net/ipv6/route.c

   1 /*
   2  *      Linux INET6 implementation
   3  *      FIB front-end.
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      $Id: route.c,v 1.56 2001/10/31 21:55:55 davem Exp $
   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 /*      Changes:
  17  *
  18  *      YOSHIFUJI Hideaki @USAGI
  19  *              reworked default router selection.
  20  *              - respect outgoing interface
  21  *              - select from (probably) reachable routers (i.e.
  22  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  23  *              - always select the same router if it is (probably)
  24  *              reachable.  otherwise, round-robin the list.
  25  */
  26
  27 #include <linux/capability.h>
  28 #include <linux/errno.h>
  29 #include <linux/types.h>
  30 #include <linux/times.h>
  31 #include <linux/socket.h>
  32 #include <linux/sockios.h>
  33 #include <linux/net.h>
  34 #include <linux/route.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/in6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39
  40 #ifdef  CONFIG_PROC_FS
  41 #include <linux/proc_fs.h>
  42 #include <linux/seq_file.h>
  43 #endif
  44
  45 #include <net/snmp.h>
  46 #include <net/ipv6.h>
  47 #include <net/ip6_fib.h>
  48 #include <net/ip6_route.h>
  49 #include <net/ndisc.h>
  50 #include <net/addrconf.h>
  51 #include <net/tcp.h>
  52 #include <linux/rtnetlink.h>
  53 #include <net/dst.h>
  54 #include <net/xfrm.h>
  55 #include <net/netevent.h>
  56 #include <net/netlink.h>
  57
  58 #include <asm/uaccess.h>
  59
  60 #ifdef CONFIG_SYSCTL
  61 #include <linux/sysctl.h>
  62 #endif
  63
  64 /* Set to 3 to get tracing. */
  65 #define RT6_DEBUG 2
  66
  67 #if RT6_DEBUG >= 3
  68 #define RDBG(x) printk x
  69 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
  70 #else
  71 #define RDBG(x)
  72 #define RT6_TRACE(x...) do { ; } while (0)
  73 #endif
  74
  75 #define CLONE_OFFLINK_ROUTE 0
  76
  77 #define RT6_SELECT_F_IFACE      0x1
  78 #define RT6_SELECT_F_REACHABLE  0x2
  79
  80 static int ip6_rt_max_size = 4096;
  81 static int ip6_rt_gc_min_interval = HZ / 2;
  82 static int ip6_rt_gc_timeout = 60*HZ;
  83 int ip6_rt_gc_interval = 30*HZ;
  84 static int ip6_rt_gc_elasticity = 9;
  85 static int ip6_rt_mtu_expires = 10*60*HZ;
  86 static int ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
  87
  88 static struct rt6_info * ip6_rt_copy(struct rt6_info *ort);
  89 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  90 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  91 static void             ip6_dst_destroy(struct dst_entry *);
  92 static void             ip6_dst_ifdown(struct dst_entry *,
  93                                        struct net_device *dev, int how);
  94 static int               ip6_dst_gc(void);
  95
  96 static int              ip6_pkt_discard(struct sk_buff *skb);
  97 static int              ip6_pkt_discard_out(struct sk_buff *skb);
  98 static void             ip6_link_failure(struct sk_buff *skb);
  99 static void             ip6_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
 100
 101 #ifdef CONFIG_IPV6_ROUTE_INFO
 102 static struct rt6_info *rt6_add_route_info(struct in6_addr *prefix, int prefixlen,
 103                                            struct in6_addr *gwaddr, int ifindex,
 104                                            unsigned pref);
 105 static struct rt6_info *rt6_get_route_info(struct in6_addr *prefix, int prefixlen,
 106                                            struct in6_addr *gwaddr, int ifindex);
 107 #endif
 108
 109 static struct dst_ops ip6_dst_ops = {
 110         .family                 =       AF_INET6,
 111         .protocol               =       __constant_htons(ETH_P_IPV6),
 112         .gc                     =       ip6_dst_gc,
 113         .gc_thresh              =       1024,
 114         .check                  =       ip6_dst_check,
 115         .destroy                =       ip6_dst_destroy,
 116         .ifdown                 =       ip6_dst_ifdown,
 117         .negative_advice        =       ip6_negative_advice,
 118         .link_failure           =       ip6_link_failure,
 119         .update_pmtu            =       ip6_rt_update_pmtu,
 120         .entry_size             =       sizeof(struct rt6_info),
 121 };
 122
 123 struct rt6_info ip6_null_entry = {
 124         .u = {
 125                 .dst = {
 126                         .__refcnt       = ATOMIC_INIT(1),
 127                         .__use          = 1,
 128                         .dev            = &loopback_dev,
 129                         .obsolete       = -1,
 130                         .error          = -ENETUNREACH,
 131                         .metrics        = { [RTAX_HOPLIMIT - 1] = 255, },
 132                         .input          = ip6_pkt_discard,
 133                         .output         = ip6_pkt_discard_out,
 134                         .ops            = &ip6_dst_ops,
 135                         .path           = (struct dst_entry*)&ip6_null_entry,
 136                 }
 137         },
 138         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 139         .rt6i_metric    = ~(u32) 0,
 140         .rt6i_ref       = ATOMIC_INIT(1),
 141 };
 142
 143 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 144
 145 struct rt6_info ip6_prohibit_entry = {
 146         .u = {
 147                 .dst = {
 148                         .__refcnt       = ATOMIC_INIT(1),
 149                         .__use          = 1,
 150                         .dev            = &loopback_dev,
 151                         .obsolete       = -1,
 152                         .error          = -EACCES,
 153                         .metrics        = { [RTAX_HOPLIMIT - 1] = 255, },
 154                         .input          = ip6_pkt_discard,
 155                         .output         = ip6_pkt_discard_out,
 156                         .ops            = &ip6_dst_ops,
 157                         .path           = (struct dst_entry*)&ip6_prohibit_entry,
 158                 }
 159         },
 160         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 161         .rt6i_metric    = ~(u32) 0,
 162         .rt6i_ref       = ATOMIC_INIT(1),
 163 };
 164
 165 struct rt6_info ip6_blk_hole_entry = {
 166         .u = {
 167                 .dst = {
 168                         .__refcnt       = ATOMIC_INIT(1),
 169                         .__use          = 1,
 170                         .dev            = &loopback_dev,
 171                         .obsolete       = -1,
 172                         .error          = -EINVAL,
 173                         .metrics        = { [RTAX_HOPLIMIT - 1] = 255, },
 174                         .input          = ip6_pkt_discard,
 175                         .output         = ip6_pkt_discard_out,
 176                         .ops            = &ip6_dst_ops,
 177                         .path           = (struct dst_entry*)&ip6_blk_hole_entry,
 178                 }
 179         },
 180         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 181         .rt6i_metric    = ~(u32) 0,
 182         .rt6i_ref       = ATOMIC_INIT(1),
 183 };
 184
 185 #endif
 186
 187 /* allocate dst with ip6_dst_ops */
 188 static __inline__ struct rt6_info *ip6_dst_alloc(void)
 189 {
 190         return (struct rt6_info *)dst_alloc(&ip6_dst_ops);
 191 }
 192
 193 static void ip6_dst_destroy(struct dst_entry *dst)
 194 {
 195         struct rt6_info *rt = (struct rt6_info *)dst;
 196         struct inet6_dev *idev = rt->rt6i_idev;
 197
 198         if (idev != NULL) {
 199                 rt->rt6i_idev = NULL;
 200                 in6_dev_put(idev);
 201         }
 202 }
 203
 204 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 205                            int how)
 206 {
 207         struct rt6_info *rt = (struct rt6_info *)dst;
 208         struct inet6_dev *idev = rt->rt6i_idev;
 209
 210         if (dev != &loopback_dev && idev != NULL && idev->dev == dev) {
 211                 struct inet6_dev *loopback_idev = in6_dev_get(&loopback_dev);
 212                 if (loopback_idev != NULL) {
 213                         rt->rt6i_idev = loopback_idev;
 214                         in6_dev_put(idev);
 215                 }
 216         }
 217 }
 218
 219 static __inline__ int rt6_check_expired(const struct rt6_info *rt)
 220 {
 221         return (rt->rt6i_flags & RTF_EXPIRES &&
 222                 time_after(jiffies, rt->rt6i_expires));
 223 }
 224
 225 static inline int rt6_need_strict(struct in6_addr *daddr)
 226 {
 227         return (ipv6_addr_type(daddr) &
 228                 (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL));
 229 }
 230
 231 /*
 232  *      Route lookup. Any table->tb6_lock is implied.
 233  */
 234
 235 static __inline__ struct rt6_info *rt6_device_match(struct rt6_info *rt,
 236                                                     int oif,
 237                                                     int strict)
 238 {
 239         struct rt6_info *local = NULL;
 240         struct rt6_info *sprt;
 241
 242         if (oif) {
 243                 for (sprt = rt; sprt; sprt = sprt->u.next) {
 244                         struct net_device *dev = sprt->rt6i_dev;
 245                         if (dev->ifindex == oif)
 246                                 return sprt;
 247                         if (dev->flags & IFF_LOOPBACK) {
 248                                 if (sprt->rt6i_idev == NULL ||
 249                                     sprt->rt6i_idev->dev->ifindex != oif) {
 250                                         if (strict && oif)
 251                                                 continue;
 252                                         if (local && (!oif ||
 253                                                       local->rt6i_idev->dev->ifindex == oif))
 254                                                 continue;
 255                                 }
 256                                 local = sprt;
 257                         }
 258                 }
 259
 260                 if (local)
 261                         return local;
 262
 263                 if (strict)
 264                         return &ip6_null_entry;
 265         }
 266         return rt;
 267 }
 268
 269 #ifdef CONFIG_IPV6_ROUTER_PREF
 270 static void rt6_probe(struct rt6_info *rt)
 271 {
 272         struct neighbour *neigh = rt ? rt->rt6i_nexthop : NULL;
 273         /*
 274          * Okay, this does not seem to be appropriate
 275          * for now, however, we need to check if it
 276          * is really so; aka Router Reachability Probing.
 277          *
 278          * Router Reachability Probe MUST be rate-limited
 279          * to no more than one per minute.
 280          */
 281         if (!neigh || (neigh->nud_state & NUD_VALID))
 282                 return;
 283         read_lock_bh(&neigh->lock);
 284         if (!(neigh->nud_state & NUD_VALID) &&
 285             time_after(jiffies, neigh->updated + rt->rt6i_idev->cnf.rtr_probe_interval)) {
 286                 struct in6_addr mcaddr;
 287                 struct in6_addr *target;
 288
 289                 neigh->updated = jiffies;
 290                 read_unlock_bh(&neigh->lock);
 291
 292                 target = (struct in6_addr *)&neigh->primary_key;
 293                 addrconf_addr_solict_mult(target, &mcaddr);
 294                 ndisc_send_ns(rt->rt6i_dev, NULL, target, &mcaddr, NULL);
 295         } else
 296                 read_unlock_bh(&neigh->lock);
 297 }
 298 #else
 299 static inline void rt6_probe(struct rt6_info *rt)
 300 {
 301         return;
 302 }
 303 #endif
 304
 305 /*
 306  * Default Router Selection (RFC 2461 6.3.6)
 307  */
 308 static int inline rt6_check_dev(struct rt6_info *rt, int oif)
 309 {
 310         struct net_device *dev = rt->rt6i_dev;
 311         if (!oif || dev->ifindex == oif)
 312                 return 2;
 313         if ((dev->flags & IFF_LOOPBACK) &&
 314             rt->rt6i_idev && rt->rt6i_idev->dev->ifindex == oif)
 315                 return 1;
 316         return 0;
 317 }
 318
 319 static int inline rt6_check_neigh(struct rt6_info *rt)
 320 {
 321         struct neighbour *neigh = rt->rt6i_nexthop;
 322         int m = 0;
 323         if (rt->rt6i_flags & RTF_NONEXTHOP ||
 324             !(rt->rt6i_flags & RTF_GATEWAY))
 325                 m = 1;
 326         else if (neigh) {
 327                 read_lock_bh(&neigh->lock);
 328                 if (neigh->nud_state & NUD_VALID)
 329                         m = 2;
 330                 read_unlock_bh(&neigh->lock);
 331         }
 332         return m;
 333 }
 334
 335 static int rt6_score_route(struct rt6_info *rt, int oif,
 336                            int strict)
 337 {
 338         int m, n;
 339
 340         m = rt6_check_dev(rt, oif);
 341         if (!m && (strict & RT6_SELECT_F_IFACE))
 342                 return -1;
 343 #ifdef CONFIG_IPV6_ROUTER_PREF
 344         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(rt->rt6i_flags)) << 2;
 345 #endif
 346         n = rt6_check_neigh(rt);
 347         if (n > 1)
 348                 m |= 16;
 349         else if (!n && strict & RT6_SELECT_F_REACHABLE)
 350                 return -1;
 351         return m;
 352 }
 353
 354 static struct rt6_info *rt6_select(struct rt6_info **head, int oif,
 355                                    int strict)
 356 {
 357         struct rt6_info *match = NULL, *last = NULL;
 358         struct rt6_info *rt, *rt0 = *head;
 359         u32 metric;
 360         int mpri = -1;
 361
 362         RT6_TRACE("%s(head=%p(*head=%p), oif=%d)\n",
 363                   __FUNCTION__, head, head ? *head : NULL, oif);
 364
 365         for (rt = rt0, metric = rt0->rt6i_metric;
 366              rt && rt->rt6i_metric == metric && (!last || rt != rt0);
 367              rt = rt->u.next) {
 368                 int m;
 369
 370                 if (rt6_check_expired(rt))
 371                         continue;
 372
 373                 last = rt;
 374
 375                 m = rt6_score_route(rt, oif, strict);
 376                 if (m < 0)
 377                         continue;
 378
 379                 if (m > mpri) {
 380                         rt6_probe(match);
 381                         match = rt;
 382                         mpri = m;
 383                 } else {
 384                         rt6_probe(rt);
 385                 }
 386         }
 387
 388         if (!match &&
 389             (strict & RT6_SELECT_F_REACHABLE) &&
 390             last && last != rt0) {
 391                 /* no entries matched; do round-robin */
 392                 static DEFINE_SPINLOCK(lock);
 393                 spin_lock(&lock);
 394                 *head = rt0->u.next;
 395                 rt0->u.next = last->u.next;
 396                 last->u.next = rt0;
 397                 spin_unlock(&lock);
 398         }
 399
 400         RT6_TRACE("%s() => %p, score=%d\n",
 401                   __FUNCTION__, match, mpri);
 402
 403         return (match ? match : &ip6_null_entry);
 404 }
 405
 406 #ifdef CONFIG_IPV6_ROUTE_INFO
 407 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 408                   struct in6_addr *gwaddr)
 409 {
 410         struct route_info *rinfo = (struct route_info *) opt;
 411         struct in6_addr prefix_buf, *prefix;
 412         unsigned int pref;
 413         u32 lifetime;
 414         struct rt6_info *rt;
 415
 416         if (len < sizeof(struct route_info)) {
 417                 return -EINVAL;
 418         }
 419
 420         /* Sanity check for prefix_len and length */
 421         if (rinfo->length > 3) {
 422                 return -EINVAL;
 423         } else if (rinfo->prefix_len > 128) {
 424                 return -EINVAL;
 425         } else if (rinfo->prefix_len > 64) {
 426                 if (rinfo->length < 2) {
 427                         return -EINVAL;
 428                 }
 429         } else if (rinfo->prefix_len > 0) {
 430                 if (rinfo->length < 1) {
 431                         return -EINVAL;
 432                 }
 433         }
 434
 435         pref = rinfo->route_pref;
 436         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 437                 pref = ICMPV6_ROUTER_PREF_MEDIUM;
 438
 439         lifetime = htonl(rinfo->lifetime);
 440         if (lifetime == 0xffffffff) {
 441                 /* infinity */
 442         } else if (lifetime > 0x7fffffff/HZ) {
 443                 /* Avoid arithmetic overflow */
 444                 lifetime = 0x7fffffff/HZ - 1;
 445         }
 446
 447         if (rinfo->length == 3)
 448                 prefix = (struct in6_addr *)rinfo->prefix;
 449         else {
 450                 /* this function is safe */
 451                 ipv6_addr_prefix(&prefix_buf,
 452                                  (struct in6_addr *)rinfo->prefix,
 453                                  rinfo->prefix_len);
 454                 prefix = &prefix_buf;
 455         }
 456
 457         rt = rt6_get_route_info(prefix, rinfo->prefix_len, gwaddr, dev->ifindex);
 458
 459         if (rt && !lifetime) {
 460                 ip6_del_rt(rt);
 461                 rt = NULL;
 462         }
 463
 464         if (!rt && lifetime)
 465                 rt = rt6_add_route_info(prefix, rinfo->prefix_len, gwaddr, dev->ifindex,
 466                                         pref);
 467         else if (rt)
 468                 rt->rt6i_flags = RTF_ROUTEINFO |
 469                                  (rt->rt6i_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 470
 471         if (rt) {
 472                 if (lifetime == 0xffffffff) {
 473                         rt->rt6i_flags &= ~RTF_EXPIRES;
 474                 } else {
 475                         rt->rt6i_expires = jiffies + HZ * lifetime;
 476                         rt->rt6i_flags |= RTF_EXPIRES;
 477                 }
 478                 dst_release(&rt->u.dst);
 479         }
 480         return 0;
 481 }
 482 #endif
 483
 484 #define BACKTRACK() \
 485 if (rt == &ip6_null_entry && flags & RT6_F_STRICT) { \
 486         while ((fn = fn->parent) != NULL) { \
 487                 if (fn->fn_flags & RTN_TL_ROOT) { \
 488                         dst_hold(&rt->u.dst); \
 489                         goto out; \
 490                 } \
 491                 if (fn->fn_flags & RTN_RTINFO) \
 492                         goto restart; \
 493         } \
 494 }
 495
 496 static struct rt6_info *ip6_pol_route_lookup(struct fib6_table *table,
 497                                              struct flowi *fl, int flags)
 498 {
 499         struct fib6_node *fn;
 500         struct rt6_info *rt;
 501
 502         read_lock_bh(&table->tb6_lock);
 503         fn = fib6_lookup(&table->tb6_root, &fl->fl6_dst, &fl->fl6_src);
 504 restart:
 505         rt = fn->leaf;
 506         rt = rt6_device_match(rt, fl->oif, flags & RT6_F_STRICT);
 507         BACKTRACK();
 508         dst_hold(&rt->u.dst);
 509 out:
 510         read_unlock_bh(&table->tb6_lock);
 511
 512         rt->u.dst.lastuse = jiffies;
 513         rt->u.dst.__use++;
 514
 515         return rt;
 516
 517 }
 518
 519 struct rt6_info *rt6_lookup(struct in6_addr *daddr, struct in6_addr *saddr,
 520                             int oif, int strict)
 521 {
 522         struct flowi fl = {
 523                 .oif = oif,
 524                 .nl_u = {
 525                         .ip6_u = {
 526                                 .daddr = *daddr,
 527                                 /* TODO: saddr */
 528                         },
 529                 },
 530         };
 531         struct dst_entry *dst;
 532         int flags = strict ? RT6_F_STRICT : 0;
 533
 534         dst = fib6_rule_lookup(&fl, flags, ip6_pol_route_lookup);
 535         if (dst->error == 0)
 536                 return (struct rt6_info *) dst;
 537
 538         dst_release(dst);
 539
 540         return NULL;
 541 }
 542
 543 /* ip6_ins_rt is called with FREE table->tb6_lock.
 544    It takes new route entry, the addition fails by any reason the
 545    route is freed. In any case, if caller does not hold it, it may
 546    be destroyed.
 547  */
 548
 549 int ip6_ins_rt(struct rt6_info *rt, struct nlmsghdr *nlh,
 550                 void *_rtattr, struct netlink_skb_parms *req)
 551 {
 552         int err;
 553         struct fib6_table *table;
 554
 555         table = rt->rt6i_table;
 556         write_lock_bh(&table->tb6_lock);
 557         err = fib6_add(&table->tb6_root, rt, nlh, _rtattr, req);
 558         write_unlock_bh(&table->tb6_lock);
 559
 560         return err;
 561 }
 562
 563 static struct rt6_info *rt6_alloc_cow(struct rt6_info *ort, struct in6_addr *daddr,
 564                                       struct in6_addr *saddr)
 565 {
 566         struct rt6_info *rt;
 567
 568         /*
 569          *      Clone the route.
 570          */
 571
 572         rt = ip6_rt_copy(ort);
 573
 574         if (rt) {
 575                 if (!(rt->rt6i_flags&RTF_GATEWAY)) {
 576                         if (rt->rt6i_dst.plen != 128 &&
 577                             ipv6_addr_equal(&rt->rt6i_dst.addr, daddr))
 578                                 rt->rt6i_flags |= RTF_ANYCAST;
 579                         ipv6_addr_copy(&rt->rt6i_gateway, daddr);
 580                 }
 581
 582                 ipv6_addr_copy(&rt->rt6i_dst.addr, daddr);
 583                 rt->rt6i_dst.plen = 128;
 584                 rt->rt6i_flags |= RTF_CACHE;
 585                 rt->u.dst.flags |= DST_HOST;
 586
 587 #ifdef CONFIG_IPV6_SUBTREES
 588                 if (rt->rt6i_src.plen && saddr) {
 589                         ipv6_addr_copy(&rt->rt6i_src.addr, saddr);
 590                         rt->rt6i_src.plen = 128;
 591                 }
 592 #endif
 593
 594                 rt->rt6i_nexthop = ndisc_get_neigh(rt->rt6i_dev, &rt->rt6i_gateway);
 595
 596         }
 597
 598         return rt;
 599 }
 600
 601 static struct rt6_info *rt6_alloc_clone(struct rt6_info *ort, struct in6_addr *daddr)
 602 {
 603         struct rt6_info *rt = ip6_rt_copy(ort);
 604         if (rt) {
 605                 ipv6_addr_copy(&rt->rt6i_dst.addr, daddr);
 606                 rt->rt6i_dst.plen = 128;
 607                 rt->rt6i_flags |= RTF_CACHE;
 608                 if (rt->rt6i_flags & RTF_REJECT)
 609                         rt->u.dst.error = ort->u.dst.error;
 610                 rt->u.dst.flags |= DST_HOST;
 611                 rt->rt6i_nexthop = neigh_clone(ort->rt6i_nexthop);
 612         }
 613         return rt;
 614 }
 615
 616 static struct rt6_info *ip6_pol_route_input(struct fib6_table *table,
 617                                             struct flowi *fl, int flags)
 618 {
 619         struct fib6_node *fn;
 620         struct rt6_info *rt, *nrt;
 621         int strict = 0;
 622         int attempts = 3;
 623         int err;
 624         int reachable = RT6_SELECT_F_REACHABLE;
 625
 626         if (flags & RT6_F_STRICT)
 627                 strict = RT6_SELECT_F_IFACE;
 628
 629 relookup:
 630         read_lock_bh(&table->tb6_lock);
 631
 632 restart_2:
 633         fn = fib6_lookup(&table->tb6_root, &fl->fl6_dst, &fl->fl6_src);
 634
 635 restart:
 636         rt = rt6_select(&fn->leaf, fl->iif, strict | reachable);
 637         BACKTRACK();
 638         if (rt == &ip6_null_entry ||
 639             rt->rt6i_flags & RTF_CACHE)
 640                 goto out;
 641
 642         dst_hold(&rt->u.dst);
 643         read_unlock_bh(&table->tb6_lock);
 644
 645         if (!rt->rt6i_nexthop && !(rt->rt6i_flags & RTF_NONEXTHOP))
 646                 nrt = rt6_alloc_cow(rt, &fl->fl6_dst, &fl->fl6_src);
 647         else {
 648 #if CLONE_OFFLINK_ROUTE
 649                 nrt = rt6_alloc_clone(rt, &fl->fl6_dst);
 650 #else
 651                 goto out2;
 652 #endif
 653         }
 654
 655         dst_release(&rt->u.dst);
 656         rt = nrt ? : &ip6_null_entry;
 657
 658         dst_hold(&rt->u.dst);
 659         if (nrt) {
 660                 err = ip6_ins_rt(nrt, NULL, NULL, NULL);
 661                 if (!err)
 662                         goto out2;
 663         }
 664
 665         if (--attempts <= 0)
 666                 goto out2;
 667
 668         /*
 669          * Race condition! In the gap, when table->tb6_lock was
 670          * released someone could insert this route.  Relookup.
 671          */
 672         dst_release(&rt->u.dst);
 673         goto relookup;
 674
 675 out:
 676         if (reachable) {
 677                 reachable = 0;
 678                 goto restart_2;
 679         }
 680         dst_hold(&rt->u.dst);
 681         read_unlock_bh(&table->tb6_lock);
 682 out2:
 683         rt->u.dst.lastuse = jiffies;
 684         rt->u.dst.__use++;
 685
 686         return rt;
 687 }
 688
 689 void ip6_route_input(struct sk_buff *skb)
 690 {
 691         struct ipv6hdr *iph = skb->nh.ipv6h;
 692         struct flowi fl = {
 693                 .iif = skb->dev->ifindex,
 694                 .nl_u = {
 695                         .ip6_u = {
 696                                 .daddr = iph->daddr,
 697                                 .saddr = iph->saddr,
 698                                 .flowlabel = (* (u32 *) iph)&IPV6_FLOWINFO_MASK,
 699                         },
 700                 },
 701                 .proto = iph->nexthdr,
 702         };
 703         int flags = 0;
 704
 705         if (rt6_need_strict(&iph->daddr))
 706                 flags |= RT6_F_STRICT;
 707
 708         skb->dst = fib6_rule_lookup(&fl, flags, ip6_pol_route_input);
 709 }
 710
 711 static struct rt6_info *ip6_pol_route_output(struct fib6_table *table,
 712                                              struct flowi *fl, int flags)
 713 {
 714         struct fib6_node *fn;
 715         struct rt6_info *rt, *nrt;
 716         int strict = 0;
 717         int attempts = 3;
 718         int err;
 719         int reachable = RT6_SELECT_F_REACHABLE;
 720
 721         if (flags & RT6_F_STRICT)
 722                 strict = RT6_SELECT_F_IFACE;
 723
 724 relookup:
 725         read_lock_bh(&table->tb6_lock);
 726
 727 restart_2:
 728         fn = fib6_lookup(&table->tb6_root, &fl->fl6_dst, &fl->fl6_src);
 729
 730 restart:
 731         rt = rt6_select(&fn->leaf, fl->oif, strict | reachable);
 732         BACKTRACK();
 733         if (rt == &ip6_null_entry ||
 734             rt->rt6i_flags & RTF_CACHE)
 735                 goto out;
 736
 737         dst_hold(&rt->u.dst);
 738         read_unlock_bh(&table->tb6_lock);
 739
 740         if (!rt->rt6i_nexthop && !(rt->rt6i_flags & RTF_NONEXTHOP))
 741                 nrt = rt6_alloc_cow(rt, &fl->fl6_dst, &fl->fl6_src);
 742         else {
 743 #if CLONE_OFFLINK_ROUTE
 744                 nrt = rt6_alloc_clone(rt, &fl->fl6_dst);
 745 #else
 746                 goto out2;
 747 #endif
 748         }
 749
 750         dst_release(&rt->u.dst);
 751         rt = nrt ? : &ip6_null_entry;
 752
 753         dst_hold(&rt->u.dst);
 754         if (nrt) {
 755                 err = ip6_ins_rt(nrt, NULL, NULL, NULL);
 756                 if (!err)
 757                         goto out2;
 758         }
 759
 760         if (--attempts <= 0)
 761                 goto out2;
 762
 763         /*
 764          * Race condition! In the gap, when table->tb6_lock was
 765          * released someone could insert this route.  Relookup.
 766          */
 767         dst_release(&rt->u.dst);
 768         goto relookup;
 769
 770 out:
 771         if (reachable) {
 772                 reachable = 0;
 773                 goto restart_2;
 774         }
 775         dst_hold(&rt->u.dst);
 776         read_unlock_bh(&table->tb6_lock);
 777 out2:
 778         rt->u.dst.lastuse = jiffies;
 779         rt->u.dst.__use++;
 780         return rt;
 781 }
 782
 783 struct dst_entry * ip6_route_output(struct sock *sk, struct flowi *fl)
 784 {
 785         int flags = 0;
 786
 787         if (rt6_need_strict(&fl->fl6_dst))
 788                 flags |= RT6_F_STRICT;
 789
 790         return fib6_rule_lookup(fl, flags, ip6_pol_route_output);
 791 }
 792
 793
 794 /*
 795  *      Destination cache support functions
 796  */
 797
 798 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
 799 {
 800         struct rt6_info *rt;
 801
 802         rt = (struct rt6_info *) dst;
 803
 804         if (rt && rt->rt6i_node && (rt->rt6i_node->fn_sernum == cookie))
 805                 return dst;
 806
 807         return NULL;
 808 }
 809
 810 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
 811 {
 812         struct rt6_info *rt = (struct rt6_info *) dst;
 813
 814         if (rt) {
 815                 if (rt->rt6i_flags & RTF_CACHE)
 816                         ip6_del_rt(rt);
 817                 else
 818                         dst_release(dst);
 819         }
 820         return NULL;
 821 }
 822
 823 static void ip6_link_failure(struct sk_buff *skb)
 824 {
 825         struct rt6_info *rt;
 826
 827         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0, skb->dev);
 828
 829         rt = (struct rt6_info *) skb->dst;
 830         if (rt) {
 831                 if (rt->rt6i_flags&RTF_CACHE) {
 832                         dst_set_expires(&rt->u.dst, 0);
 833                         rt->rt6i_flags |= RTF_EXPIRES;
 834                 } else if (rt->rt6i_node && (rt->rt6i_flags & RTF_DEFAULT))
 835                         rt->rt6i_node->fn_sernum = -1;
 836         }
 837 }
 838
 839 static void ip6_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
 840 {
 841         struct rt6_info *rt6 = (struct rt6_info*)dst;
 842
 843         if (mtu < dst_mtu(dst) && rt6->rt6i_dst.plen == 128) {
 844                 rt6->rt6i_flags |= RTF_MODIFIED;
 845                 if (mtu < IPV6_MIN_MTU) {
 846                         mtu = IPV6_MIN_MTU;
 847                         dst->metrics[RTAX_FEATURES-1] |= RTAX_FEATURE_ALLFRAG;
 848                 }
 849                 dst->metrics[RTAX_MTU-1] = mtu;
 850                 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
 851         }
 852 }
 853
 854 static int ipv6_get_mtu(struct net_device *dev);
 855
 856 static inline unsigned int ipv6_advmss(unsigned int mtu)
 857 {
 858         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
 859
 860         if (mtu < ip6_rt_min_advmss)
 861                 mtu = ip6_rt_min_advmss;
 862
 863         /*
 864          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
 865          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
 866          * IPV6_MAXPLEN is also valid and means: "any MSS,
 867          * rely only on pmtu discovery"
 868          */
 869         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
 870                 mtu = IPV6_MAXPLEN;
 871         return mtu;
 872 }
 873
 874 static struct dst_entry *ndisc_dst_gc_list;
 875 static DEFINE_SPINLOCK(ndisc_lock);
 876
 877 struct dst_entry *ndisc_dst_alloc(struct net_device *dev,
 878                                   struct neighbour *neigh,
 879                                   struct in6_addr *addr,
 880                                   int (*output)(struct sk_buff *))
 881 {
 882         struct rt6_info *rt;
 883         struct inet6_dev *idev = in6_dev_get(dev);
 884
 885         if (unlikely(idev == NULL))
 886                 return NULL;
 887
 888         rt = ip6_dst_alloc();
 889         if (unlikely(rt == NULL)) {
 890                 in6_dev_put(idev);
 891                 goto out;
 892         }
 893
 894         dev_hold(dev);
 895         if (neigh)
 896                 neigh_hold(neigh);
 897         else
 898                 neigh = ndisc_get_neigh(dev, addr);
 899
 900         rt->rt6i_dev      = dev;
 901         rt->rt6i_idev     = idev;
 902         rt->rt6i_nexthop  = neigh;
 903         atomic_set(&rt->u.dst.__refcnt, 1);
 904         rt->u.dst.metrics[RTAX_HOPLIMIT-1] = 255;
 905         rt->u.dst.metrics[RTAX_MTU-1] = ipv6_get_mtu(rt->rt6i_dev);
 906         rt->u.dst.metrics[RTAX_ADVMSS-1] = ipv6_advmss(dst_mtu(&rt->u.dst));
 907         rt->u.dst.output  = output;
 908
 909 #if 0   /* there's no chance to use these for ndisc */
 910         rt->u.dst.flags   = ipv6_addr_type(addr) & IPV6_ADDR_UNICAST
 911                                 ? DST_HOST
 912                                 : 0;
 913         ipv6_addr_copy(&rt->rt6i_dst.addr, addr);
 914         rt->rt6i_dst.plen = 128;
 915 #endif
 916
 917         spin_lock_bh(&ndisc_lock);
 918         rt->u.dst.next = ndisc_dst_gc_list;
 919         ndisc_dst_gc_list = &rt->u.dst;
 920         spin_unlock_bh(&ndisc_lock);
 921
 922         fib6_force_start_gc();
 923
 924 out:
 925         return (struct dst_entry *)rt;
 926 }
 927
 928 int ndisc_dst_gc(int *more)
 929 {
 930         struct dst_entry *dst, *next, **pprev;
 931         int freed;
 932
 933         next = NULL;
 934         freed = 0;
 935
 936         spin_lock_bh(&ndisc_lock);
 937         pprev = &ndisc_dst_gc_list;
 938
 939         while ((dst = *pprev) != NULL) {
 940                 if (!atomic_read(&dst->__refcnt)) {
 941                         *pprev = dst->next;
 942                         dst_free(dst);
 943                         freed++;
 944                 } else {
 945                         pprev = &dst->next;
 946                         (*more)++;
 947                 }
 948         }
 949
 950         spin_unlock_bh(&ndisc_lock);
 951
 952         return freed;
 953 }
 954
 955 static int ip6_dst_gc(void)
 956 {
 957         static unsigned expire = 30*HZ;
 958         static unsigned long last_gc;
 959         unsigned long now = jiffies;
 960
 961         if (time_after(last_gc + ip6_rt_gc_min_interval, now) &&
 962             atomic_read(&ip6_dst_ops.entries) <= ip6_rt_max_size)
 963                 goto out;
 964
 965         expire++;
 966         fib6_run_gc(expire);
 967         last_gc = now;
 968         if (atomic_read(&ip6_dst_ops.entries) < ip6_dst_ops.gc_thresh)
 969                 expire = ip6_rt_gc_timeout>>1;
 970
 971 out:
 972         expire -= expire>>ip6_rt_gc_elasticity;
 973         return (atomic_read(&ip6_dst_ops.entries) > ip6_rt_max_size);
 974 }
 975
 976 /* Clean host part of a prefix. Not necessary in radix tree,
 977    but results in cleaner routing tables.
 978
 979    Remove it only when all the things will work!
 980  */
 981
 982 static int ipv6_get_mtu(struct net_device *dev)
 983 {
 984         int mtu = IPV6_MIN_MTU;
 985         struct inet6_dev *idev;
 986
 987         idev = in6_dev_get(dev);
 988         if (idev) {
 989                 mtu = idev->cnf.mtu6;
 990                 in6_dev_put(idev);
 991         }
 992         return mtu;
 993 }
 994
 995 int ipv6_get_hoplimit(struct net_device *dev)
 996 {
 997         int hoplimit = ipv6_devconf.hop_limit;
 998         struct inet6_dev *idev;
 999
1000         idev = in6_dev_get(dev);
1001         if (idev) {
1002                 hoplimit = idev->cnf.hop_limit;
1003                 in6_dev_put(idev);
1004         }
1005         return hoplimit;
1006 }
1007
1008 /*
1009  *
1010  */
1011
1012 int ip6_route_add(struct in6_rtmsg *rtmsg, struct nlmsghdr *nlh,
1013                   void *_rtattr, struct netlink_skb_parms *req,
1014                   u32 table_id)
1015 {
1016         int err;
1017         struct rtmsg *r;
1018         struct rtattr **rta;
1019         struct rt6_info *rt = NULL;
1020         struct net_device *dev = NULL;
1021         struct inet6_dev *idev = NULL;
1022         struct fib6_table *table;
1023         int addr_type;
1024
1025         rta = (struct rtattr **) _rtattr;
1026
1027         if (rtmsg->rtmsg_dst_len > 128 || rtmsg->rtmsg_src_len > 128)
1028                 return -EINVAL;
1029 #ifndef CONFIG_IPV6_SUBTREES
1030         if (rtmsg->rtmsg_src_len)
1031                 return -EINVAL;
1032 #endif
1033         if (rtmsg->rtmsg_ifindex) {
1034                 err = -ENODEV;
1035                 dev = dev_get_by_index(rtmsg->rtmsg_ifindex);
1036                 if (!dev)
1037                         goto out;
1038                 idev = in6_dev_get(dev);
1039                 if (!idev)
1040                         goto out;
1041         }
1042
1043         if (rtmsg->rtmsg_metric == 0)
1044                 rtmsg->rtmsg_metric = IP6_RT_PRIO_USER;
1045
1046         table = fib6_new_table(table_id);
1047         if (table == NULL) {
1048                 err = -ENOBUFS;
1049                 goto out;
1050         }
1051
1052         rt = ip6_dst_alloc();
1053
1054         if (rt == NULL) {
1055                 err = -ENOMEM;
1056                 goto out;
1057         }
1058
1059         rt->u.dst.obsolete = -1;
1060         rt->rt6i_expires = jiffies + clock_t_to_jiffies(rtmsg->rtmsg_info);
1061         if (nlh && (r = NLMSG_DATA(nlh))) {
1062                 rt->rt6i_protocol = r->rtm_protocol;
1063         } else {
1064                 rt->rt6i_protocol = RTPROT_BOOT;
1065         }
1066
1067         addr_type = ipv6_addr_type(&rtmsg->rtmsg_dst);
1068
1069         if (addr_type & IPV6_ADDR_MULTICAST)
1070                 rt->u.dst.input = ip6_mc_input;
1071         else
1072                 rt->u.dst.input = ip6_forward;
1073
1074         rt->u.dst.output = ip6_output;
1075
1076         ipv6_addr_prefix(&rt->rt6i_dst.addr,
1077                          &rtmsg->rtmsg_dst, rtmsg->rtmsg_dst_len);
1078         rt->rt6i_dst.plen = rtmsg->rtmsg_dst_len;
1079         if (rt->rt6i_dst.plen == 128)
1080                rt->u.dst.flags = DST_HOST;
1081
1082 #ifdef CONFIG_IPV6_SUBTREES
1083         ipv6_addr_prefix(&rt->rt6i_src.addr,
1084                          &rtmsg->rtmsg_src, rtmsg->rtmsg_src_len);
1085         rt->rt6i_src.plen = rtmsg->rtmsg_src_len;
1086 #endif
1087
1088         rt->rt6i_metric = rtmsg->rtmsg_metric;
1089
1090         /* We cannot add true routes via loopback here,
1091            they would result in kernel looping; promote them to reject routes
1092          */
1093         if ((rtmsg->rtmsg_flags&RTF_REJECT) ||
1094             (dev && (dev->flags&IFF_LOOPBACK) && !(addr_type&IPV6_ADDR_LOOPBACK))) {
1095                 /* hold loopback dev/idev if we haven't done so. */
1096                 if (dev != &loopback_dev) {
1097                         if (dev) {
1098                                 dev_put(dev);
1099                                 in6_dev_put(idev);
1100                         }
1101                         dev = &loopback_dev;
1102                         dev_hold(dev);
1103                         idev = in6_dev_get(dev);
1104                         if (!idev) {
1105                                 err = -ENODEV;
1106                                 goto out;
1107                         }
1108                 }
1109                 rt->u.dst.output = ip6_pkt_discard_out;
1110                 rt->u.dst.input = ip6_pkt_discard;
1111                 rt->u.dst.error = -ENETUNREACH;
1112                 rt->rt6i_flags = RTF_REJECT|RTF_NONEXTHOP;
1113                 goto install_route;
1114         }
1115
1116         if (rtmsg->rtmsg_flags & RTF_GATEWAY) {
1117                 struct in6_addr *gw_addr;
1118                 int gwa_type;
1119
1120                 gw_addr = &rtmsg->rtmsg_gateway;
1121                 ipv6_addr_copy(&rt->rt6i_gateway, &rtmsg->rtmsg_gateway);
1122                 gwa_type = ipv6_addr_type(gw_addr);
1123
1124                 if (gwa_type != (IPV6_ADDR_LINKLOCAL|IPV6_ADDR_UNICAST)) {
1125                         struct rt6_info *grt;
1126
1127                         /* IPv6 strictly inhibits using not link-local
1128                            addresses as nexthop address.
1129                            Otherwise, router will not able to send redirects.
1130                            It is very good, but in some (rare!) circumstances
1131                            (SIT, PtP, NBMA NOARP links) it is handy to allow
1132                            some exceptions. --ANK
1133                          */
1134                         err = -EINVAL;
1135                         if (!(gwa_type&IPV6_ADDR_UNICAST))
1136                                 goto out;
1137
1138                         grt = rt6_lookup(gw_addr, NULL, rtmsg->rtmsg_ifindex, 1);
1139
1140                         err = -EHOSTUNREACH;
1141                         if (grt == NULL)
1142                                 goto out;
1143                         if (dev) {
1144                                 if (dev != grt->rt6i_dev) {
1145                                         dst_release(&grt->u.dst);
1146                                         goto out;
1147                                 }
1148                         } else {
1149                                 dev = grt->rt6i_dev;
1150                                 idev = grt->rt6i_idev;
1151                                 dev_hold(dev);
1152                                 in6_dev_hold(grt->rt6i_idev);
1153                         }
1154                         if (!(grt->rt6i_flags&RTF_GATEWAY))
1155                                 err = 0;
1156                         dst_release(&grt->u.dst);
1157
1158                         if (err)
1159                                 goto out;
1160                 }
1161                 err = -EINVAL;
1162                 if (dev == NULL || (dev->flags&IFF_LOOPBACK))
1163                         goto out;
1164         }
1165
1166         err = -ENODEV;
1167         if (dev == NULL)
1168                 goto out;
1169
1170         if (rtmsg->rtmsg_flags & (RTF_GATEWAY|RTF_NONEXTHOP)) {
1171                 rt->rt6i_nexthop = __neigh_lookup_errno(&nd_tbl, &rt->rt6i_gateway, dev);
1172                 if (IS_ERR(rt->rt6i_nexthop)) {
1173                         err = PTR_ERR(rt->rt6i_nexthop);
1174                         rt->rt6i_nexthop = NULL;
1175                         goto out;
1176                 }
1177         }
1178
1179         rt->rt6i_flags = rtmsg->rtmsg_flags;
1180
1181 install_route:
1182         if (rta && rta[RTA_METRICS-1]) {
1183                 int attrlen = RTA_PAYLOAD(rta[RTA_METRICS-1]);
1184                 struct rtattr *attr = RTA_DATA(rta[RTA_METRICS-1]);
1185
1186                 while (RTA_OK(attr, attrlen)) {
1187                         unsigned flavor = attr->rta_type;
1188                         if (flavor) {
1189                                 if (flavor > RTAX_MAX) {
1190                                         err = -EINVAL;
1191                                         goto out;
1192                                 }
1193                                 rt->u.dst.metrics[flavor-1] =
1194                                         *(u32 *)RTA_DATA(attr);
1195                         }
1196                         attr = RTA_NEXT(attr, attrlen);
1197                 }
1198         }
1199
1200         if (rt->u.dst.metrics[RTAX_HOPLIMIT-1] == 0)
1201                 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = -1;
1202         if (!rt->u.dst.metrics[RTAX_MTU-1])
1203                 rt->u.dst.metrics[RTAX_MTU-1] = ipv6_get_mtu(dev);
1204         if (!rt->u.dst.metrics[RTAX_ADVMSS-1])
1205                 rt->u.dst.metrics[RTAX_ADVMSS-1] = ipv6_advmss(dst_mtu(&rt->u.dst));
1206         rt->u.dst.dev = dev;
1207         rt->rt6i_idev = idev;
1208         rt->rt6i_table = table;
1209         return ip6_ins_rt(rt, nlh, _rtattr, req);
1210
1211 out:
1212         if (dev)
1213                 dev_put(dev);
1214         if (idev)
1215                 in6_dev_put(idev);
1216         if (rt)
1217                 dst_free((struct dst_entry *) rt);
1218         return err;
1219 }
1220
1221 static int __ip6_del_rt(struct rt6_info *rt, struct nlmsghdr *nlh,
1222                         void *_rtattr, struct netlink_skb_parms *req)
1223 {
1224         int err;
1225         struct fib6_table *table;
1226
1227         if (rt == &ip6_null_entry)
1228                 return -ENOENT;
1229
1230         table = rt->rt6i_table;
1231         write_lock_bh(&table->tb6_lock);
1232
1233         err = fib6_del(rt, nlh, _rtattr, req);
1234         dst_release(&rt->u.dst);
1235
1236         write_unlock_bh(&table->tb6_lock);
1237
1238         return err;
1239 }
1240
1241 int ip6_del_rt(struct rt6_info *rt)
1242 {
1243         return __ip6_del_rt(rt, NULL, NULL, NULL);
1244 }
1245
1246 static int ip6_route_del(struct in6_rtmsg *rtmsg, struct nlmsghdr *nlh,
1247                          void *_rtattr, struct netlink_skb_parms *req,
1248                          u32 table_id)
1249 {
1250         struct fib6_table *table;
1251         struct fib6_node *fn;
1252         struct rt6_info *rt;
1253         int err = -ESRCH;
1254
1255         table = fib6_get_table(table_id);
1256         if (table == NULL)
1257                 return err;
1258
1259         read_lock_bh(&table->tb6_lock);
1260
1261         fn = fib6_locate(&table->tb6_root,
1262                          &rtmsg->rtmsg_dst, rtmsg->rtmsg_dst_len,
1263                          &rtmsg->rtmsg_src, rtmsg->rtmsg_src_len);
1264
1265         if (fn) {
1266                 for (rt = fn->leaf; rt; rt = rt->u.next) {
1267                         if (rtmsg->rtmsg_ifindex &&
1268                             (rt->rt6i_dev == NULL ||
1269                              rt->rt6i_dev->ifindex != rtmsg->rtmsg_ifindex))
1270                                 continue;
1271                         if (rtmsg->rtmsg_flags&RTF_GATEWAY &&
1272                             !ipv6_addr_equal(&rtmsg->rtmsg_gateway, &rt->rt6i_gateway))
1273                                 continue;
1274                         if (rtmsg->rtmsg_metric &&
1275                             rtmsg->rtmsg_metric != rt->rt6i_metric)
1276                                 continue;
1277                         dst_hold(&rt->u.dst);
1278                         read_unlock_bh(&table->tb6_lock);
1279
1280                         return __ip6_del_rt(rt, nlh, _rtattr, req);
1281                 }
1282         }
1283         read_unlock_bh(&table->tb6_lock);
1284
1285         return err;
1286 }
1287
1288 /*
1289  *      Handle redirects
1290  */
1291 void rt6_redirect(struct in6_addr *dest, struct in6_addr *saddr,
1292                   struct neighbour *neigh, u8 *lladdr, int on_link)
1293 {
1294         struct rt6_info *rt, *nrt = NULL;
1295         struct fib6_node *fn;
1296         struct fib6_table *table;
1297         struct netevent_redirect netevent;
1298
1299         /* TODO: Very lazy, might need to check all tables */
1300         table = fib6_get_table(RT6_TABLE_MAIN);
1301         if (table == NULL)
1302                 return;
1303
1304         /*
1305          * Get the "current" route for this destination and
1306          * check if the redirect has come from approriate router.
1307          *
1308          * RFC 2461 specifies that redirects should only be
1309          * accepted if they come from the nexthop to the target.
1310          * Due to the way the routes are chosen, this notion
1311          * is a bit fuzzy and one might need to check all possible
1312          * routes.
1313          */
1314
1315         read_lock_bh(&table->tb6_lock);
1316         fn = fib6_lookup(&table->tb6_root, dest, NULL);
1317 restart:
1318         for (rt = fn->leaf; rt; rt = rt->u.next) {
1319                 /*
1320                  * Current route is on-link; redirect is always invalid.
1321                  *
1322                  * Seems, previous statement is not true. It could
1323                  * be node, which looks for us as on-link (f.e. proxy ndisc)
1324                  * But then router serving it might decide, that we should
1325                  * know truth 8)8) --ANK (980726).
1326                  */
1327                 if (rt6_check_expired(rt))
1328                         continue;
1329                 if (!(rt->rt6i_flags & RTF_GATEWAY))
1330                         continue;
1331                 if (neigh->dev != rt->rt6i_dev)
1332                         continue;
1333                 if (!ipv6_addr_equal(saddr, &rt->rt6i_gateway))
1334                         continue;
1335                 break;
1336         }
1337         if (rt)
1338                 dst_hold(&rt->u.dst);
1339         else if (rt6_need_strict(dest)) {
1340                 while ((fn = fn->parent) != NULL) {
1341                         if (fn->fn_flags & RTN_ROOT)
1342                                 break;
1343                         if (fn->fn_flags & RTN_RTINFO)
1344                                 goto restart;
1345                 }
1346         }
1347         read_unlock_bh(&table->tb6_lock);
1348
1349         if (!rt) {
1350                 if (net_ratelimit())
1351                         printk(KERN_DEBUG "rt6_redirect: source isn't a valid nexthop "
1352                                "for redirect target\n");
1353                 return;
1354         }
1355
1356         /*
1357          *      We have finally decided to accept it.
1358          */
1359
1360         neigh_update(neigh, lladdr, NUD_STALE,
1361                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
1362                      NEIGH_UPDATE_F_OVERRIDE|
1363                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
1364                                      NEIGH_UPDATE_F_ISROUTER))
1365                      );
1366
1367         /*
1368          * Redirect received -> path was valid.
1369          * Look, redirects are sent only in response to data packets,
1370          * so that this nexthop apparently is reachable. --ANK
1371          */
1372         dst_confirm(&rt->u.dst);
1373
1374         /* Duplicate redirect: silently ignore. */
1375         if (neigh == rt->u.dst.neighbour)
1376                 goto out;
1377
1378         nrt = ip6_rt_copy(rt);
1379         if (nrt == NULL)
1380                 goto out;
1381
1382         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
1383         if (on_link)
1384                 nrt->rt6i_flags &= ~RTF_GATEWAY;
1385
1386         ipv6_addr_copy(&nrt->rt6i_dst.addr, dest);
1387         nrt->rt6i_dst.plen = 128;
1388         nrt->u.dst.flags |= DST_HOST;
1389
1390         ipv6_addr_copy(&nrt->rt6i_gateway, (struct in6_addr*)neigh->primary_key);
1391         nrt->rt6i_nexthop = neigh_clone(neigh);
1392         /* Reset pmtu, it may be better */
1393         nrt->u.dst.metrics[RTAX_MTU-1] = ipv6_get_mtu(neigh->dev);
1394         nrt->u.dst.metrics[RTAX_ADVMSS-1] = ipv6_advmss(dst_mtu(&nrt->u.dst));
1395
1396         if (ip6_ins_rt(nrt, NULL, NULL, NULL))
1397                 goto out;
1398
1399         netevent.old = &rt->u.dst;
1400         netevent.new = &nrt->u.dst;
1401         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
1402
1403         if (rt->rt6i_flags&RTF_CACHE) {
1404                 ip6_del_rt(rt);
1405                 return;
1406         }
1407
1408 out:
1409         dst_release(&rt->u.dst);
1410         return;
1411 }
1412
1413 /*
1414  *      Handle ICMP "packet too big" messages
1415  *      i.e. Path MTU discovery
1416  */
1417
1418 void rt6_pmtu_discovery(struct in6_addr *daddr, struct in6_addr *saddr,
1419                         struct net_device *dev, u32 pmtu)
1420 {
1421         struct rt6_info *rt, *nrt;
1422         int allfrag = 0;
1423
1424         rt = rt6_lookup(daddr, saddr, dev->ifindex, 0);
1425         if (rt == NULL)
1426                 return;
1427
1428         if (pmtu >= dst_mtu(&rt->u.dst))
1429                 goto out;
1430
1431         if (pmtu < IPV6_MIN_MTU) {
1432                 /*
1433                  * According to RFC2460, PMTU is set to the IPv6 Minimum Link
1434                  * MTU (1280) and a fragment header should always be included
1435                  * after a node receiving Too Big message reporting PMTU is
1436                  * less than the IPv6 Minimum Link MTU.
1437                  */
1438                 pmtu = IPV6_MIN_MTU;
1439                 allfrag = 1;
1440         }
1441
1442         /* New mtu received -> path was valid.
1443            They are sent only in response to data packets,
1444            so that this nexthop apparently is reachable. --ANK
1445          */
1446         dst_confirm(&rt->u.dst);
1447
1448         /* Host route. If it is static, it would be better
1449            not to override it, but add new one, so that
1450            when cache entry will expire old pmtu
1451            would return automatically.
1452          */
1453         if (rt->rt6i_flags & RTF_CACHE) {
1454                 rt->u.dst.metrics[RTAX_MTU-1] = pmtu;
1455                 if (allfrag)
1456                         rt->u.dst.metrics[RTAX_FEATURES-1] |= RTAX_FEATURE_ALLFRAG;
1457                 dst_set_expires(&rt->u.dst, ip6_rt_mtu_expires);
1458                 rt->rt6i_flags |= RTF_MODIFIED|RTF_EXPIRES;
1459                 goto out;
1460         }
1461
1462         /* Network route.
1463            Two cases are possible:
1464            1. It is connected route. Action: COW
1465            2. It is gatewayed route or NONEXTHOP route. Action: clone it.
1466          */
1467         if (!rt->rt6i_nexthop && !(rt->rt6i_flags & RTF_NONEXTHOP))
1468                 nrt = rt6_alloc_cow(rt, daddr, saddr);
1469         else
1470                 nrt = rt6_alloc_clone(rt, daddr);
1471
1472         if (nrt) {
1473                 nrt->u.dst.metrics[RTAX_MTU-1] = pmtu;
1474                 if (allfrag)
1475                         nrt->u.dst.metrics[RTAX_FEATURES-1] |= RTAX_FEATURE_ALLFRAG;
1476
1477                 /* According to RFC 1981, detecting PMTU increase shouldn't be
1478                  * happened within 5 mins, the recommended timer is 10 mins.
1479                  * Here this route expiration time is set to ip6_rt_mtu_expires
1480                  * which is 10 mins. After 10 mins the decreased pmtu is expired
1481                  * and detecting PMTU increase will be automatically happened.
1482                  */
1483                 dst_set_expires(&nrt->u.dst, ip6_rt_mtu_expires);
1484                 nrt->rt6i_flags |= RTF_DYNAMIC|RTF_EXPIRES;
1485
1486                 ip6_ins_rt(nrt, NULL, NULL, NULL);
1487         }
1488 out:
1489         dst_release(&rt->u.dst);
1490 }
1491
1492 /*
1493  *      Misc support functions
1494  */
1495
1496 static struct rt6_info * ip6_rt_copy(struct rt6_info *ort)
1497 {
1498         struct rt6_info *rt = ip6_dst_alloc();
1499
1500         if (rt) {
1501                 rt->u.dst.input = ort->u.dst.input;
1502                 rt->u.dst.output = ort->u.dst.output;
1503
1504                 memcpy(rt->u.dst.metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
1505                 rt->u.dst.dev = ort->u.dst.dev;
1506                 if (rt->u.dst.dev)
1507                         dev_hold(rt->u.dst.dev);
1508                 rt->rt6i_idev = ort->rt6i_idev;
1509                 if (rt->rt6i_idev)
1510                         in6_dev_hold(rt->rt6i_idev);
1511                 rt->u.dst.lastuse = jiffies;
1512                 rt->rt6i_expires = 0;
1513
1514                 ipv6_addr_copy(&rt->rt6i_gateway, &ort->rt6i_gateway);
1515                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_EXPIRES;
1516                 rt->rt6i_metric = 0;
1517
1518                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
1519 #ifdef CONFIG_IPV6_SUBTREES
1520                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
1521 #endif
1522                 rt->rt6i_table = ort->rt6i_table;
1523         }
1524         return rt;
1525 }
1526
1527 #ifdef CONFIG_IPV6_ROUTE_INFO
1528 static struct rt6_info *rt6_get_route_info(struct in6_addr *prefix, int prefixlen,
1529                                            struct in6_addr *gwaddr, int ifindex)
1530 {
1531         struct fib6_node *fn;
1532         struct rt6_info *rt = NULL;
1533         struct fib6_table *table;
1534
1535         table = fib6_get_table(RT6_TABLE_INFO);
1536         if (table == NULL)
1537                 return NULL;
1538
1539         write_lock_bh(&table->tb6_lock);
1540         fn = fib6_locate(&table->tb6_root, prefix ,prefixlen, NULL, 0);
1541         if (!fn)
1542                 goto out;
1543
1544         for (rt = fn->leaf; rt; rt = rt->u.next) {
1545                 if (rt->rt6i_dev->ifindex != ifindex)
1546                         continue;
1547                 if ((rt->rt6i_flags & (RTF_ROUTEINFO|RTF_GATEWAY)) != (RTF_ROUTEINFO|RTF_GATEWAY))
1548                         continue;
1549                 if (!ipv6_addr_equal(&rt->rt6i_gateway, gwaddr))
1550                         continue;
1551                 dst_hold(&rt->u.dst);
1552                 break;
1553         }
1554 out:
1555         write_unlock_bh(&table->tb6_lock);
1556         return rt;
1557 }
1558
1559 static struct rt6_info *rt6_add_route_info(struct in6_addr *prefix, int prefixlen,
1560                                            struct in6_addr *gwaddr, int ifindex,
1561                                            unsigned pref)
1562 {
1563         struct in6_rtmsg rtmsg;
1564
1565         memset(&rtmsg, 0, sizeof(rtmsg));
1566         rtmsg.rtmsg_type = RTMSG_NEWROUTE;
1567         ipv6_addr_copy(&rtmsg.rtmsg_dst, prefix);
1568         rtmsg.rtmsg_dst_len = prefixlen;
1569         ipv6_addr_copy(&rtmsg.rtmsg_gateway, gwaddr);
1570         rtmsg.rtmsg_metric = 1024;
1571         rtmsg.rtmsg_flags = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO | RTF_UP | RTF_PREF(pref);
1572         /* We should treat it as a default route if prefix length is 0. */
1573         if (!prefixlen)
1574                 rtmsg.rtmsg_flags |= RTF_DEFAULT;
1575         rtmsg.rtmsg_ifindex = ifindex;
1576
1577         ip6_route_add(&rtmsg, NULL, NULL, NULL, RT6_TABLE_INFO);
1578
1579         return rt6_get_route_info(prefix, prefixlen, gwaddr, ifindex);
1580 }
1581 #endif
1582
1583 struct rt6_info *rt6_get_dflt_router(struct in6_addr *addr, struct net_device *dev)
1584 {
1585         struct rt6_info *rt;
1586         struct fib6_table *table;
1587
1588         table = fib6_get_table(RT6_TABLE_DFLT);
1589         if (table == NULL)
1590                 return NULL;
1591
1592         write_lock_bh(&table->tb6_lock);
1593         for (rt = table->tb6_root.leaf; rt; rt=rt->u.next) {
1594                 if (dev == rt->rt6i_dev &&
1595                     ((rt->rt6i_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
1596                     ipv6_addr_equal(&rt->rt6i_gateway, addr))
1597                         break;
1598         }
1599         if (rt)
1600                 dst_hold(&rt->u.dst);
1601         write_unlock_bh(&table->tb6_lock);
1602         return rt;
1603 }
1604
1605 struct rt6_info *rt6_add_dflt_router(struct in6_addr *gwaddr,
1606                                      struct net_device *dev,
1607                                      unsigned int pref)
1608 {
1609         struct in6_rtmsg rtmsg;
1610
1611         memset(&rtmsg, 0, sizeof(struct in6_rtmsg));
1612         rtmsg.rtmsg_type = RTMSG_NEWROUTE;
1613         ipv6_addr_copy(&rtmsg.rtmsg_gateway, gwaddr);
1614         rtmsg.rtmsg_metric = 1024;
1615         rtmsg.rtmsg_flags = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT | RTF_UP | RTF_EXPIRES |
1616                             RTF_PREF(pref);
1617
1618         rtmsg.rtmsg_ifindex = dev->ifindex;
1619
1620         ip6_route_add(&rtmsg, NULL, NULL, NULL, RT6_TABLE_DFLT);
1621         return rt6_get_dflt_router(gwaddr, dev);
1622 }
1623
1624 void rt6_purge_dflt_routers(void)
1625 {
1626         struct rt6_info *rt;
1627         struct fib6_table *table;
1628
1629         /* NOTE: Keep consistent with rt6_get_dflt_router */
1630         table = fib6_get_table(RT6_TABLE_DFLT);
1631         if (table == NULL)
1632                 return;
1633
1634 restart:
1635         read_lock_bh(&table->tb6_lock);
1636         for (rt = table->tb6_root.leaf; rt; rt = rt->u.next) {
1637                 if (rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF)) {
1638                         dst_hold(&rt->u.dst);
1639                         read_unlock_bh(&table->tb6_lock);
1640                         ip6_del_rt(rt);
1641                         goto restart;
1642                 }
1643         }
1644         read_unlock_bh(&table->tb6_lock);
1645 }
1646
1647 int ipv6_route_ioctl(unsigned int cmd, void __user *arg)
1648 {
1649         struct in6_rtmsg rtmsg;
1650         int err;
1651
1652         switch(cmd) {
1653         case SIOCADDRT:         /* Add a route */
1654         case SIOCDELRT:         /* Delete a route */
1655                 if (!capable(CAP_NET_ADMIN))
1656                         return -EPERM;
1657                 err = copy_from_user(&rtmsg, arg,
1658                                      sizeof(struct in6_rtmsg));
1659                 if (err)
1660                         return -EFAULT;
1661
1662                 rtnl_lock();
1663                 switch (cmd) {
1664                 case SIOCADDRT:
1665                         err = ip6_route_add(&rtmsg, NULL, NULL, NULL,
1666                                             RT6_TABLE_MAIN);
1667                         break;
1668                 case SIOCDELRT:
1669                         err = ip6_route_del(&rtmsg, NULL, NULL, NULL,
1670                                             RT6_TABLE_MAIN);
1671                         break;
1672                 default:
1673                         err = -EINVAL;
1674                 }
1675                 rtnl_unlock();
1676
1677                 return err;
1678         };
1679
1680         return -EINVAL;
1681 }
1682
1683 /*
1684  *      Drop the packet on the floor
1685  */
1686
1687 static int ip6_pkt_discard(struct sk_buff *skb)
1688 {
1689         int type = ipv6_addr_type(&skb->nh.ipv6h->daddr);
1690         if (type == IPV6_ADDR_ANY || type == IPV6_ADDR_RESERVED)
1691                 IP6_INC_STATS(IPSTATS_MIB_INADDRERRORS);
1692
1693         IP6_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
1694         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_NOROUTE, 0, skb->dev);
1695         kfree_skb(skb);
1696         return 0;
1697 }
1698
1699 static int ip6_pkt_discard_out(struct sk_buff *skb)
1700 {
1701         skb->dev = skb->dst->dev;
1702         return ip6_pkt_discard(skb);
1703 }
1704
1705 /*
1706  *      Allocate a dst for local (unicast / anycast) address.
1707  */
1708
1709 struct rt6_info *addrconf_dst_alloc(struct inet6_dev *idev,
1710                                     const struct in6_addr *addr,
1711                                     int anycast)
1712 {
1713         struct rt6_info *rt = ip6_dst_alloc();
1714
1715         if (rt == NULL)
1716                 return ERR_PTR(-ENOMEM);
1717
1718         dev_hold(&loopback_dev);
1719         in6_dev_hold(idev);
1720
1721         rt->u.dst.flags = DST_HOST;
1722         rt->u.dst.input = ip6_input;
1723         rt->u.dst.output = ip6_output;
1724         rt->rt6i_dev = &loopback_dev;
1725         rt->rt6i_idev = idev;
1726         rt->u.dst.metrics[RTAX_MTU-1] = ipv6_get_mtu(rt->rt6i_dev);
1727         rt->u.dst.metrics[RTAX_ADVMSS-1] = ipv6_advmss(dst_mtu(&rt->u.dst));
1728         rt->u.dst.metrics[RTAX_HOPLIMIT-1] = -1;
1729         rt->u.dst.obsolete = -1;
1730
1731         rt->rt6i_flags = RTF_UP | RTF_NONEXTHOP;
1732         if (anycast)
1733                 rt->rt6i_flags |= RTF_ANYCAST;
1734         else
1735                 rt->rt6i_flags |= RTF_LOCAL;
1736         rt->rt6i_nexthop = ndisc_get_neigh(rt->rt6i_dev, &rt->rt6i_gateway);
1737         if (rt->rt6i_nexthop == NULL) {
1738                 dst_free((struct dst_entry *) rt);
1739                 return ERR_PTR(-ENOMEM);
1740         }
1741
1742         ipv6_addr_copy(&rt->rt6i_dst.addr, addr);
1743         rt->rt6i_dst.plen = 128;
1744         rt->rt6i_table = fib6_get_table(RT6_TABLE_LOCAL);
1745
1746         atomic_set(&rt->u.dst.__refcnt, 1);
1747
1748         return rt;
1749 }
1750
1751 static int fib6_ifdown(struct rt6_info *rt, void *arg)
1752 {
1753         if (((void*)rt->rt6i_dev == arg || arg == NULL) &&
1754             rt != &ip6_null_entry) {
1755                 RT6_TRACE("deleted by ifdown %p\n", rt);
1756                 return -1;
1757         }
1758         return 0;
1759 }
1760
1761 void rt6_ifdown(struct net_device *dev)
1762 {
1763         fib6_clean_all(fib6_ifdown, 0, dev);
1764 }
1765
1766 struct rt6_mtu_change_arg
1767 {
1768         struct net_device *dev;
1769         unsigned mtu;
1770 };
1771
1772 static int rt6_mtu_change_route(struct rt6_info *rt, void *p_arg)
1773 {
1774         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
1775         struct inet6_dev *idev;
1776
1777         /* In IPv6 pmtu discovery is not optional,
1778            so that RTAX_MTU lock cannot disable it.
1779            We still use this lock to block changes
1780            caused by addrconf/ndisc.
1781         */
1782
1783         idev = __in6_dev_get(arg->dev);
1784         if (idev == NULL)
1785                 return 0;
1786
1787         /* For administrative MTU increase, there is no way to discover
1788            IPv6 PMTU increase, so PMTU increase should be updated here.
1789            Since RFC 1981 doesn't include administrative MTU increase
1790            update PMTU increase is a MUST. (i.e. jumbo frame)
1791          */
1792         /*
1793            If new MTU is less than route PMTU, this new MTU will be the
1794            lowest MTU in the path, update the route PMTU to reflect PMTU
1795            decreases; if new MTU is greater than route PMTU, and the
1796            old MTU is the lowest MTU in the path, update the route PMTU
1797            to reflect the increase. In this case if the other nodes' MTU
1798            also have the lowest MTU, TOO BIG MESSAGE will be lead to
1799            PMTU discouvery.
1800          */
1801         if (rt->rt6i_dev == arg->dev &&
1802             !dst_metric_locked(&rt->u.dst, RTAX_MTU) &&
1803             (dst_mtu(&rt->u.dst) > arg->mtu ||
1804              (dst_mtu(&rt->u.dst) < arg->mtu &&
1805               dst_mtu(&rt->u.dst) == idev->cnf.mtu6)))
1806                 rt->u.dst.metrics[RTAX_MTU-1] = arg->mtu;
1807         rt->u.dst.metrics[RTAX_ADVMSS-1] = ipv6_advmss(arg->mtu);
1808         return 0;
1809 }
1810
1811 void rt6_mtu_change(struct net_device *dev, unsigned mtu)
1812 {
1813         struct rt6_mtu_change_arg arg = {
1814                 .dev = dev,
1815                 .mtu = mtu,
1816         };
1817
1818         fib6_clean_all(rt6_mtu_change_route, 0, &arg);
1819 }
1820
1821 static int inet6_rtm_to_rtmsg(struct rtmsg *r, struct rtattr **rta,
1822                               struct in6_rtmsg *rtmsg)
1823 {
1824         memset(rtmsg, 0, sizeof(*rtmsg));
1825
1826         rtmsg->rtmsg_dst_len = r->rtm_dst_len;
1827         rtmsg->rtmsg_src_len = r->rtm_src_len;
1828         rtmsg->rtmsg_flags = RTF_UP;
1829         if (r->rtm_type == RTN_UNREACHABLE)
1830                 rtmsg->rtmsg_flags |= RTF_REJECT;
1831
1832         if (rta[RTA_GATEWAY-1]) {
1833                 if (rta[RTA_GATEWAY-1]->rta_len != RTA_LENGTH(16))
1834                         return -EINVAL;
1835                 memcpy(&rtmsg->rtmsg_gateway, RTA_DATA(rta[RTA_GATEWAY-1]), 16);
1836                 rtmsg->rtmsg_flags |= RTF_GATEWAY;
1837         }
1838         if (rta[RTA_DST-1]) {
1839                 if (RTA_PAYLOAD(rta[RTA_DST-1]) < ((r->rtm_dst_len+7)>>3))
1840                         return -EINVAL;
1841                 memcpy(&rtmsg->rtmsg_dst, RTA_DATA(rta[RTA_DST-1]), ((r->rtm_dst_len+7)>>3));
1842         }
1843         if (rta[RTA_SRC-1]) {
1844                 if (RTA_PAYLOAD(rta[RTA_SRC-1]) < ((r->rtm_src_len+7)>>3))
1845                         return -EINVAL;
1846                 memcpy(&rtmsg->rtmsg_src, RTA_DATA(rta[RTA_SRC-1]), ((r->rtm_src_len+7)>>3));
1847         }
1848         if (rta[RTA_OIF-1]) {
1849                 if (rta[RTA_OIF-1]->rta_len != RTA_LENGTH(sizeof(int)))
1850                         return -EINVAL;
1851                 memcpy(&rtmsg->rtmsg_ifindex, RTA_DATA(rta[RTA_OIF-1]), sizeof(int));
1852         }
1853         if (rta[RTA_PRIORITY-1]) {
1854                 if (rta[RTA_PRIORITY-1]->rta_len != RTA_LENGTH(4))
1855                         return -EINVAL;
1856                 memcpy(&rtmsg->rtmsg_metric, RTA_DATA(rta[RTA_PRIORITY-1]), 4);
1857         }
1858         return 0;
1859 }
1860
1861 int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
1862 {
1863         struct rtmsg *r = NLMSG_DATA(nlh);
1864         struct in6_rtmsg rtmsg;
1865
1866         if (inet6_rtm_to_rtmsg(r, arg, &rtmsg))
1867                 return -EINVAL;
1868         return ip6_route_del(&rtmsg, nlh, arg, &NETLINK_CB(skb),
1869                              rtm_get_table(arg, r->rtm_table));
1870 }
1871
1872 int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg)
1873 {
1874         struct rtmsg *r = NLMSG_DATA(nlh);
1875         struct in6_rtmsg rtmsg;
1876
1877         if (inet6_rtm_to_rtmsg(r, arg, &rtmsg))
1878                 return -EINVAL;
1879         return ip6_route_add(&rtmsg, nlh, arg, &NETLINK_CB(skb),
1880                              rtm_get_table(arg, r->rtm_table));
1881 }
1882
1883 static int rt6_fill_node(struct sk_buff *skb, struct rt6_info *rt,
1884                          struct in6_addr *dst, struct in6_addr *src,
1885                          int iif, int type, u32 pid, u32 seq,
1886                          int prefix, unsigned int flags)
1887 {
1888         struct rtmsg *rtm;
1889         struct nlmsghdr  *nlh;
1890         unsigned char    *b = skb->tail;
1891         struct rta_cacheinfo ci;
1892         u32 table;
1893
1894         if (prefix) {   /* user wants prefix routes only */
1895                 if (!(rt->rt6i_flags & RTF_PREFIX_RT)) {
1896                         /* success since this is not a prefix route */
1897                         return 1;
1898                 }
1899         }
1900
1901         nlh = NLMSG_NEW(skb, pid, seq, type, sizeof(*rtm), flags);
1902         rtm = NLMSG_DATA(nlh);
1903         rtm->rtm_family = AF_INET6;
1904         rtm->rtm_dst_len = rt->rt6i_dst.plen;
1905         rtm->rtm_src_len = rt->rt6i_src.plen;
1906         rtm->rtm_tos = 0;
1907         if (rt->rt6i_table)
1908                 table = rt->rt6i_table->tb6_id;
1909         else
1910                 table = RT6_TABLE_UNSPEC;
1911         rtm->rtm_table = table;
1912         RTA_PUT_U32(skb, RTA_TABLE, table);
1913         if (rt->rt6i_flags&RTF_REJECT)
1914                 rtm->rtm_type = RTN_UNREACHABLE;
1915         else if (rt->rt6i_dev && (rt->rt6i_dev->flags&IFF_LOOPBACK))
1916                 rtm->rtm_type = RTN_LOCAL;
1917         else
1918                 rtm->rtm_type = RTN_UNICAST;
1919         rtm->rtm_flags = 0;
1920         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
1921         rtm->rtm_protocol = rt->rt6i_protocol;
1922         if (rt->rt6i_flags&RTF_DYNAMIC)
1923                 rtm->rtm_protocol = RTPROT_REDIRECT;
1924         else if (rt->rt6i_flags & RTF_ADDRCONF)
1925                 rtm->rtm_protocol = RTPROT_KERNEL;
1926         else if (rt->rt6i_flags&RTF_DEFAULT)
1927                 rtm->rtm_protocol = RTPROT_RA;
1928
1929         if (rt->rt6i_flags&RTF_CACHE)
1930                 rtm->rtm_flags |= RTM_F_CLONED;
1931
1932         if (dst) {
1933                 RTA_PUT(skb, RTA_DST, 16, dst);
1934                 rtm->rtm_dst_len = 128;
1935         } else if (rtm->rtm_dst_len)
1936                 RTA_PUT(skb, RTA_DST, 16, &rt->rt6i_dst.addr);
1937 #ifdef CONFIG_IPV6_SUBTREES
1938         if (src) {
1939                 RTA_PUT(skb, RTA_SRC, 16, src);
1940                 rtm->rtm_src_len = 128;
1941         } else if (rtm->rtm_src_len)
1942                 RTA_PUT(skb, RTA_SRC, 16, &rt->rt6i_src.addr);
1943 #endif
1944         if (iif)
1945                 RTA_PUT(skb, RTA_IIF, 4, &iif);
1946         else if (dst) {
1947                 struct in6_addr saddr_buf;
1948                 if (ipv6_get_saddr(&rt->u.dst, dst, &saddr_buf) == 0)
1949                         RTA_PUT(skb, RTA_PREFSRC, 16, &saddr_buf);
1950         }
1951         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1952                 goto rtattr_failure;
1953         if (rt->u.dst.neighbour)
1954                 RTA_PUT(skb, RTA_GATEWAY, 16, &rt->u.dst.neighbour->primary_key);
1955         if (rt->u.dst.dev)
1956                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->rt6i_dev->ifindex);
1957         RTA_PUT(skb, RTA_PRIORITY, 4, &rt->rt6i_metric);
1958         ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1959         if (rt->rt6i_expires)
1960                 ci.rta_expires = jiffies_to_clock_t(rt->rt6i_expires - jiffies);
1961         else
1962                 ci.rta_expires = 0;
1963         ci.rta_used = rt->u.dst.__use;
1964         ci.rta_clntref = atomic_read(&rt->u.dst.__refcnt);
1965         ci.rta_error = rt->u.dst.error;
1966         ci.rta_id = 0;
1967         ci.rta_ts = 0;
1968         ci.rta_tsage = 0;
1969         RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1970         nlh->nlmsg_len = skb->tail - b;
1971         return skb->len;
1972
1973 nlmsg_failure:
1974 rtattr_failure:
1975         skb_trim(skb, b - skb->data);
1976         return -1;
1977 }
1978
1979 int rt6_dump_route(struct rt6_info *rt, void *p_arg)
1980 {
1981         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
1982         int prefix;
1983
1984         if (arg->cb->nlh->nlmsg_len >= NLMSG_LENGTH(sizeof(struct rtmsg))) {
1985                 struct rtmsg *rtm = NLMSG_DATA(arg->cb->nlh);
1986                 prefix = (rtm->rtm_flags & RTM_F_PREFIX) != 0;
1987         } else
1988                 prefix = 0;
1989
1990         return rt6_fill_node(arg->skb, rt, NULL, NULL, 0, RTM_NEWROUTE,
1991                      NETLINK_CB(arg->cb->skb).pid, arg->cb->nlh->nlmsg_seq,
1992                      prefix, NLM_F_MULTI);
1993 }
1994
1995 int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
1996 {
1997         struct rtattr **rta = arg;
1998         int iif = 0;
1999         int err = -ENOBUFS;
2000         struct sk_buff *skb;
2001         struct flowi fl;
2002         struct rt6_info *rt;
2003
2004         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2005         if (skb == NULL)
2006                 goto out;
2007
2008         /* Reserve room for dummy headers, this skb can pass
2009            through good chunk of routing engine.
2010          */
2011         skb->mac.raw = skb->data;
2012         skb_reserve(skb, MAX_HEADER + sizeof(struct ipv6hdr));
2013
2014         memset(&fl, 0, sizeof(fl));
2015         if (rta[RTA_SRC-1])
2016                 ipv6_addr_copy(&fl.fl6_src,
2017                                (struct in6_addr*)RTA_DATA(rta[RTA_SRC-1]));
2018         if (rta[RTA_DST-1])
2019                 ipv6_addr_copy(&fl.fl6_dst,
2020                                (struct in6_addr*)RTA_DATA(rta[RTA_DST-1]));
2021
2022         if (rta[RTA_IIF-1])
2023                 memcpy(&iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
2024
2025         if (iif) {
2026                 struct net_device *dev;
2027                 dev = __dev_get_by_index(iif);
2028                 if (!dev) {
2029                         err = -ENODEV;
2030                         goto out_free;
2031                 }
2032         }
2033
2034         fl.oif = 0;
2035         if (rta[RTA_OIF-1])
2036                 memcpy(&fl.oif, RTA_DATA(rta[RTA_OIF-1]), sizeof(int));
2037
2038         rt = (struct rt6_info*)ip6_route_output(NULL, &fl);
2039
2040         skb->dst = &rt->u.dst;
2041
2042         NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
2043         err = rt6_fill_node(skb, rt,
2044                             &fl.fl6_dst, &fl.fl6_src,
2045                             iif,
2046                             RTM_NEWROUTE, NETLINK_CB(in_skb).pid,
2047                             nlh->nlmsg_seq, 0, 0);
2048         if (err < 0) {
2049                 err = -EMSGSIZE;
2050                 goto out_free;
2051         }
2052
2053         err = rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
2054 out:
2055         return err;
2056 out_free:
2057         kfree_skb(skb);
2058         goto out;
2059 }
2060
2061 void inet6_rt_notify(int event, struct rt6_info *rt, struct nlmsghdr *nlh,
2062                         struct netlink_skb_parms *req)
2063 {
2064         struct sk_buff *skb;
2065         u32 pid = req ? req->pid : 0;
2066         u32 seq = nlh ? nlh->nlmsg_seq : 0;
2067         int payload = sizeof(struct rtmsg) + 256;
2068         int err = -ENOBUFS;
2069
2070         skb = nlmsg_new(nlmsg_total_size(payload), gfp_any());
2071         if (skb == NULL)
2072                 goto errout;
2073
2074         err = rt6_fill_node(skb, rt, NULL, NULL, 0, event, pid, seq, 0, 0);
2075         if (err < 0) {
2076                 kfree_skb(skb);
2077                 goto errout;
2078         }
2079
2080         err = rtnl_notify(skb, pid, RTNLGRP_IPV6_ROUTE, nlh, gfp_any());
2081 errout:
2082         if (err < 0)
2083                 rtnl_set_sk_err(RTNLGRP_IPV6_ROUTE, err);
2084 }
2085
2086 /*
2087  *      /proc
2088  */
2089
2090 #ifdef CONFIG_PROC_FS
2091
2092 #define RT6_INFO_LEN (32 + 4 + 32 + 4 + 32 + 40 + 5 + 1)
2093
2094 struct rt6_proc_arg
2095 {
2096         char *buffer;
2097         int offset;
2098         int length;
2099         int skip;
2100         int len;
2101 };
2102
2103 static int rt6_info_route(struct rt6_info *rt, void *p_arg)
2104 {
2105         struct rt6_proc_arg *arg = (struct rt6_proc_arg *) p_arg;
2106         int i;
2107
2108         if (arg->skip < arg->offset / RT6_INFO_LEN) {
2109                 arg->skip++;
2110                 return 0;
2111         }
2112
2113         if (arg->len >= arg->length)
2114                 return 0;
2115
2116         for (i=0; i<16; i++) {
2117                 sprintf(arg->buffer + arg->len, "%02x",
2118                         rt->rt6i_dst.addr.s6_addr[i]);
2119                 arg->len += 2;
2120         }
2121         arg->len += sprintf(arg->buffer + arg->len, " %02x ",
2122                             rt->rt6i_dst.plen);
2123
2124 #ifdef CONFIG_IPV6_SUBTREES
2125         for (i=0; i<16; i++) {
2126                 sprintf(arg->buffer + arg->len, "%02x",
2127                         rt->rt6i_src.addr.s6_addr[i]);
2128                 arg->len += 2;
2129         }
2130         arg->len += sprintf(arg->buffer + arg->len, " %02x ",
2131                             rt->rt6i_src.plen);
2132 #else
2133         sprintf(arg->buffer + arg->len,
2134                 "00000000000000000000000000000000 00 ");
2135         arg->len += 36;
2136 #endif
2137
2138         if (rt->rt6i_nexthop) {
2139                 for (i=0; i<16; i++) {
2140                         sprintf(arg->buffer + arg->len, "%02x",
2141                                 rt->rt6i_nexthop->primary_key[i]);
2142                         arg->len += 2;
2143                 }
2144         } else {
2145                 sprintf(arg->buffer + arg->len,
2146                         "00000000000000000000000000000000");
2147                 arg->len += 32;
2148         }
2149         arg->len += sprintf(arg->buffer + arg->len,
2150                             " %08x %08x %08x %08x %8s\n",
2151                             rt->rt6i_metric, atomic_read(&rt->u.dst.__refcnt),
2152                             rt->u.dst.__use, rt->rt6i_flags,
2153                             rt->rt6i_dev ? rt->rt6i_dev->name : "");
2154         return 0;
2155 }
2156
2157 static int rt6_proc_info(char *buffer, char **start, off_t offset, int length)
2158 {
2159         struct rt6_proc_arg arg = {
2160                 .buffer = buffer,
2161                 .offset = offset,
2162                 .length = length,
2163         };
2164
2165         fib6_clean_all(rt6_info_route, 0, &arg);
2166
2167         *start = buffer;
2168         if (offset)
2169                 *start += offset % RT6_INFO_LEN;
2170
2171         arg.len -= offset % RT6_INFO_LEN;
2172
2173         if (arg.len > length)
2174                 arg.len = length;
2175         if (arg.len < 0)
2176                 arg.len = 0;
2177
2178         return arg.len;
2179 }
2180
2181 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
2182 {
2183         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
2184                       rt6_stats.fib_nodes, rt6_stats.fib_route_nodes,
2185                       rt6_stats.fib_rt_alloc, rt6_stats.fib_rt_entries,
2186                       rt6_stats.fib_rt_cache,
2187                       atomic_read(&ip6_dst_ops.entries),
2188                       rt6_stats.fib_discarded_routes);
2189
2190         return 0;
2191 }
2192
2193 static int rt6_stats_seq_open(struct inode *inode, struct file *file)
2194 {
2195         return single_open(file, rt6_stats_seq_show, NULL);
2196 }
2197
2198 static struct file_operations rt6_stats_seq_fops = {
2199         .owner   = THIS_MODULE,
2200         .open    = rt6_stats_seq_open,
2201         .read    = seq_read,
2202         .llseek  = seq_lseek,
2203         .release = single_release,
2204 };
2205 #endif  /* CONFIG_PROC_FS */
2206
2207 #ifdef CONFIG_SYSCTL
2208
2209 static int flush_delay;
2210
2211 static
2212 int ipv6_sysctl_rtcache_flush(ctl_table *ctl, int write, struct file * filp,
2213                               void __user *buffer, size_t *lenp, loff_t *ppos)
2214 {
2215         if (write) {
2216                 proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
2217                 fib6_run_gc(flush_delay <= 0 ? ~0UL : (unsigned long)flush_delay);
2218                 return 0;
2219         } else
2220                 return -EINVAL;
2221 }
2222
2223 ctl_table ipv6_route_table[] = {
2224         {
2225                 .ctl_name       =       NET_IPV6_ROUTE_FLUSH,
2226                 .procname       =       "flush",
2227                 .data           =       &flush_delay,
2228                 .maxlen         =       sizeof(int),
2229                 .mode           =       0200,
2230                 .proc_handler   =       &ipv6_sysctl_rtcache_flush
2231         },
2232         {
2233                 .ctl_name       =       NET_IPV6_ROUTE_GC_THRESH,
2234                 .procname       =       "gc_thresh",
2235                 .data           =       &ip6_dst_ops.gc_thresh,
2236                 .maxlen         =       sizeof(int),
2237                 .mode           =       0644,
2238                 .proc_handler   =       &proc_dointvec,
2239         },
2240         {
2241                 .ctl_name       =       NET_IPV6_ROUTE_MAX_SIZE,
2242                 .procname       =       "max_size",
2243                 .data           =       &ip6_rt_max_size,
2244                 .maxlen         =       sizeof(int),
2245                 .mode           =       0644,
2246                 .proc_handler   =       &proc_dointvec,
2247         },
2248         {
2249                 .ctl_name       =       NET_IPV6_ROUTE_GC_MIN_INTERVAL,
2250                 .procname       =       "gc_min_interval",
2251                 .data           =       &ip6_rt_gc_min_interval,
2252                 .maxlen         =       sizeof(int),
2253                 .mode           =       0644,
2254                 .proc_handler   =       &proc_dointvec_jiffies,
2255                 .strategy       =       &sysctl_jiffies,
2256         },
2257         {
2258                 .ctl_name       =       NET_IPV6_ROUTE_GC_TIMEOUT,
2259                 .procname       =       "gc_timeout",
2260                 .data           =       &ip6_rt_gc_timeout,
2261                 .maxlen         =       sizeof(int),
2262                 .mode           =       0644,
2263                 .proc_handler   =       &proc_dointvec_jiffies,
2264                 .strategy       =       &sysctl_jiffies,
2265         },
2266         {
2267                 .ctl_name       =       NET_IPV6_ROUTE_GC_INTERVAL,
2268                 .procname       =       "gc_interval",
2269                 .data           =       &ip6_rt_gc_interval,
2270                 .maxlen         =       sizeof(int),
2271                 .mode           =       0644,
2272                 .proc_handler   =       &proc_dointvec_jiffies,
2273                 .strategy       =       &sysctl_jiffies,
2274         },
2275         {
2276                 .ctl_name       =       NET_IPV6_ROUTE_GC_ELASTICITY,
2277                 .procname       =       "gc_elasticity",
2278                 .data           =       &ip6_rt_gc_elasticity,
2279                 .maxlen         =       sizeof(int),
2280                 .mode           =       0644,
2281                 .proc_handler   =       &proc_dointvec_jiffies,
2282                 .strategy       =       &sysctl_jiffies,
2283         },
2284         {
2285                 .ctl_name       =       NET_IPV6_ROUTE_MTU_EXPIRES,
2286                 .procname       =       "mtu_expires",
2287                 .data           =       &ip6_rt_mtu_expires,
2288                 .maxlen         =       sizeof(int),
2289                 .mode           =       0644,
2290                 .proc_handler   =       &proc_dointvec_jiffies,
2291                 .strategy       =       &sysctl_jiffies,
2292         },
2293         {
2294                 .ctl_name       =       NET_IPV6_ROUTE_MIN_ADVMSS,
2295                 .procname       =       "min_adv_mss",
2296                 .data           =       &ip6_rt_min_advmss,
2297                 .maxlen         =       sizeof(int),
2298                 .mode           =       0644,
2299                 .proc_handler   =       &proc_dointvec_jiffies,
2300                 .strategy       =       &sysctl_jiffies,
2301         },
2302         {
2303                 .ctl_name       =       NET_IPV6_ROUTE_GC_MIN_INTERVAL_MS,
2304                 .procname       =       "gc_min_interval_ms",
2305                 .data           =       &ip6_rt_gc_min_interval,
2306                 .maxlen         =       sizeof(int),
2307                 .mode           =       0644,
2308                 .proc_handler   =       &proc_dointvec_ms_jiffies,
2309                 .strategy       =       &sysctl_ms_jiffies,
2310         },
2311         { .ctl_name = 0 }
2312 };
2313
2314 #endif
2315
2316 void __init ip6_route_init(void)
2317 {
2318         struct proc_dir_entry *p;
2319
2320         ip6_dst_ops.kmem_cachep = kmem_cache_create("ip6_dst_cache",
2321                                                      sizeof(struct rt6_info),
2322                                                      0, SLAB_HWCACHE_ALIGN,
2323                                                      NULL, NULL);
2324         if (!ip6_dst_ops.kmem_cachep)
2325                 panic("cannot create ip6_dst_cache");
2326
2327         fib6_init();
2328 #ifdef  CONFIG_PROC_FS
2329         p = proc_net_create("ipv6_route", 0, rt6_proc_info);
2330         if (p)
2331                 p->owner = THIS_MODULE;
2332
2333         proc_net_fops_create("rt6_stats", S_IRUGO, &rt6_stats_seq_fops);
2334 #endif
2335 #ifdef CONFIG_XFRM
2336         xfrm6_init();
2337 #endif
2338 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2339         fib6_rules_init();
2340 #endif
2341 }
2342
2343 void ip6_route_cleanup(void)
2344 {
2345 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2346         fib6_rules_cleanup();
2347 #endif
2348 #ifdef CONFIG_PROC_FS
2349         proc_net_remove("ipv6_route");
2350         proc_net_remove("rt6_stats");
2351 #endif
2352 #ifdef CONFIG_XFRM
2353         xfrm6_fini();
2354 #endif
2355         rt6_ifdown(NULL);
2356         fib6_gc_cleanup();
2357         kmem_cache_destroy(ip6_dst_ops.kmem_cachep);
2358 }