11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * INET An implementation of the TCP/IP protocol suite for the LINUX 31da177e4SLinus Torvalds * operating system. INET is implemented using the BSD Socket 41da177e4SLinus Torvalds * interface as the means of communication with the user level. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * IPv4 Forwarding Information Base: policy rules. 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 9e1ef4bf2SThomas Graf * Thomas Graf <tgraf@suug.ch> 101da177e4SLinus Torvalds * 111da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or 121da177e4SLinus Torvalds * modify it under the terms of the GNU General Public License 131da177e4SLinus Torvalds * as published by the Free Software Foundation; either version 141da177e4SLinus Torvalds * 2 of the License, or (at your option) any later version. 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * Fixes: 171da177e4SLinus Torvalds * Rani Assaf : local_rule cannot be deleted 181da177e4SLinus Torvalds * Marc Boucher : routing by fwmark 191da177e4SLinus Torvalds */ 201da177e4SLinus Torvalds 211da177e4SLinus Torvalds #include <linux/types.h> 221da177e4SLinus Torvalds #include <linux/kernel.h> 231da177e4SLinus Torvalds #include <linux/netdevice.h> 241da177e4SLinus Torvalds #include <linux/netlink.h> 25e1ef4bf2SThomas Graf #include <linux/inetdevice.h> 261da177e4SLinus Torvalds #include <linux/init.h> 277b204afdSRobert Olsson #include <linux/list.h> 287b204afdSRobert Olsson #include <linux/rcupdate.h> 291da177e4SLinus Torvalds #include <net/ip.h> 301da177e4SLinus Torvalds #include <net/route.h> 311da177e4SLinus Torvalds #include <net/tcp.h> 321da177e4SLinus Torvalds #include <net/ip_fib.h> 33e1ef4bf2SThomas Graf #include <net/fib_rules.h> 341da177e4SLinus Torvalds 35e1ef4bf2SThomas Graf static struct fib_rules_ops fib4_rules_ops; 361da177e4SLinus Torvalds 37e1ef4bf2SThomas Graf struct fib4_rule 381da177e4SLinus Torvalds { 39e1ef4bf2SThomas Graf struct fib_rule common; 40e1ef4bf2SThomas Graf u8 dst_len; 41e1ef4bf2SThomas Graf u8 src_len; 42e1ef4bf2SThomas Graf u8 tos; 43e1ef4bf2SThomas Graf u32 src; 44e1ef4bf2SThomas Graf u32 srcmask; 45e1ef4bf2SThomas Graf u32 dst; 46e1ef4bf2SThomas Graf u32 dstmask; 471da177e4SLinus Torvalds #ifdef CONFIG_IP_ROUTE_FWMARK 48e1ef4bf2SThomas Graf u32 fwmark; 491da177e4SLinus Torvalds #endif 501da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ROUTE 51e1ef4bf2SThomas Graf u32 tclassid; 521da177e4SLinus Torvalds #endif 531da177e4SLinus Torvalds }; 541da177e4SLinus Torvalds 55e1ef4bf2SThomas Graf static struct fib4_rule default_rule = { 56e1ef4bf2SThomas Graf .common = { 57e1ef4bf2SThomas Graf .refcnt = ATOMIC_INIT(2), 58e1ef4bf2SThomas Graf .pref = 0x7FFF, 59e1ef4bf2SThomas Graf .table = RT_TABLE_DEFAULT, 60e1ef4bf2SThomas Graf .action = FR_ACT_TO_TBL, 61e1ef4bf2SThomas Graf }, 621da177e4SLinus Torvalds }; 631da177e4SLinus Torvalds 64e1ef4bf2SThomas Graf static struct fib4_rule main_rule = { 65e1ef4bf2SThomas Graf .common = { 66e1ef4bf2SThomas Graf .refcnt = ATOMIC_INIT(2), 67e1ef4bf2SThomas Graf .pref = 0x7FFE, 68e1ef4bf2SThomas Graf .table = RT_TABLE_MAIN, 69e1ef4bf2SThomas Graf .action = FR_ACT_TO_TBL, 70e1ef4bf2SThomas Graf }, 711da177e4SLinus Torvalds }; 721da177e4SLinus Torvalds 73e1ef4bf2SThomas Graf static struct fib4_rule local_rule = { 74e1ef4bf2SThomas Graf .common = { 75e1ef4bf2SThomas Graf .refcnt = ATOMIC_INIT(2), 76e1ef4bf2SThomas Graf .table = RT_TABLE_LOCAL, 77e1ef4bf2SThomas Graf .action = FR_ACT_TO_TBL, 78e1ef4bf2SThomas Graf .flags = FIB_RULE_PERMANENT, 79e1ef4bf2SThomas Graf }, 801da177e4SLinus Torvalds }; 811da177e4SLinus Torvalds 82e1ef4bf2SThomas Graf static LIST_HEAD(fib4_rules); 837b204afdSRobert Olsson 84e1ef4bf2SThomas Graf #ifdef CONFIG_NET_CLS_ROUTE 85e1ef4bf2SThomas Graf u32 fib_rules_tclass(struct fib_result *res) 861da177e4SLinus Torvalds { 87e1ef4bf2SThomas Graf return res->r ? ((struct fib4_rule *) res->r)->tclassid : 0; 88e1ef4bf2SThomas Graf } 891da177e4SLinus Torvalds #endif 901da177e4SLinus Torvalds 91e1ef4bf2SThomas Graf int fib_lookup(struct flowi *flp, struct fib_result *res) 92e1ef4bf2SThomas Graf { 93e1ef4bf2SThomas Graf struct fib_lookup_arg arg = { 94e1ef4bf2SThomas Graf .result = res, 95e1ef4bf2SThomas Graf }; 96e1ef4bf2SThomas Graf int err; 97e1ef4bf2SThomas Graf 98e1ef4bf2SThomas Graf err = fib_rules_lookup(&fib4_rules_ops, flp, 0, &arg); 99e1ef4bf2SThomas Graf res->r = arg.rule; 100e1ef4bf2SThomas Graf 1011da177e4SLinus Torvalds return err; 1021da177e4SLinus Torvalds } 1031da177e4SLinus Torvalds 104*8ce11e6aSAdrian Bunk static int fib4_rule_action(struct fib_rule *rule, struct flowi *flp, 105*8ce11e6aSAdrian Bunk int flags, struct fib_lookup_arg *arg) 106e1ef4bf2SThomas Graf { 107e1ef4bf2SThomas Graf int err = -EAGAIN; 108e1ef4bf2SThomas Graf struct fib_table *tbl; 109e1ef4bf2SThomas Graf 110e1ef4bf2SThomas Graf switch (rule->action) { 111e1ef4bf2SThomas Graf case FR_ACT_TO_TBL: 112e1ef4bf2SThomas Graf break; 113e1ef4bf2SThomas Graf 114e1ef4bf2SThomas Graf case FR_ACT_UNREACHABLE: 115e1ef4bf2SThomas Graf err = -ENETUNREACH; 116e1ef4bf2SThomas Graf goto errout; 117e1ef4bf2SThomas Graf 118e1ef4bf2SThomas Graf case FR_ACT_PROHIBIT: 119e1ef4bf2SThomas Graf err = -EACCES; 120e1ef4bf2SThomas Graf goto errout; 121e1ef4bf2SThomas Graf 122e1ef4bf2SThomas Graf case FR_ACT_BLACKHOLE: 123e1ef4bf2SThomas Graf default: 124e1ef4bf2SThomas Graf err = -EINVAL; 125e1ef4bf2SThomas Graf goto errout; 126e1ef4bf2SThomas Graf } 127e1ef4bf2SThomas Graf 128e1ef4bf2SThomas Graf if ((tbl = fib_get_table(rule->table)) == NULL) 129e1ef4bf2SThomas Graf goto errout; 130e1ef4bf2SThomas Graf 131e1ef4bf2SThomas Graf err = tbl->tb_lookup(tbl, flp, (struct fib_result *) arg->result); 132e1ef4bf2SThomas Graf if (err > 0) 133e1ef4bf2SThomas Graf err = -EAGAIN; 134e1ef4bf2SThomas Graf errout: 135e1ef4bf2SThomas Graf return err; 136e1ef4bf2SThomas Graf } 137e1ef4bf2SThomas Graf 138e1ef4bf2SThomas Graf 139e1ef4bf2SThomas Graf void fib_select_default(const struct flowi *flp, struct fib_result *res) 140e1ef4bf2SThomas Graf { 141e1ef4bf2SThomas Graf if (res->r && res->r->action == FR_ACT_TO_TBL && 142e1ef4bf2SThomas Graf FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) { 143e1ef4bf2SThomas Graf struct fib_table *tb; 144e1ef4bf2SThomas Graf if ((tb = fib_get_table(res->r->table)) != NULL) 145e1ef4bf2SThomas Graf tb->tb_select_default(tb, flp, res); 146e1ef4bf2SThomas Graf } 147e1ef4bf2SThomas Graf } 148e1ef4bf2SThomas Graf 149e1ef4bf2SThomas Graf static int fib4_rule_match(struct fib_rule *rule, struct flowi *fl, int flags) 150e1ef4bf2SThomas Graf { 151e1ef4bf2SThomas Graf struct fib4_rule *r = (struct fib4_rule *) rule; 152e1ef4bf2SThomas Graf u32 daddr = fl->fl4_dst; 153e1ef4bf2SThomas Graf u32 saddr = fl->fl4_src; 154e1ef4bf2SThomas Graf 155e1ef4bf2SThomas Graf if (((saddr ^ r->src) & r->srcmask) || 156e1ef4bf2SThomas Graf ((daddr ^ r->dst) & r->dstmask)) 157e1ef4bf2SThomas Graf return 0; 158e1ef4bf2SThomas Graf 159e1ef4bf2SThomas Graf if (r->tos && (r->tos != fl->fl4_tos)) 160e1ef4bf2SThomas Graf return 0; 161e1ef4bf2SThomas Graf 162e1ef4bf2SThomas Graf #ifdef CONFIG_IP_ROUTE_FWMARK 163e1ef4bf2SThomas Graf if (r->fwmark && (r->fwmark != fl->fl4_fwmark)) 164e1ef4bf2SThomas Graf return 0; 165e1ef4bf2SThomas Graf #endif 166e1ef4bf2SThomas Graf 167e1ef4bf2SThomas Graf return 1; 168e1ef4bf2SThomas Graf } 1691da177e4SLinus Torvalds 1701da177e4SLinus Torvalds static struct fib_table *fib_empty_table(void) 1711da177e4SLinus Torvalds { 1721da177e4SLinus Torvalds int id; 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds for (id = 1; id <= RT_TABLE_MAX; id++) 1751da177e4SLinus Torvalds if (fib_tables[id] == NULL) 1761da177e4SLinus Torvalds return __fib_new_table(id); 1771da177e4SLinus Torvalds return NULL; 1781da177e4SLinus Torvalds } 1791da177e4SLinus Torvalds 180e1ef4bf2SThomas Graf static struct nla_policy fib4_rule_policy[FRA_MAX+1] __read_mostly = { 181e1ef4bf2SThomas Graf [FRA_IFNAME] = { .type = NLA_STRING }, 182e1ef4bf2SThomas Graf [FRA_PRIORITY] = { .type = NLA_U32 }, 183e1ef4bf2SThomas Graf [FRA_SRC] = { .type = NLA_U32 }, 184e1ef4bf2SThomas Graf [FRA_DST] = { .type = NLA_U32 }, 185e1ef4bf2SThomas Graf [FRA_FWMARK] = { .type = NLA_U32 }, 186e1ef4bf2SThomas Graf [FRA_FLOW] = { .type = NLA_U32 }, 1871da177e4SLinus Torvalds }; 1881da177e4SLinus Torvalds 189e1ef4bf2SThomas Graf static int fib4_rule_configure(struct fib_rule *rule, struct sk_buff *skb, 190e1ef4bf2SThomas Graf struct nlmsghdr *nlh, struct fib_rule_hdr *frh, 191e1ef4bf2SThomas Graf struct nlattr **tb) 1921da177e4SLinus Torvalds { 193e1ef4bf2SThomas Graf int err = -EINVAL; 194e1ef4bf2SThomas Graf struct fib4_rule *rule4 = (struct fib4_rule *) rule; 1951da177e4SLinus Torvalds 196e1ef4bf2SThomas Graf if (frh->src_len > 32 || frh->dst_len > 32 || 197e1ef4bf2SThomas Graf (frh->tos & ~IPTOS_TOS_MASK)) 198e1ef4bf2SThomas Graf goto errout; 199e1ef4bf2SThomas Graf 200e1ef4bf2SThomas Graf if (rule->table == RT_TABLE_UNSPEC) { 201e1ef4bf2SThomas Graf if (rule->action == FR_ACT_TO_TBL) { 202e1ef4bf2SThomas Graf struct fib_table *table; 203e1ef4bf2SThomas Graf 204e1ef4bf2SThomas Graf table = fib_empty_table(); 205e1ef4bf2SThomas Graf if (table == NULL) { 206e1ef4bf2SThomas Graf err = -ENOBUFS; 207e1ef4bf2SThomas Graf goto errout; 208e1ef4bf2SThomas Graf } 209e1ef4bf2SThomas Graf 210e1ef4bf2SThomas Graf rule->table = table->tb_id; 211e1ef4bf2SThomas Graf } 212e1ef4bf2SThomas Graf } 213e1ef4bf2SThomas Graf 214e1ef4bf2SThomas Graf if (tb[FRA_SRC]) 215e1ef4bf2SThomas Graf rule4->src = nla_get_u32(tb[FRA_SRC]); 216e1ef4bf2SThomas Graf 217e1ef4bf2SThomas Graf if (tb[FRA_DST]) 218e1ef4bf2SThomas Graf rule4->dst = nla_get_u32(tb[FRA_DST]); 219e1ef4bf2SThomas Graf 2201da177e4SLinus Torvalds #ifdef CONFIG_IP_ROUTE_FWMARK 221e1ef4bf2SThomas Graf if (tb[FRA_FWMARK]) 222e1ef4bf2SThomas Graf rule4->fwmark = nla_get_u32(tb[FRA_FWMARK]); 2231da177e4SLinus Torvalds #endif 2241da177e4SLinus Torvalds 2251da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ROUTE 226e1ef4bf2SThomas Graf if (tb[FRA_FLOW]) 227e1ef4bf2SThomas Graf rule4->tclassid = nla_get_u32(tb[FRA_FLOW]); 2281da177e4SLinus Torvalds #endif 2291da177e4SLinus Torvalds 230e1ef4bf2SThomas Graf rule4->src_len = frh->src_len; 231e1ef4bf2SThomas Graf rule4->srcmask = inet_make_mask(rule4->src_len); 232e1ef4bf2SThomas Graf rule4->dst_len = frh->dst_len; 233e1ef4bf2SThomas Graf rule4->dstmask = inet_make_mask(rule4->dst_len); 234e1ef4bf2SThomas Graf rule4->tos = frh->tos; 235e1ef4bf2SThomas Graf 236e1ef4bf2SThomas Graf err = 0; 237e1ef4bf2SThomas Graf errout: 238e1ef4bf2SThomas Graf return err; 2391da177e4SLinus Torvalds } 2401da177e4SLinus Torvalds 241e1ef4bf2SThomas Graf static int fib4_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh, 242e1ef4bf2SThomas Graf struct nlattr **tb) 243a5cdc030SPatrick McHardy { 244e1ef4bf2SThomas Graf struct fib4_rule *rule4 = (struct fib4_rule *) rule; 245a5cdc030SPatrick McHardy 246e1ef4bf2SThomas Graf if (frh->src_len && (rule4->src_len != frh->src_len)) 247e1ef4bf2SThomas Graf return 0; 248e1ef4bf2SThomas Graf 249e1ef4bf2SThomas Graf if (frh->dst_len && (rule4->dst_len != frh->dst_len)) 250e1ef4bf2SThomas Graf return 0; 251e1ef4bf2SThomas Graf 252e1ef4bf2SThomas Graf if (frh->tos && (rule4->tos != frh->tos)) 253e1ef4bf2SThomas Graf return 0; 254e1ef4bf2SThomas Graf 255e1ef4bf2SThomas Graf #ifdef CONFIG_IP_ROUTE_FWMARK 256e1ef4bf2SThomas Graf if (tb[FRA_FWMARK] && (rule4->fwmark != nla_get_u32(tb[FRA_FWMARK]))) 257e1ef4bf2SThomas Graf return 0; 258e1ef4bf2SThomas Graf #endif 259e1ef4bf2SThomas Graf 260e1ef4bf2SThomas Graf #ifdef CONFIG_NET_CLS_ROUTE 261e1ef4bf2SThomas Graf if (tb[FRA_FLOW] && (rule4->tclassid != nla_get_u32(tb[FRA_FLOW]))) 262e1ef4bf2SThomas Graf return 0; 263e1ef4bf2SThomas Graf #endif 264e1ef4bf2SThomas Graf 265e1ef4bf2SThomas Graf if (tb[FRA_SRC] && (rule4->src != nla_get_u32(tb[FRA_SRC]))) 266e1ef4bf2SThomas Graf return 0; 267e1ef4bf2SThomas Graf 268e1ef4bf2SThomas Graf if (tb[FRA_DST] && (rule4->dst != nla_get_u32(tb[FRA_DST]))) 269e1ef4bf2SThomas Graf return 0; 270e1ef4bf2SThomas Graf 271e1ef4bf2SThomas Graf return 1; 272a5cdc030SPatrick McHardy } 273a5cdc030SPatrick McHardy 274e1ef4bf2SThomas Graf static int fib4_rule_fill(struct fib_rule *rule, struct sk_buff *skb, 275e1ef4bf2SThomas Graf struct nlmsghdr *nlh, struct fib_rule_hdr *frh) 2761da177e4SLinus Torvalds { 277e1ef4bf2SThomas Graf struct fib4_rule *rule4 = (struct fib4_rule *) rule; 2781da177e4SLinus Torvalds 279e1ef4bf2SThomas Graf frh->family = AF_INET; 280e1ef4bf2SThomas Graf frh->dst_len = rule4->dst_len; 281e1ef4bf2SThomas Graf frh->src_len = rule4->src_len; 282e1ef4bf2SThomas Graf frh->tos = rule4->tos; 2831da177e4SLinus Torvalds 284e1ef4bf2SThomas Graf #ifdef CONFIG_IP_ROUTE_FWMARK 285e1ef4bf2SThomas Graf if (rule4->fwmark) 286e1ef4bf2SThomas Graf NLA_PUT_U32(skb, FRA_FWMARK, rule4->fwmark); 287e1ef4bf2SThomas Graf #endif 288e1ef4bf2SThomas Graf 289e1ef4bf2SThomas Graf if (rule4->dst_len) 290e1ef4bf2SThomas Graf NLA_PUT_U32(skb, FRA_DST, rule4->dst); 291e1ef4bf2SThomas Graf 292e1ef4bf2SThomas Graf if (rule4->src_len) 293e1ef4bf2SThomas Graf NLA_PUT_U32(skb, FRA_SRC, rule4->src); 294e1ef4bf2SThomas Graf 295e1ef4bf2SThomas Graf #ifdef CONFIG_NET_CLS_ROUTE 296e1ef4bf2SThomas Graf if (rule4->tclassid) 297e1ef4bf2SThomas Graf NLA_PUT_U32(skb, FRA_FLOW, rule4->tclassid); 298e1ef4bf2SThomas Graf #endif 299e1ef4bf2SThomas Graf return 0; 300e1ef4bf2SThomas Graf 301e1ef4bf2SThomas Graf nla_put_failure: 302e1ef4bf2SThomas Graf return -ENOBUFS; 3031da177e4SLinus Torvalds } 3041da177e4SLinus Torvalds 305e1ef4bf2SThomas Graf int fib4_rules_dump(struct sk_buff *skb, struct netlink_callback *cb) 3061da177e4SLinus Torvalds { 307e1ef4bf2SThomas Graf return fib_rules_dump(skb, cb, AF_INET); 308e1ef4bf2SThomas Graf } 309e1ef4bf2SThomas Graf 310e1ef4bf2SThomas Graf static u32 fib4_rule_default_pref(void) 311e1ef4bf2SThomas Graf { 312e1ef4bf2SThomas Graf struct list_head *pos; 313e1ef4bf2SThomas Graf struct fib_rule *rule; 314e1ef4bf2SThomas Graf 315e1ef4bf2SThomas Graf if (!list_empty(&fib4_rules)) { 316e1ef4bf2SThomas Graf pos = fib4_rules.next; 317e1ef4bf2SThomas Graf if (pos->next != &fib4_rules) { 318e1ef4bf2SThomas Graf rule = list_entry(pos->next, struct fib_rule, list); 319e1ef4bf2SThomas Graf if (rule->pref) 320e1ef4bf2SThomas Graf return rule->pref - 1; 321e1ef4bf2SThomas Graf } 322e1ef4bf2SThomas Graf } 323e1ef4bf2SThomas Graf 324e1ef4bf2SThomas Graf return 0; 325e1ef4bf2SThomas Graf } 326e1ef4bf2SThomas Graf 327e1ef4bf2SThomas Graf static struct fib_rules_ops fib4_rules_ops = { 328e1ef4bf2SThomas Graf .family = AF_INET, 329e1ef4bf2SThomas Graf .rule_size = sizeof(struct fib4_rule), 330e1ef4bf2SThomas Graf .action = fib4_rule_action, 331e1ef4bf2SThomas Graf .match = fib4_rule_match, 332e1ef4bf2SThomas Graf .configure = fib4_rule_configure, 333e1ef4bf2SThomas Graf .compare = fib4_rule_compare, 334e1ef4bf2SThomas Graf .fill = fib4_rule_fill, 335e1ef4bf2SThomas Graf .default_pref = fib4_rule_default_pref, 336e1ef4bf2SThomas Graf .nlgroup = RTNLGRP_IPV4_RULE, 337e1ef4bf2SThomas Graf .policy = fib4_rule_policy, 338e1ef4bf2SThomas Graf .rules_list = &fib4_rules, 339e1ef4bf2SThomas Graf .owner = THIS_MODULE, 340e1ef4bf2SThomas Graf }; 341e1ef4bf2SThomas Graf 342e1ef4bf2SThomas Graf void __init fib4_rules_init(void) 343e1ef4bf2SThomas Graf { 344e1ef4bf2SThomas Graf list_add_tail(&local_rule.common.list, &fib4_rules); 345e1ef4bf2SThomas Graf list_add_tail(&main_rule.common.list, &fib4_rules); 346e1ef4bf2SThomas Graf list_add_tail(&default_rule.common.list, &fib4_rules); 347e1ef4bf2SThomas Graf 348e1ef4bf2SThomas Graf fib_rules_register(&fib4_rules_ops); 349e1ef4bf2SThomas Graf } 350e1ef4bf2SThomas Graf 351e1ef4bf2SThomas Graf void __exit fib4_rules_cleanup(void) 352e1ef4bf2SThomas Graf { 353e1ef4bf2SThomas Graf fib_rules_unregister(&fib4_rules_ops); 3541da177e4SLinus Torvalds } 355