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