net/ipv4/fib_rules.c

   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  *              IPv4 Forwarding Information Base: policy rules.
   7  *
   8  * Version:     $Id: fib_rules.c,v 1.11 1999/06/09 10:10:47 davem Exp $
   9  *
  10  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * Fixes:
  18  *              Rani Assaf      :       local_rule cannot be deleted
  19  *              Marc Boucher    :       routing by fwmark
  20  */
  21
  22 #include <linux/config.h>
  23 #include <asm/uaccess.h>
  24 #include <asm/system.h>
  25 #include <asm/bitops.h>
  26 #include <linux/types.h>
  27 #include <linux/kernel.h>
  28 #include <linux/sched.h>
  29 #include <linux/mm.h>
  30 #include <linux/string.h>
  31 #include <linux/socket.h>
  32 #include <linux/sockios.h>
  33 #include <linux/errno.h>
  34 #include <linux/in.h>
  35 #include <linux/inet.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/if_arp.h>
  38 #include <linux/proc_fs.h>
  39 #include <linux/skbuff.h>
  40 #include <linux/netlink.h>
  41 #include <linux/init.h>
  42
  43 #include <net/ip.h>
  44 #include <net/protocol.h>
  45 #include <net/route.h>
  46 #include <net/tcp.h>
  47 #include <net/sock.h>
  48 #include <net/ip_fib.h>
  49
  50 #define FRprintk(a...)
  51
  52 struct fib_rule
  53 {
  54         struct fib_rule *r_next;
  55         u32             r_preference;
  56         unsigned char   r_table;
  57         unsigned char   r_action;
  58         unsigned char   r_dst_len;
  59         unsigned char   r_src_len;
  60         u32             r_src;
  61         u32             r_srcmask;
  62         u32             r_dst;
  63         u32             r_dstmask;
  64         u32             r_srcmap;
  65         u8              r_flags;
  66         u8              r_tos;
  67 #ifdef CONFIG_IP_ROUTE_FWMARK
  68         u32             r_fwmark;
  69 #endif
  70         int             r_ifindex;
  71 #ifdef CONFIG_NET_CLS_ROUTE
  72         __u32           r_tclassid;
  73 #endif
  74         char            r_ifname[IFNAMSIZ];
  75 };
  76
  77 static struct fib_rule default_rule = { NULL, 0x7FFF, RT_TABLE_DEFAULT, RTN_UNICAST, };
  78 static struct fib_rule main_rule = { &default_rule, 0x7FFE, RT_TABLE_MAIN, RTN_UNICAST, };
  79 static struct fib_rule local_rule = { &main_rule, 0, RT_TABLE_LOCAL, RTN_UNICAST, };
  80
  81 static struct fib_rule *fib_rules = &local_rule;
  82 static rwlock_t fib_rules_lock = RW_LOCK_UNLOCKED;
  83
  84 int inet_rtm_delrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
  85 {
  86         struct rtattr **rta = arg;
  87         struct rtmsg *rtm = NLMSG_DATA(nlh);
  88         struct fib_rule *r, **rp;
  89         int err = -ESRCH;
  90
  91         for (rp=&fib_rules; (r=*rp) != NULL; rp=&r->r_next) {
  92                 if ((!rta[RTA_SRC-1] || memcmp(RTA_DATA(rta[RTA_SRC-1]), &r->r_src, 4) == 0) &&
  93                     rtm->rtm_src_len == r->r_src_len &&
  94                     rtm->rtm_dst_len == r->r_dst_len &&
  95                     (!rta[RTA_DST-1] || memcmp(RTA_DATA(rta[RTA_DST-1]), &r->r_dst, 4) == 0) &&
  96                     rtm->rtm_tos == r->r_tos &&
  97 #ifdef CONFIG_IP_ROUTE_FWMARK
  98                     (!rta[RTA_PROTOINFO-1] || memcmp(RTA_DATA(rta[RTA_PROTOINFO-1]), &r->r_fwmark, 4) == 0) &&
  99 #endif
 100                     (!rtm->rtm_type || rtm->rtm_type == r->r_action) &&
 101                     (!rta[RTA_PRIORITY-1] || memcmp(RTA_DATA(rta[RTA_PRIORITY-1]), &r->r_preference, 4) == 0) &&
 102                     (!rta[RTA_IIF-1] || strcmp(RTA_DATA(rta[RTA_IIF-1]), r->r_ifname) == 0) &&
 103                     (!rtm->rtm_table || (r && rtm->rtm_table == r->r_table))) {
 104                         err = -EPERM;
 105                         if (r == &local_rule)
 106                                 break;
 107
 108                         write_lock_bh(&fib_rules_lock);
 109                         *rp = r->r_next;
 110                         write_unlock_bh(&fib_rules_lock);
 111                         if (r != &default_rule && r != &main_rule)
 112                                 kfree(r);
 113                         err = 0;
 114                         break;
 115                 }
 116         }
 117         return err;
 118 }
 119
 120 /* Allocate new unique table id */
 121
 122 static struct fib_table *fib_empty_table(void)
 123 {
 124         int id;
 125
 126         for (id = 1; id <= RT_TABLE_MAX; id++)
 127                 if (fib_tables[id] == NULL)
 128                         return __fib_new_table(id);
 129         return NULL;
 130 }
 131
 132
 133 int inet_rtm_newrule(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
 134 {
 135         struct rtattr **rta = arg;
 136         struct rtmsg *rtm = NLMSG_DATA(nlh);
 137         struct fib_rule *r, *new_r, **rp;
 138         unsigned char table_id;
 139
 140         if (rtm->rtm_src_len > 32 || rtm->rtm_dst_len > 32 ||
 141             (rtm->rtm_tos & ~IPTOS_TOS_MASK))
 142                 return -EINVAL;
 143
 144         if (rta[RTA_IIF-1] && RTA_PAYLOAD(rta[RTA_IIF-1]) > IFNAMSIZ)
 145                 return -EINVAL;
 146
 147         table_id = rtm->rtm_table;
 148         if (table_id == RT_TABLE_UNSPEC) {
 149                 struct fib_table *table;
 150                 if (rtm->rtm_type == RTN_UNICAST || rtm->rtm_type == RTN_NAT) {
 151                         if ((table = fib_empty_table()) == NULL)
 152                                 return -ENOBUFS;
 153                         table_id = table->tb_id;
 154                 }
 155         }
 156
 157         new_r = kmalloc(sizeof(*new_r), GFP_KERNEL);
 158         if (!new_r)
 159                 return -ENOMEM;
 160         memset(new_r, 0, sizeof(*new_r));
 161         if (rta[RTA_SRC-1])
 162                 memcpy(&new_r->r_src, RTA_DATA(rta[RTA_SRC-1]), 4);
 163         if (rta[RTA_DST-1])
 164                 memcpy(&new_r->r_dst, RTA_DATA(rta[RTA_DST-1]), 4);
 165         if (rta[RTA_GATEWAY-1])
 166                 memcpy(&new_r->r_srcmap, RTA_DATA(rta[RTA_GATEWAY-1]), 4);
 167         new_r->r_src_len = rtm->rtm_src_len;
 168         new_r->r_dst_len = rtm->rtm_dst_len;
 169         new_r->r_srcmask = inet_make_mask(rtm->rtm_src_len);
 170         new_r->r_dstmask = inet_make_mask(rtm->rtm_dst_len);
 171         new_r->r_tos = rtm->rtm_tos;
 172 #ifdef CONFIG_IP_ROUTE_FWMARK
 173         if (rta[RTA_PROTOINFO-1])
 174                 memcpy(&new_r->r_fwmark, RTA_DATA(rta[RTA_PROTOINFO-1]), 4);
 175 #endif
 176         new_r->r_action = rtm->rtm_type;
 177         new_r->r_flags = rtm->rtm_flags;
 178         if (rta[RTA_PRIORITY-1])
 179                 memcpy(&new_r->r_preference, RTA_DATA(rta[RTA_PRIORITY-1]), 4);
 180         new_r->r_table = table_id;
 181         if (rta[RTA_IIF-1]) {
 182                 struct device *dev;
 183                 memcpy(new_r->r_ifname, RTA_DATA(rta[RTA_IIF-1]), IFNAMSIZ);
 184                 new_r->r_ifname[IFNAMSIZ-1] = 0;
 185                 new_r->r_ifindex = -1;
 186                 dev = dev_get(new_r->r_ifname);
 187                 if (dev)
 188                         new_r->r_ifindex = dev->ifindex;
 189         }
 190 #ifdef CONFIG_NET_CLS_ROUTE
 191         if (rta[RTA_FLOW-1])
 192                 memcpy(&new_r->r_tclassid, RTA_DATA(rta[RTA_FLOW-1]), 4);
 193 #endif
 194
 195         rp = &fib_rules;
 196         if (!new_r->r_preference) {
 197                 r = fib_rules;
 198                 if (r && (r = r->r_next) != NULL) {
 199                         rp = &fib_rules->r_next;
 200                         if (r->r_preference)
 201                                 new_r->r_preference = r->r_preference - 1;
 202                 }
 203         }
 204
 205         while ( (r = *rp) != NULL ) {
 206                 if (r->r_preference > new_r->r_preference)
 207                         break;
 208                 rp = &r->r_next;
 209         }
 210
 211         new_r->r_next = r;
 212         write_lock_bh(&fib_rules_lock);
 213         *rp = new_r;
 214         write_unlock_bh(&fib_rules_lock);
 215         return 0;
 216 }
 217
 218 u32 fib_rules_map_destination(u32 daddr, struct fib_result *res)
 219 {
 220         u32 mask = inet_make_mask(res->prefixlen);
 221         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 222 }
 223
 224 u32 fib_rules_policy(u32 saddr, struct fib_result *res, unsigned *flags)
 225 {
 226         struct fib_rule *r = res->r;
 227
 228         if (r->r_action == RTN_NAT) {
 229                 int addrtype = inet_addr_type(r->r_srcmap);
 230
 231                 if (addrtype == RTN_NAT) {
 232                         /* Packet is from  translated source; remember it */
 233                         saddr = (saddr&~r->r_srcmask)|r->r_srcmap;
 234                         *flags |= RTCF_SNAT;
 235                 } else if (addrtype == RTN_LOCAL || r->r_srcmap == 0) {
 236                         /* Packet is from masqueraded source; remember it */
 237                         saddr = r->r_srcmap;
 238                         *flags |= RTCF_MASQ;
 239                 }
 240         }
 241         return saddr;
 242 }
 243
 244 #ifdef CONFIG_NET_CLS_ROUTE
 245 u32 fib_rules_tclass(struct fib_result *res)
 246 {
 247         if (res->r)
 248                 return res->r->r_tclassid;
 249         return 0;
 250 }
 251 #endif
 252
 253
 254 static void fib_rules_detach(struct device *dev)
 255 {
 256         struct fib_rule *r;
 257
 258         for (r=fib_rules; r; r=r->r_next) {
 259                 if (r->r_ifindex == dev->ifindex) {
 260                         write_lock_bh(&fib_rules_lock);
 261                         r->r_ifindex = -1;
 262                         write_unlock_bh(&fib_rules_lock);
 263                 }
 264         }
 265 }
 266
 267 static void fib_rules_attach(struct device *dev)
 268 {
 269         struct fib_rule *r;
 270
 271         for (r=fib_rules; r; r=r->r_next) {
 272                 if (r->r_ifindex == -1 && strcmp(dev->name, r->r_ifname) == 0) {
 273                         write_lock_bh(&fib_rules_lock);
 274                         r->r_ifindex = dev->ifindex;
 275                         write_unlock_bh(&fib_rules_lock);
 276                 }
 277         }
 278 }
 279
 280 int fib_lookup(const struct rt_key *key, struct fib_result *res)
 281 {
 282         int err;
 283         struct fib_rule *r, *policy;
 284         struct fib_table *tb;
 285
 286         u32 daddr = key->dst;
 287         u32 saddr = key->src;
 288
 289 FRprintk("Lookup: %08x <- %08x ", key->dst, key->src);
 290         read_lock(&fib_rules_lock);
 291         for (r = fib_rules; r; r=r->r_next) {
 292                 if (((saddr^r->r_src) & r->r_srcmask) ||
 293                     ((daddr^r->r_dst) & r->r_dstmask) ||
 294 #ifdef CONFIG_IP_ROUTE_TOS
 295                     (r->r_tos && r->r_tos != key->tos) ||
 296 #endif
 297 #ifdef CONFIG_IP_ROUTE_FWMARK
 298                     (r->r_fwmark && r->r_fwmark != key->fwmark) ||
 299 #endif
 300                     (r->r_ifindex && r->r_ifindex != key->iif))
 301                         continue;
 302
 303 FRprintk("tb %d r %d ", r->r_table, r->r_action);
 304                 switch (r->r_action) {
 305                 case RTN_UNICAST:
 306                 case RTN_NAT:
 307                         policy = r;
 308                         break;
 309                 case RTN_UNREACHABLE:
 310                         read_unlock(&fib_rules_lock);
 311                         return -ENETUNREACH;
 312                 default:
 313                 case RTN_BLACKHOLE:
 314                         read_unlock(&fib_rules_lock);
 315                         return -EINVAL;
 316                 case RTN_PROHIBIT:
 317                         read_unlock(&fib_rules_lock);
 318                         return -EACCES;
 319                 }
 320
 321                 if ((tb = fib_get_table(r->r_table)) == NULL)
 322                         continue;
 323                 err = tb->tb_lookup(tb, key, res);
 324                 if (err == 0) {
 325 FRprintk("ok\n");
 326                         res->r = policy;
 327                         read_unlock(&fib_rules_lock);
 328                         return 0;
 329                 }
 330                 if (err < 0 && err != -EAGAIN) {
 331                         read_unlock(&fib_rules_lock);
 332                         return err;
 333                 }
 334         }
 335 FRprintk("FAILURE\n");
 336         read_unlock(&fib_rules_lock);
 337         return -ENETUNREACH;
 338 }
 339
 340 void fib_select_default(const struct rt_key *key, struct fib_result *res)
 341 {
 342         if (res->r && res->r->r_action == RTN_UNICAST &&
 343             FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) {
 344                 struct fib_table *tb;
 345                 if ((tb = fib_get_table(res->r->r_table)) != NULL)
 346                         tb->tb_select_default(tb, key, res);
 347         }
 348 }
 349
 350 static int fib_rules_event(struct notifier_block *this, unsigned long event, void *ptr)
 351 {
 352         struct device *dev = ptr;
 353
 354         if (event == NETDEV_UNREGISTER)
 355                 fib_rules_detach(dev);
 356         else if (event == NETDEV_REGISTER)
 357                 fib_rules_attach(dev);
 358         return NOTIFY_DONE;
 359 }
 360
 361
 362 struct notifier_block fib_rules_notifier = {
 363         fib_rules_event,
 364         NULL,
 365         0
 366 };
 367
 368 #ifdef CONFIG_RTNETLINK
 369
 370 extern __inline__ int inet_fill_rule(struct sk_buff *skb,
 371                                      struct fib_rule *r,
 372                                      struct netlink_callback *cb)
 373 {
 374         struct rtmsg *rtm;
 375         struct nlmsghdr  *nlh;
 376         unsigned char    *b = skb->tail;
 377
 378         nlh = NLMSG_PUT(skb, NETLINK_CREDS(cb->skb)->pid, cb->nlh->nlmsg_seq, RTM_NEWRULE, sizeof(*rtm));
 379         rtm = NLMSG_DATA(nlh);
 380         rtm->rtm_family = AF_INET;
 381         rtm->rtm_dst_len = r->r_dst_len;
 382         rtm->rtm_src_len = r->r_src_len;
 383         rtm->rtm_tos = r->r_tos;
 384 #ifdef CONFIG_IP_ROUTE_FWMARK
 385         if (r->r_fwmark)
 386                 RTA_PUT(skb, RTA_PROTOINFO, 4, &r->r_fwmark);
 387 #endif
 388         rtm->rtm_table = r->r_table;
 389         rtm->rtm_protocol = 0;
 390         rtm->rtm_scope = 0;
 391         rtm->rtm_type = r->r_action;
 392         rtm->rtm_flags = r->r_flags;
 393
 394         if (r->r_dst_len)
 395                 RTA_PUT(skb, RTA_DST, 4, &r->r_dst);
 396         if (r->r_src_len)
 397                 RTA_PUT(skb, RTA_SRC, 4, &r->r_src);
 398         if (r->r_ifname[0])
 399                 RTA_PUT(skb, RTA_IIF, IFNAMSIZ, &r->r_ifname);
 400         if (r->r_preference)
 401                 RTA_PUT(skb, RTA_PRIORITY, 4, &r->r_preference);
 402         if (r->r_srcmap)
 403                 RTA_PUT(skb, RTA_GATEWAY, 4, &r->r_srcmap);
 404 #ifdef CONFIG_NET_CLS_ROUTE
 405         if (r->r_tclassid)
 406                 RTA_PUT(skb, RTA_FLOW, 4, &r->r_tclassid);
 407 #endif
 408         nlh->nlmsg_len = skb->tail - b;
 409         return skb->len;
 410
 411 nlmsg_failure:
 412 rtattr_failure:
 413         skb_put(skb, b - skb->tail);
 414         return -1;
 415 }
 416
 417 int inet_dump_rules(struct sk_buff *skb, struct netlink_callback *cb)
 418 {
 419         int idx;
 420         int s_idx = cb->args[0];
 421         struct fib_rule *r;
 422
 423         read_lock(&fib_rules_lock);
 424         for (r=fib_rules, idx=0; r; r = r->r_next, idx++) {
 425                 if (idx < s_idx)
 426                         continue;
 427                 if (inet_fill_rule(skb, r, cb) < 0)
 428                         break;
 429         }
 430         read_unlock(&fib_rules_lock);
 431         cb->args[0] = idx;
 432
 433         return skb->len;
 434 }
 435
 436 #endif /* CONFIG_RTNETLINK */
 437
 438 __initfunc(void fib_rules_init(void))
 439 {
 440         register_netdevice_notifier(&fib_rules_notifier);
 441 }