 230
 231 static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
 232                              struct fib_result *res, unsigned int flags)
 233 {
 234         struct fib_table *tb;
 235         int err = -ENETUNREACH;
 236
 237         rcu_read_lock();
 238
 239         tb = fib_get_table(net, RT_TABLE_MAIN);
 240         if (tb)
 241                 err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
 242
 243         if (err == -EAGAIN)
 244                 err = -ENETUNREACH;
 245
 246         rcu_read_unlock();
 247
 248         return err;
 249 }
 250
 251 #else /* CONFIG_IP_MULTIPLE_TABLES */
 252 int __net_init fib4_rules_init(struct net *net);
 253 void __net_exit fib4_rules_exit(struct net *net);
 254
 255 struct fib_table *fib_new_table(struct net *net, u32 id);
 256 struct fib_table *fib_get_table(struct net *net, u32 id);
 257
 258 int __fib_lookup(struct net *net, struct flowi4 *flp,
 259                  struct fib_result *res, unsigned int flags);
 260
 261 static inline int fib_lookup(struct net *net, struct flowi4 *flp,
 262                              struct fib_result *res, unsigned int flags)
 263 {
 264         struct fib_table *tb;
 265         int err = -ENETUNREACH;
 266
 267         flags |= FIB_LOOKUP_NOREF;
 268         if (net->ipv4.fib_has_custom_rules)
 269                 return __fib_lookup(net, flp, res, flags);
 270
 271         rcu_read_lock();
 272
 273         res->tclassid = 0;
 274
 275         tb = rcu_dereference_rtnl(net->ipv4.fib_main);
 276         if (tb)
 277                 err = fib_table_lookup(tb, flp, res, flags);
 278
 279         if (!err)
 280                 goto out;
 281
 282         tb = rcu_dereference_rtnl(net->ipv4.fib_default);
 283         if (tb)
 284                 err = fib_table_lookup(tb, flp, res, flags);
 285
 286 out:
 287         if (err == -EAGAIN)
 288                 err = -ENETUNREACH;
 289
 290         rcu_read_unlock();
 291
 292         return err;
 293 }
 294
 295 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 296
 297 /* Exported by fib_frontend.c */
 298 extern const struct nla_policy rtm_ipv4_policy[];
 299 void ip_fib_init(void);
 300 __be32 fib_compute_spec_dst(struct sk_buff *skb);
 301 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 302                         u8 tos, int oif, struct net_device *dev,
 303                         struct in_device *idev, u32 *itag);
 304 void fib_select_default(const struct flowi4 *flp, struct fib_result *res);
 305 #ifdef CONFIG_IP_ROUTE_CLASSID
 306 static inline int fib_num_tclassid_users(struct net *net)
 307 {
 308         return net->ipv4.fib_num_tclassid_users;
 309 }
 310 #else
 311 static inline int fib_num_tclassid_users(struct net *net)
 312 {
 313         return 0;
 314 }
 315 #endif
 316 int fib_unmerge(struct net *net);
 317 void fib_flush_external(struct net *net);
 318
 319 /* Exported by fib_semantics.c */
 320 int ip_fib_check_default(__be32 gw, struct net_device *dev);
 321 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
 322 int fib_sync_down_addr(struct net *net, __be32 local);
 323 int fib_sync_up(struct net_device *dev, unsigned int nh_flags);
 324
 325 extern u32 fib_multipath_secret __read_mostly;
 326
 327 static inline int fib_multipath_hash(__be32 saddr, __be32 daddr)
 328 {
 329         return jhash_2words((__force u32)saddr, (__force u32)daddr,
 330                             fib_multipath_secret) >> 1;
 331 }
 332
 333 void fib_select_multipath(struct fib_result *res, int hash);
 334 void fib_select_path(struct net *net, struct fib_result *res,
 335                      struct flowi4 *fl4, int mp_hash);
 336
 337 /* Exported by fib_trie.c */
 338 void fib_trie_init(void);
 339 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
 340
 341 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
 342 {
 343 #ifdef CONFIG_IP_ROUTE_CLASSID
 344 #ifdef CONFIG_IP_MULTIPLE_TABLES
 345         u32 rtag;
 346 #endif
 347         *itag = FIB_RES_NH(*res).nh_tclassid<<16;
 348 #ifdef CONFIG_IP_MULTIPLE_TABLES
 349         rtag = res->tclassid;
 350         if (*itag == 0)
 351                 *itag = (rtag<<16);
 352         *itag |= (rtag>>16);
 353 #endif
 354 #endif
 355 }
 356
 357 void free_fib_info(struct fib_info *fi);
 358
 359 static inline void fib_info_put(struct fib_info *fi)
 360 {
 361         if (atomic_dec_and_test(&fi->fib_clntref))
 362                 free_fib_info(fi);
 363 }
 364
 365 #ifdef CONFIG_PROC_FS
 366 int __net_init fib_proc_init(struct net *net);
 367 void __net_exit fib_proc_exit(struct net *net);
 368 #else
 369 static inline int fib_proc_init(struct net *net)
 370 {
 371         return 0;
 372 }
 373 static inline void fib_proc_exit(struct net *net)
 374 {
 375 }
 376 #endif
 377
 378 #endif  /* _NET_FIB_H */