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: FIB frontend. 7 * 8 * Version: $Id: fib_frontend.c,v 1.26 2001/10/31 21:55:54 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 18 #include <linux/config.h> 19 #include <linux/module.h> 20 #include <asm/uaccess.h> 21 #include <asm/system.h> 22 #include <linux/bitops.h> 23 #include <linux/types.h> 24 #include <linux/kernel.h> 25 #include <linux/sched.h> 26 #include <linux/mm.h> 27 #include <linux/string.h> 28 #include <linux/socket.h> 29 #include <linux/sockios.h> 30 #include <linux/errno.h> 31 #include <linux/in.h> 32 #include <linux/inet.h> 33 #include <linux/netdevice.h> 34 #include <linux/if_arp.h> 35 #include <linux/skbuff.h> 36 #include <linux/netlink.h> 37 #include <linux/init.h> 38 39 #include <net/ip.h> 40 #include <net/protocol.h> 41 #include <net/route.h> 42 #include <net/tcp.h> 43 #include <net/sock.h> 44 #include <net/icmp.h> 45 #include <net/arp.h> 46 #include <net/ip_fib.h> 47 48 #define FFprint(a...) printk(KERN_DEBUG a) 49 50 #ifndef CONFIG_IP_MULTIPLE_TABLES 51 52 #define RT_TABLE_MIN RT_TABLE_MAIN 53 54 struct fib_table *ip_fib_local_table; 55 struct fib_table *ip_fib_main_table; 56 57 #else 58 59 #define RT_TABLE_MIN 1 60 61 struct fib_table *fib_tables[RT_TABLE_MAX+1]; 62 63 struct fib_table *__fib_new_table(int id) 64 { 65 struct fib_table *tb; 66 67 tb = fib_hash_init(id); 68 if (!tb) 69 return NULL; 70 fib_tables[id] = tb; 71 return tb; 72 } 73 74 75 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 76 77 78 static void fib_flush(void) 79 { 80 int flushed = 0; 81 #ifdef CONFIG_IP_MULTIPLE_TABLES 82 struct fib_table *tb; 83 int id; 84 85 for (id = RT_TABLE_MAX; id>0; id--) { 86 if ((tb = fib_get_table(id))==NULL) 87 continue; 88 flushed += tb->tb_flush(tb); 89 } 90 #else /* CONFIG_IP_MULTIPLE_TABLES */ 91 flushed += ip_fib_main_table->tb_flush(ip_fib_main_table); 92 flushed += ip_fib_local_table->tb_flush(ip_fib_local_table); 93 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 94 95 if (flushed) 96 rt_cache_flush(-1); 97 } 98 99 /* 100 * Find the first device with a given source address. 101 */ 102 103 struct net_device * ip_dev_find(u32 addr) 104 { 105 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = addr } } }; 106 struct fib_result res; 107 struct net_device *dev = NULL; 108 109 #ifdef CONFIG_IP_MULTIPLE_TABLES 110 res.r = NULL; 111 #endif 112 113 if (!ip_fib_local_table || 114 ip_fib_local_table->tb_lookup(ip_fib_local_table, &fl, &res)) 115 return NULL; 116 if (res.type != RTN_LOCAL) 117 goto out; 118 dev = FIB_RES_DEV(res); 119 120 if (dev) 121 dev_hold(dev); 122 out: 123 fib_res_put(&res); 124 return dev; 125 } 126 127 unsigned inet_addr_type(u32 addr) 128 { 129 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = addr } } }; 130 struct fib_result res; 131 unsigned ret = RTN_BROADCAST; 132 133 if (ZERONET(addr) || BADCLASS(addr)) 134 return RTN_BROADCAST; 135 if (MULTICAST(addr)) 136 return RTN_MULTICAST; 137 138 #ifdef CONFIG_IP_MULTIPLE_TABLES 139 res.r = NULL; 140 #endif 141 142 if (ip_fib_local_table) { 143 ret = RTN_UNICAST; 144 if (!ip_fib_local_table->tb_lookup(ip_fib_local_table, 145 &fl, &res)) { 146 ret = res.type; 147 fib_res_put(&res); 148 } 149 } 150 return ret; 151 } 152 153 /* Given (packet source, input interface) and optional (dst, oif, tos): 154 - (main) check, that source is valid i.e. not broadcast or our local 155 address. 156 - figure out what "logical" interface this packet arrived 157 and calculate "specific destination" address. 158 - check, that packet arrived from expected physical interface. 159 */ 160 161 int fib_validate_source(u32 src, u32 dst, u8 tos, int oif, 162 struct net_device *dev, u32 *spec_dst, u32 *itag) 163 { 164 struct in_device *in_dev; 165 struct flowi fl = { .nl_u = { .ip4_u = 166 { .daddr = src, 167 .saddr = dst, 168 .tos = tos } }, 169 .iif = oif }; 170 struct fib_result res; 171 int no_addr, rpf; 172 int ret; 173 174 no_addr = rpf = 0; 175 rcu_read_lock(); 176 in_dev = __in_dev_get(dev); 177 if (in_dev) { 178 no_addr = in_dev->ifa_list == NULL; 179 rpf = IN_DEV_RPFILTER(in_dev); 180 } 181 rcu_read_unlock(); 182 183 if (in_dev == NULL) 184 goto e_inval; 185 186 if (fib_lookup(&fl, &res)) 187 goto last_resort; 188 if (res.type != RTN_UNICAST) 189 goto e_inval_res; 190 *spec_dst = FIB_RES_PREFSRC(res); 191 fib_combine_itag(itag, &res); 192 #ifdef CONFIG_IP_ROUTE_MULTIPATH 193 if (FIB_RES_DEV(res) == dev || res.fi->fib_nhs > 1) 194 #else 195 if (FIB_RES_DEV(res) == dev) 196 #endif 197 { 198 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 199 fib_res_put(&res); 200 return ret; 201 } 202 fib_res_put(&res); 203 if (no_addr) 204 goto last_resort; 205 if (rpf) 206 goto e_inval; 207 fl.oif = dev->ifindex; 208 209 ret = 0; 210 if (fib_lookup(&fl, &res) == 0) { 211 if (res.type == RTN_UNICAST) { 212 *spec_dst = FIB_RES_PREFSRC(res); 213 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 214 } 215 fib_res_put(&res); 216 } 217 return ret; 218 219 last_resort: 220 if (rpf) 221 goto e_inval; 222 *spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); 223 *itag = 0; 224 return 0; 225 226 e_inval_res: 227 fib_res_put(&res); 228 e_inval: 229 return -EINVAL; 230 } 231 232 #ifndef CONFIG_IP_NOSIOCRT 233 234 /* 235 * Handle IP routing ioctl calls. These are used to manipulate the routing tables 236 */ 237 238 int ip_rt_ioctl(unsigned int cmd, void __user *arg) 239 { 240 int err; 241 struct kern_rta rta; 242 struct rtentry r; 243 struct { 244 struct nlmsghdr nlh; 245 struct rtmsg rtm; 246 } req; 247 248 switch (cmd) { 249 case SIOCADDRT: /* Add a route */ 250 case SIOCDELRT: /* Delete a route */ 251 if (!capable(CAP_NET_ADMIN)) 252 return -EPERM; 253 if (copy_from_user(&r, arg, sizeof(struct rtentry))) 254 return -EFAULT; 255 rtnl_lock(); 256 err = fib_convert_rtentry(cmd, &req.nlh, &req.rtm, &rta, &r); 257 if (err == 0) { 258 if (cmd == SIOCDELRT) { 259 struct fib_table *tb = fib_get_table(req.rtm.rtm_table); 260 err = -ESRCH; 261 if (tb) 262 err = tb->tb_delete(tb, &req.rtm, &rta, &req.nlh, NULL); 263 } else { 264 struct fib_table *tb = fib_new_table(req.rtm.rtm_table); 265 err = -ENOBUFS; 266 if (tb) 267 err = tb->tb_insert(tb, &req.rtm, &rta, &req.nlh, NULL); 268 } 269 if (rta.rta_mx) 270 kfree(rta.rta_mx); 271 } 272 rtnl_unlock(); 273 return err; 274 } 275 return -EINVAL; 276 } 277 278 #else 279 280 int ip_rt_ioctl(unsigned int cmd, void *arg) 281 { 282 return -EINVAL; 283 } 284 285 #endif 286 287 static int inet_check_attr(struct rtmsg *r, struct rtattr **rta) 288 { 289 int i; 290 291 for (i=1; i<=RTA_MAX; i++) { 292 struct rtattr *attr = rta[i-1]; 293 if (attr) { 294 if (RTA_PAYLOAD(attr) < 4) 295 return -EINVAL; 296 if (i != RTA_MULTIPATH && i != RTA_METRICS) 297 rta[i-1] = (struct rtattr*)RTA_DATA(attr); 298 } 299 } 300 return 0; 301 } 302 303 int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg) 304 { 305 struct fib_table * tb; 306 struct rtattr **rta = arg; 307 struct rtmsg *r = NLMSG_DATA(nlh); 308 309 if (inet_check_attr(r, rta)) 310 return -EINVAL; 311 312 tb = fib_get_table(r->rtm_table); 313 if (tb) 314 return tb->tb_delete(tb, r, (struct kern_rta*)rta, nlh, &NETLINK_CB(skb)); 315 return -ESRCH; 316 } 317 318 int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg) 319 { 320 struct fib_table * tb; 321 struct rtattr **rta = arg; 322 struct rtmsg *r = NLMSG_DATA(nlh); 323 324 if (inet_check_attr(r, rta)) 325 return -EINVAL; 326 327 tb = fib_new_table(r->rtm_table); 328 if (tb) 329 return tb->tb_insert(tb, r, (struct kern_rta*)rta, nlh, &NETLINK_CB(skb)); 330 return -ENOBUFS; 331 } 332 333 int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 334 { 335 int t; 336 int s_t; 337 struct fib_table *tb; 338 339 if (NLMSG_PAYLOAD(cb->nlh, 0) >= sizeof(struct rtmsg) && 340 ((struct rtmsg*)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED) 341 return ip_rt_dump(skb, cb); 342 343 s_t = cb->args[0]; 344 if (s_t == 0) 345 s_t = cb->args[0] = RT_TABLE_MIN; 346 347 for (t=s_t; t<=RT_TABLE_MAX; t++) { 348 if (t < s_t) continue; 349 if (t > s_t) 350 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 351 if ((tb = fib_get_table(t))==NULL) 352 continue; 353 if (tb->tb_dump(tb, skb, cb) < 0) 354 break; 355 } 356 357 cb->args[0] = t; 358 359 return skb->len; 360 } 361 362 /* Prepare and feed intra-kernel routing request. 363 Really, it should be netlink message, but :-( netlink 364 can be not configured, so that we feed it directly 365 to fib engine. It is legal, because all events occur 366 only when netlink is already locked. 367 */ 368 369 static void fib_magic(int cmd, int type, u32 dst, int dst_len, struct in_ifaddr *ifa) 370 { 371 struct fib_table * tb; 372 struct { 373 struct nlmsghdr nlh; 374 struct rtmsg rtm; 375 } req; 376 struct kern_rta rta; 377 378 memset(&req.rtm, 0, sizeof(req.rtm)); 379 memset(&rta, 0, sizeof(rta)); 380 381 if (type == RTN_UNICAST) 382 tb = fib_new_table(RT_TABLE_MAIN); 383 else 384 tb = fib_new_table(RT_TABLE_LOCAL); 385 386 if (tb == NULL) 387 return; 388 389 req.nlh.nlmsg_len = sizeof(req); 390 req.nlh.nlmsg_type = cmd; 391 req.nlh.nlmsg_flags = NLM_F_REQUEST|NLM_F_CREATE|NLM_F_APPEND; 392 req.nlh.nlmsg_pid = 0; 393 req.nlh.nlmsg_seq = 0; 394 395 req.rtm.rtm_dst_len = dst_len; 396 req.rtm.rtm_table = tb->tb_id; 397 req.rtm.rtm_protocol = RTPROT_KERNEL; 398 req.rtm.rtm_scope = (type != RTN_LOCAL ? RT_SCOPE_LINK : RT_SCOPE_HOST); 399 req.rtm.rtm_type = type; 400 401 rta.rta_dst = &dst; 402 rta.rta_prefsrc = &ifa->ifa_local; 403 rta.rta_oif = &ifa->ifa_dev->dev->ifindex; 404 405 if (cmd == RTM_NEWROUTE) 406 tb->tb_insert(tb, &req.rtm, &rta, &req.nlh, NULL); 407 else 408 tb->tb_delete(tb, &req.rtm, &rta, &req.nlh, NULL); 409 } 410 411 static void fib_add_ifaddr(struct in_ifaddr *ifa) 412 { 413 struct in_device *in_dev = ifa->ifa_dev; 414 struct net_device *dev = in_dev->dev; 415 struct in_ifaddr *prim = ifa; 416 u32 mask = ifa->ifa_mask; 417 u32 addr = ifa->ifa_local; 418 u32 prefix = ifa->ifa_address&mask; 419 420 if (ifa->ifa_flags&IFA_F_SECONDARY) { 421 prim = inet_ifa_byprefix(in_dev, prefix, mask); 422 if (prim == NULL) { 423 printk(KERN_DEBUG "fib_add_ifaddr: bug: prim == NULL\n"); 424 return; 425 } 426 } 427 428 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 429 430 if (!(dev->flags&IFF_UP)) 431 return; 432 433 /* Add broadcast address, if it is explicitly assigned. */ 434 if (ifa->ifa_broadcast && ifa->ifa_broadcast != 0xFFFFFFFF) 435 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 436 437 if (!ZERONET(prefix) && !(ifa->ifa_flags&IFA_F_SECONDARY) && 438 (prefix != addr || ifa->ifa_prefixlen < 32)) { 439 fib_magic(RTM_NEWROUTE, dev->flags&IFF_LOOPBACK ? RTN_LOCAL : 440 RTN_UNICAST, prefix, ifa->ifa_prefixlen, prim); 441 442 /* Add network specific broadcasts, when it takes a sense */ 443 if (ifa->ifa_prefixlen < 31) { 444 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 445 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix|~mask, 32, prim); 446 } 447 } 448 } 449 450 static void fib_del_ifaddr(struct in_ifaddr *ifa) 451 { 452 struct in_device *in_dev = ifa->ifa_dev; 453 struct net_device *dev = in_dev->dev; 454 struct in_ifaddr *ifa1; 455 struct in_ifaddr *prim = ifa; 456 u32 brd = ifa->ifa_address|~ifa->ifa_mask; 457 u32 any = ifa->ifa_address&ifa->ifa_mask; 458 #define LOCAL_OK 1 459 #define BRD_OK 2 460 #define BRD0_OK 4 461 #define BRD1_OK 8 462 unsigned ok = 0; 463 464 if (!(ifa->ifa_flags&IFA_F_SECONDARY)) 465 fib_magic(RTM_DELROUTE, dev->flags&IFF_LOOPBACK ? RTN_LOCAL : 466 RTN_UNICAST, any, ifa->ifa_prefixlen, prim); 467 else { 468 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 469 if (prim == NULL) { 470 printk(KERN_DEBUG "fib_del_ifaddr: bug: prim == NULL\n"); 471 return; 472 } 473 } 474 475 /* Deletion is more complicated than add. 476 We should take care of not to delete too much :-) 477 478 Scan address list to be sure that addresses are really gone. 479 */ 480 481 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 482 if (ifa->ifa_local == ifa1->ifa_local) 483 ok |= LOCAL_OK; 484 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 485 ok |= BRD_OK; 486 if (brd == ifa1->ifa_broadcast) 487 ok |= BRD1_OK; 488 if (any == ifa1->ifa_broadcast) 489 ok |= BRD0_OK; 490 } 491 492 if (!(ok&BRD_OK)) 493 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 494 if (!(ok&BRD1_OK)) 495 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 496 if (!(ok&BRD0_OK)) 497 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 498 if (!(ok&LOCAL_OK)) { 499 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 500 501 /* Check, that this local address finally disappeared. */ 502 if (inet_addr_type(ifa->ifa_local) != RTN_LOCAL) { 503 /* And the last, but not the least thing. 504 We must flush stray FIB entries. 505 506 First of all, we scan fib_info list searching 507 for stray nexthop entries, then ignite fib_flush. 508 */ 509 if (fib_sync_down(ifa->ifa_local, NULL, 0)) 510 fib_flush(); 511 } 512 } 513 #undef LOCAL_OK 514 #undef BRD_OK 515 #undef BRD0_OK 516 #undef BRD1_OK 517 } 518 519 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb ) 520 { 521 522 struct fib_result res; 523 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = frn->fl_addr, 524 .fwmark = frn->fl_fwmark, 525 .tos = frn->fl_tos, 526 .scope = frn->fl_scope } } }; 527 if (tb) { 528 local_bh_disable(); 529 530 frn->tb_id = tb->tb_id; 531 frn->err = tb->tb_lookup(tb, &fl, &res); 532 533 if (!frn->err) { 534 frn->prefixlen = res.prefixlen; 535 frn->nh_sel = res.nh_sel; 536 frn->type = res.type; 537 frn->scope = res.scope; 538 } 539 local_bh_enable(); 540 } 541 } 542 543 static void nl_fib_input(struct sock *sk, int len) 544 { 545 struct sk_buff *skb = NULL; 546 struct nlmsghdr *nlh = NULL; 547 struct fib_result_nl *frn; 548 int err; 549 u32 pid; 550 struct fib_table *tb; 551 552 skb = skb_recv_datagram(sk, 0, 0, &err); 553 nlh = (struct nlmsghdr *)skb->data; 554 555 frn = (struct fib_result_nl *) NLMSG_DATA(nlh); 556 tb = fib_get_table(frn->tb_id_in); 557 558 nl_fib_lookup(frn, tb); 559 560 pid = nlh->nlmsg_pid; /*pid of sending process */ 561 NETLINK_CB(skb).pid = 0; /* from kernel */ 562 NETLINK_CB(skb).dst_pid = pid; 563 NETLINK_CB(skb).dst_group = 0; /* unicast */ 564 netlink_unicast(sk, skb, pid, MSG_DONTWAIT); 565 } 566 567 static void nl_fib_lookup_init(void) 568 { 569 netlink_kernel_create(NETLINK_FIB_LOOKUP, 0, nl_fib_input, THIS_MODULE); 570 } 571 572 static void fib_disable_ip(struct net_device *dev, int force) 573 { 574 if (fib_sync_down(0, dev, force)) 575 fib_flush(); 576 rt_cache_flush(0); 577 arp_ifdown(dev); 578 } 579 580 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 581 { 582 struct in_ifaddr *ifa = (struct in_ifaddr*)ptr; 583 584 switch (event) { 585 case NETDEV_UP: 586 fib_add_ifaddr(ifa); 587 #ifdef CONFIG_IP_ROUTE_MULTIPATH 588 fib_sync_up(ifa->ifa_dev->dev); 589 #endif 590 rt_cache_flush(-1); 591 break; 592 case NETDEV_DOWN: 593 fib_del_ifaddr(ifa); 594 if (ifa->ifa_dev && ifa->ifa_dev->ifa_list == NULL) { 595 /* Last address was deleted from this interface. 596 Disable IP. 597 */ 598 fib_disable_ip(ifa->ifa_dev->dev, 1); 599 } else { 600 rt_cache_flush(-1); 601 } 602 break; 603 } 604 return NOTIFY_DONE; 605 } 606 607 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 608 { 609 struct net_device *dev = ptr; 610 struct in_device *in_dev = __in_dev_get(dev); 611 612 if (event == NETDEV_UNREGISTER) { 613 fib_disable_ip(dev, 2); 614 return NOTIFY_DONE; 615 } 616 617 if (!in_dev) 618 return NOTIFY_DONE; 619 620 switch (event) { 621 case NETDEV_UP: 622 for_ifa(in_dev) { 623 fib_add_ifaddr(ifa); 624 } endfor_ifa(in_dev); 625 #ifdef CONFIG_IP_ROUTE_MULTIPATH 626 fib_sync_up(dev); 627 #endif 628 rt_cache_flush(-1); 629 break; 630 case NETDEV_DOWN: 631 fib_disable_ip(dev, 0); 632 break; 633 case NETDEV_CHANGEMTU: 634 case NETDEV_CHANGE: 635 rt_cache_flush(0); 636 break; 637 } 638 return NOTIFY_DONE; 639 } 640 641 static struct notifier_block fib_inetaddr_notifier = { 642 .notifier_call =fib_inetaddr_event, 643 }; 644 645 static struct notifier_block fib_netdev_notifier = { 646 .notifier_call =fib_netdev_event, 647 }; 648 649 void __init ip_fib_init(void) 650 { 651 #ifndef CONFIG_IP_MULTIPLE_TABLES 652 ip_fib_local_table = fib_hash_init(RT_TABLE_LOCAL); 653 ip_fib_main_table = fib_hash_init(RT_TABLE_MAIN); 654 #else 655 fib_rules_init(); 656 #endif 657 658 register_netdevice_notifier(&fib_netdev_notifier); 659 register_inetaddr_notifier(&fib_inetaddr_notifier); 660 nl_fib_lookup_init(); 661 } 662 663 EXPORT_SYMBOL(inet_addr_type); 664 EXPORT_SYMBOL(ip_rt_ioctl); 665