net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <net/net_namespace.h>
  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/dst_metadata.h>
  55 #include <net/xfrm.h>
  56 #include <net/netevent.h>
  57 #include <net/netlink.h>
  58 #include <net/rtnh.h>
  59 #include <net/lwtunnel.h>
  60 #include <net/ip_tunnels.h>
  61 #include <net/l3mdev.h>
  62 #include <net/ip.h>
  63 #include <linux/uaccess.h>
  64 #include <linux/btf_ids.h>
  65
  66 #ifdef CONFIG_SYSCTL
  67 #include <linux/sysctl.h>
  68 #endif
  69
  70 static int ip6_rt_type_to_error(u8 fib6_type);
  71
  72 #define CREATE_TRACE_POINTS
  73 #include <trace/events/fib6.h>
  74 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  75 #undef CREATE_TRACE_POINTS
  76
  77 enum rt6_nud_state {
  78         RT6_NUD_FAIL_HARD = -3,
  79         RT6_NUD_FAIL_PROBE = -2,
  80         RT6_NUD_FAIL_DO_RR = -1,
  81         RT6_NUD_SUCCEED = 1
  82 };
  83
  84 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  85 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  86 static unsigned int      ip6_mtu(const struct dst_entry *dst);
  87 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  88 static void             ip6_dst_destroy(struct dst_entry *);
  89 static void             ip6_dst_ifdown(struct dst_entry *,
  90                                        struct net_device *dev, int how);
  91 static int               ip6_dst_gc(struct dst_ops *ops);
  92
  93 static int              ip6_pkt_discard(struct sk_buff *skb);
  94 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  95 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  96 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  97 static void             ip6_link_failure(struct sk_buff *skb);
  98 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
  99                                            struct sk_buff *skb, u32 mtu,
 100                                            bool confirm_neigh);
 101 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 102                                         struct sk_buff *skb);
 103 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 104                            int strict);
 105 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 106 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 107                          struct fib6_info *rt, struct dst_entry *dst,
 108                          struct in6_addr *dest, struct in6_addr *src,
 109                          int iif, int type, u32 portid, u32 seq,
 110                          unsigned int flags);
 111 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 112                                            const struct in6_addr *daddr,
 113                                            const struct in6_addr *saddr);
 114
 115 #ifdef CONFIG_IPV6_ROUTE_INFO
 116 static struct fib6_info *rt6_add_route_info(struct net *net,
 117                                            const struct in6_addr *prefix, int prefixlen,
 118                                            const struct in6_addr *gwaddr,
 119                                            struct net_device *dev,
 120                                            unsigned int pref);
 121 static struct fib6_info *rt6_get_route_info(struct net *net,
 122                                            const struct in6_addr *prefix, int prefixlen,
 123                                            const struct in6_addr *gwaddr,
 124                                            struct net_device *dev);
 125 #endif
 126
 127 struct uncached_list {
 128         spinlock_t              lock;
 129         struct list_head        head;
 130 };
 131
 132 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 133
 134 void rt6_uncached_list_add(struct rt6_info *rt)
 135 {
 136         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 137
 138         rt->rt6i_uncached_list = ul;
 139
 140         spin_lock_bh(&ul->lock);
 141         list_add_tail(&rt->rt6i_uncached, &ul->head);
 142         spin_unlock_bh(&ul->lock);
 143 }
 144
 145 void rt6_uncached_list_del(struct rt6_info *rt)
 146 {
 147         if (!list_empty(&rt->rt6i_uncached)) {
 148                 struct uncached_list *ul = rt->rt6i_uncached_list;
 149                 struct net *net = dev_net(rt->dst.dev);
 150
 151                 spin_lock_bh(&ul->lock);
 152                 list_del(&rt->rt6i_uncached);
 153                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 154                 spin_unlock_bh(&ul->lock);
 155         }
 156 }
 157
 158 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 159 {
 160         struct net_device *loopback_dev = net->loopback_dev;
 161         int cpu;
 162
 163         if (dev == loopback_dev)
 164                 return;
 165
 166         for_each_possible_cpu(cpu) {
 167                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 168                 struct rt6_info *rt;
 169
 170                 spin_lock_bh(&ul->lock);
 171                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 172                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 173                         struct net_device *rt_dev = rt->dst.dev;
 174
 175                         if (rt_idev->dev == dev) {
 176                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 177                                 in6_dev_put(rt_idev);
 178                         }
 179
 180                         if (rt_dev == dev) {
 181                                 rt->dst.dev = blackhole_netdev;
 182                                 dev_hold(rt->dst.dev);
 183                                 dev_put(rt_dev);
 184                         }
 185                 }
 186                 spin_unlock_bh(&ul->lock);
 187         }
 188 }
 189
 190 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 191                                              struct sk_buff *skb,
 192                                              const void *daddr)
 193 {
 194         if (!ipv6_addr_any(p))
 195                 return (const void *) p;
 196         else if (skb)
 197                 return &ipv6_hdr(skb)->daddr;
 198         return daddr;
 199 }
 200
 201 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 202                                    struct net_device *dev,
 203                                    struct sk_buff *skb,
 204                                    const void *daddr)
 205 {
 206         struct neighbour *n;
 207
 208         daddr = choose_neigh_daddr(gw, skb, daddr);
 209         n = __ipv6_neigh_lookup(dev, daddr);
 210         if (n)
 211                 return n;
 212
 213         n = neigh_create(&nd_tbl, daddr, dev);
 214         return IS_ERR(n) ? NULL : n;
 215 }
 216
 217 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 218                                               struct sk_buff *skb,
 219                                               const void *daddr)
 220 {
 221         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 222
 223         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 224                                 dst->dev, skb, daddr);
 225 }
 226
 227 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 228 {
 229         struct net_device *dev = dst->dev;
 230         struct rt6_info *rt = (struct rt6_info *)dst;
 231
 232         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 233         if (!daddr)
 234                 return;
 235         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 236                 return;
 237         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 238                 return;
 239         __ipv6_confirm_neigh(dev, daddr);
 240 }
 241
 242 static struct dst_ops ip6_dst_ops_template = {
 243         .family                 =       AF_INET6,
 244         .gc                     =       ip6_dst_gc,
 245         .gc_thresh              =       1024,
 246         .check                  =       ip6_dst_check,
 247         .default_advmss         =       ip6_default_advmss,
 248         .mtu                    =       ip6_mtu,
 249         .cow_metrics            =       dst_cow_metrics_generic,
 250         .destroy                =       ip6_dst_destroy,
 251         .ifdown                 =       ip6_dst_ifdown,
 252         .negative_advice        =       ip6_negative_advice,
 253         .link_failure           =       ip6_link_failure,
 254         .update_pmtu            =       ip6_rt_update_pmtu,
 255         .redirect               =       rt6_do_redirect,
 256         .local_out              =       __ip6_local_out,
 257         .neigh_lookup           =       ip6_dst_neigh_lookup,
 258         .confirm_neigh          =       ip6_confirm_neigh,
 259 };
 260
 261 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 262 {
 263         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 264
 265         return mtu ? : dst->dev->mtu;
 266 }
 267
 268 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 269                                          struct sk_buff *skb, u32 mtu,
 270                                          bool confirm_neigh)
 271 {
 272 }
 273
 274 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 275                                       struct sk_buff *skb)
 276 {
 277 }
 278
 279 static struct dst_ops ip6_dst_blackhole_ops = {
 280         .family                 =       AF_INET6,
 281         .destroy                =       ip6_dst_destroy,
 282         .check                  =       ip6_dst_check,
 283         .mtu                    =       ip6_blackhole_mtu,
 284         .default_advmss         =       ip6_default_advmss,
 285         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 286         .redirect               =       ip6_rt_blackhole_redirect,
 287         .cow_metrics            =       dst_cow_metrics_generic,
 288         .neigh_lookup           =       ip6_dst_neigh_lookup,
 289 };
 290
 291 static const u32 ip6_template_metrics[RTAX_MAX] = {
 292         [RTAX_HOPLIMIT - 1] = 0,
 293 };
 294
 295 static const struct fib6_info fib6_null_entry_template = {
 296         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 297         .fib6_protocol  = RTPROT_KERNEL,
 298         .fib6_metric    = ~(u32)0,
 299         .fib6_ref       = REFCOUNT_INIT(1),
 300         .fib6_type      = RTN_UNREACHABLE,
 301         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 302 };
 303
 304 static const struct rt6_info ip6_null_entry_template = {
 305         .dst = {
 306                 .__refcnt       = ATOMIC_INIT(1),
 307                 .__use          = 1,
 308                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 309                 .error          = -ENETUNREACH,
 310                 .input          = ip6_pkt_discard,
 311                 .output         = ip6_pkt_discard_out,
 312         },
 313         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 314 };
 315
 316 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 317
 318 static const struct rt6_info ip6_prohibit_entry_template = {
 319         .dst = {
 320                 .__refcnt       = ATOMIC_INIT(1),
 321                 .__use          = 1,
 322                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 323                 .error          = -EACCES,
 324                 .input          = ip6_pkt_prohibit,
 325                 .output         = ip6_pkt_prohibit_out,
 326         },
 327         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 328 };
 329
 330 static const struct rt6_info ip6_blk_hole_entry_template = {
 331         .dst = {
 332                 .__refcnt       = ATOMIC_INIT(1),
 333                 .__use          = 1,
 334                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 335                 .error          = -EINVAL,
 336                 .input          = dst_discard,
 337                 .output         = dst_discard_out,
 338         },
 339         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 340 };
 341
 342 #endif
 343
 344 static void rt6_info_init(struct rt6_info *rt)
 345 {
 346         struct dst_entry *dst = &rt->dst;
 347
 348         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 349         INIT_LIST_HEAD(&rt->rt6i_uncached);
 350 }
 351
 352 /* allocate dst with ip6_dst_ops */
 353 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 354                                int flags)
 355 {
 356         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 357                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 358
 359         if (rt) {
 360                 rt6_info_init(rt);
 361                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 362         }
 363
 364         return rt;
 365 }
 366 EXPORT_SYMBOL(ip6_dst_alloc);
 367
 368 static void ip6_dst_destroy(struct dst_entry *dst)
 369 {
 370         struct rt6_info *rt = (struct rt6_info *)dst;
 371         struct fib6_info *from;
 372         struct inet6_dev *idev;
 373
 374         ip_dst_metrics_put(dst);
 375         rt6_uncached_list_del(rt);
 376
 377         idev = rt->rt6i_idev;
 378         if (idev) {
 379                 rt->rt6i_idev = NULL;
 380                 in6_dev_put(idev);
 381         }
 382
 383         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 384         fib6_info_release(from);
 385 }
 386
 387 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 388                            int how)
 389 {
 390         struct rt6_info *rt = (struct rt6_info *)dst;
 391         struct inet6_dev *idev = rt->rt6i_idev;
 392         struct net_device *loopback_dev =
 393                 dev_net(dev)->loopback_dev;
 394
 395         if (idev && idev->dev != loopback_dev) {
 396                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 397                 if (loopback_idev) {
 398                         rt->rt6i_idev = loopback_idev;
 399                         in6_dev_put(idev);
 400                 }
 401         }
 402 }
 403
 404 static bool __rt6_check_expired(const struct rt6_info *rt)
 405 {
 406         if (rt->rt6i_flags & RTF_EXPIRES)
 407                 return time_after(jiffies, rt->dst.expires);
 408         else
 409                 return false;
 410 }
 411
 412 static bool rt6_check_expired(const struct rt6_info *rt)
 413 {
 414         struct fib6_info *from;
 415
 416         from = rcu_dereference(rt->from);
 417
 418         if (rt->rt6i_flags & RTF_EXPIRES) {
 419                 if (time_after(jiffies, rt->dst.expires))
 420                         return true;
 421         } else if (from) {
 422                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 423                         fib6_check_expired(from);
 424         }
 425         return false;
 426 }
 427
 428 void fib6_select_path(const struct net *net, struct fib6_result *res,
 429                       struct flowi6 *fl6, int oif, bool have_oif_match,
 430                       const struct sk_buff *skb, int strict)
 431 {
 432         struct fib6_info *sibling, *next_sibling;
 433         struct fib6_info *match = res->f6i;
 434
 435         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 436                 goto out;
 437
 438         if (match->nh && have_oif_match && res->nh)
 439                 return;
 440
 441         /* We might have already computed the hash for ICMPv6 errors. In such
 442          * case it will always be non-zero. Otherwise now is the time to do it.
 443          */
 444         if (!fl6->mp_hash &&
 445             (!match->nh || nexthop_is_multipath(match->nh)))
 446                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 447
 448         if (unlikely(match->nh)) {
 449                 nexthop_path_fib6_result(res, fl6->mp_hash);
 450                 return;
 451         }
 452
 453         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 454                 goto out;
 455
 456         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 457                                  fib6_siblings) {
 458                 const struct fib6_nh *nh = sibling->fib6_nh;
 459                 int nh_upper_bound;
 460
 461                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 462                 if (fl6->mp_hash > nh_upper_bound)
 463                         continue;
 464                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 465                         break;
 466                 match = sibling;
 467                 break;
 468         }
 469
 470 out:
 471         res->f6i = match;
 472         res->nh = match->fib6_nh;
 473 }
 474
 475 /*
 476  *      Route lookup. rcu_read_lock() should be held.
 477  */
 478
 479 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 480                                const struct in6_addr *saddr, int oif, int flags)
 481 {
 482         const struct net_device *dev;
 483
 484         if (nh->fib_nh_flags & RTNH_F_DEAD)
 485                 return false;
 486
 487         dev = nh->fib_nh_dev;
 488         if (oif) {
 489                 if (dev->ifindex == oif)
 490                         return true;
 491         } else {
 492                 if (ipv6_chk_addr(net, saddr, dev,
 493                                   flags & RT6_LOOKUP_F_IFACE))
 494                         return true;
 495         }
 496
 497         return false;
 498 }
 499
 500 struct fib6_nh_dm_arg {
 501         struct net              *net;
 502         const struct in6_addr   *saddr;
 503         int                     oif;
 504         int                     flags;
 505         struct fib6_nh          *nh;
 506 };
 507
 508 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 509 {
 510         struct fib6_nh_dm_arg *arg = _arg;
 511
 512         arg->nh = nh;
 513         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 514                                   arg->flags);
 515 }
 516
 517 /* returns fib6_nh from nexthop or NULL */
 518 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 519                                         struct fib6_result *res,
 520                                         const struct in6_addr *saddr,
 521                                         int oif, int flags)
 522 {
 523         struct fib6_nh_dm_arg arg = {
 524                 .net   = net,
 525                 .saddr = saddr,
 526                 .oif   = oif,
 527                 .flags = flags,
 528         };
 529
 530         if (nexthop_is_blackhole(nh))
 531                 return NULL;
 532
 533         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 534                 return arg.nh;
 535
 536         return NULL;
 537 }
 538
 539 static void rt6_device_match(struct net *net, struct fib6_result *res,
 540                              const struct in6_addr *saddr, int oif, int flags)
 541 {
 542         struct fib6_info *f6i = res->f6i;
 543         struct fib6_info *spf6i;
 544         struct fib6_nh *nh;
 545
 546         if (!oif && ipv6_addr_any(saddr)) {
 547                 if (unlikely(f6i->nh)) {
 548                         nh = nexthop_fib6_nh(f6i->nh);
 549                         if (nexthop_is_blackhole(f6i->nh))
 550                                 goto out_blackhole;
 551                 } else {
 552                         nh = f6i->fib6_nh;
 553                 }
 554                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 555                         goto out;
 556         }
 557
 558         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 559                 bool matched = false;
 560
 561                 if (unlikely(spf6i->nh)) {
 562                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 563                                               oif, flags);
 564                         if (nh)
 565                                 matched = true;
 566                 } else {
 567                         nh = spf6i->fib6_nh;
 568                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 569                                 matched = true;
 570                 }
 571                 if (matched) {
 572                         res->f6i = spf6i;
 573                         goto out;
 574                 }
 575         }
 576
 577         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 578                 res->f6i = net->ipv6.fib6_null_entry;
 579                 nh = res->f6i->fib6_nh;
 580                 goto out;
 581         }
 582
 583         if (unlikely(f6i->nh)) {
 584                 nh = nexthop_fib6_nh(f6i->nh);
 585                 if (nexthop_is_blackhole(f6i->nh))
 586                         goto out_blackhole;
 587         } else {
 588                 nh = f6i->fib6_nh;
 589         }
 590
 591         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 592                 res->f6i = net->ipv6.fib6_null_entry;
 593                 nh = res->f6i->fib6_nh;
 594         }
 595 out:
 596         res->nh = nh;
 597         res->fib6_type = res->f6i->fib6_type;
 598         res->fib6_flags = res->f6i->fib6_flags;
 599         return;
 600
 601 out_blackhole:
 602         res->fib6_flags |= RTF_REJECT;
 603         res->fib6_type = RTN_BLACKHOLE;
 604         res->nh = nh;
 605 }
 606
 607 #ifdef CONFIG_IPV6_ROUTER_PREF
 608 struct __rt6_probe_work {
 609         struct work_struct work;
 610         struct in6_addr target;
 611         struct net_device *dev;
 612 };
 613
 614 static void rt6_probe_deferred(struct work_struct *w)
 615 {
 616         struct in6_addr mcaddr;
 617         struct __rt6_probe_work *work =
 618                 container_of(w, struct __rt6_probe_work, work);
 619
 620         addrconf_addr_solict_mult(&work->target, &mcaddr);
 621         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 622         dev_put(work->dev);
 623         kfree(work);
 624 }
 625
 626 static void rt6_probe(struct fib6_nh *fib6_nh)
 627 {
 628         struct __rt6_probe_work *work = NULL;
 629         const struct in6_addr *nh_gw;
 630         unsigned long last_probe;
 631         struct neighbour *neigh;
 632         struct net_device *dev;
 633         struct inet6_dev *idev;
 634
 635         /*
 636          * Okay, this does not seem to be appropriate
 637          * for now, however, we need to check if it
 638          * is really so; aka Router Reachability Probing.
 639          *
 640          * Router Reachability Probe MUST be rate-limited
 641          * to no more than one per minute.
 642          */
 643         if (!fib6_nh->fib_nh_gw_family)
 644                 return;
 645
 646         nh_gw = &fib6_nh->fib_nh_gw6;
 647         dev = fib6_nh->fib_nh_dev;
 648         rcu_read_lock_bh();
 649         last_probe = READ_ONCE(fib6_nh->last_probe);
 650         idev = __in6_dev_get(dev);
 651         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 652         if (neigh) {
 653                 if (neigh->nud_state & NUD_VALID)
 654                         goto out;
 655
 656                 write_lock(&neigh->lock);
 657                 if (!(neigh->nud_state & NUD_VALID) &&
 658                     time_after(jiffies,
 659                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 660                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 661                         if (work)
 662                                 __neigh_set_probe_once(neigh);
 663                 }
 664                 write_unlock(&neigh->lock);
 665         } else if (time_after(jiffies, last_probe +
 666                                        idev->cnf.rtr_probe_interval)) {
 667                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 668         }
 669
 670         if (!work || cmpxchg(&fib6_nh->last_probe,
 671                              last_probe, jiffies) != last_probe) {
 672                 kfree(work);
 673         } else {
 674                 INIT_WORK(&work->work, rt6_probe_deferred);
 675                 work->target = *nh_gw;
 676                 dev_hold(dev);
 677                 work->dev = dev;
 678                 schedule_work(&work->work);
 679         }
 680
 681 out:
 682         rcu_read_unlock_bh();
 683 }
 684 #else
 685 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 686 {
 687 }
 688 #endif
 689
 690 /*
 691  * Default Router Selection (RFC 2461 6.3.6)
 692  */
 693 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 694 {
 695         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 696         struct neighbour *neigh;
 697
 698         rcu_read_lock_bh();
 699         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 700                                           &fib6_nh->fib_nh_gw6);
 701         if (neigh) {
 702                 read_lock(&neigh->lock);
 703                 if (neigh->nud_state & NUD_VALID)
 704                         ret = RT6_NUD_SUCCEED;
 705 #ifdef CONFIG_IPV6_ROUTER_PREF
 706                 else if (!(neigh->nud_state & NUD_FAILED))
 707                         ret = RT6_NUD_SUCCEED;
 708                 else
 709                         ret = RT6_NUD_FAIL_PROBE;
 710 #endif
 711                 read_unlock(&neigh->lock);
 712         } else {
 713                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 714                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 715         }
 716         rcu_read_unlock_bh();
 717
 718         return ret;
 719 }
 720
 721 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 722                            int strict)
 723 {
 724         int m = 0;
 725
 726         if (!oif || nh->fib_nh_dev->ifindex == oif)
 727                 m = 2;
 728
 729         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 730                 return RT6_NUD_FAIL_HARD;
 731 #ifdef CONFIG_IPV6_ROUTER_PREF
 732         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 733 #endif
 734         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 735             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 736                 int n = rt6_check_neigh(nh);
 737                 if (n < 0)
 738                         return n;
 739         }
 740         return m;
 741 }
 742
 743 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 744                        int oif, int strict, int *mpri, bool *do_rr)
 745 {
 746         bool match_do_rr = false;
 747         bool rc = false;
 748         int m;
 749
 750         if (nh->fib_nh_flags & RTNH_F_DEAD)
 751                 goto out;
 752
 753         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 754             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 755             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 756                 goto out;
 757
 758         m = rt6_score_route(nh, fib6_flags, oif, strict);
 759         if (m == RT6_NUD_FAIL_DO_RR) {
 760                 match_do_rr = true;
 761                 m = 0; /* lowest valid score */
 762         } else if (m == RT6_NUD_FAIL_HARD) {
 763                 goto out;
 764         }
 765
 766         if (strict & RT6_LOOKUP_F_REACHABLE)
 767                 rt6_probe(nh);
 768
 769         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 770         if (m > *mpri) {
 771                 *do_rr = match_do_rr;
 772                 *mpri = m;
 773                 rc = true;
 774         }
 775 out:
 776         return rc;
 777 }
 778
 779 struct fib6_nh_frl_arg {
 780         u32             flags;
 781         int             oif;
 782         int             strict;
 783         int             *mpri;
 784         bool            *do_rr;
 785         struct fib6_nh  *nh;
 786 };
 787
 788 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 789 {
 790         struct fib6_nh_frl_arg *arg = _arg;
 791
 792         arg->nh = nh;
 793         return find_match(nh, arg->flags, arg->oif, arg->strict,
 794                           arg->mpri, arg->do_rr);
 795 }
 796
 797 static void __find_rr_leaf(struct fib6_info *f6i_start,
 798                            struct fib6_info *nomatch, u32 metric,
 799                            struct fib6_result *res, struct fib6_info **cont,
 800                            int oif, int strict, bool *do_rr, int *mpri)
 801 {
 802         struct fib6_info *f6i;
 803
 804         for (f6i = f6i_start;
 805              f6i && f6i != nomatch;
 806              f6i = rcu_dereference(f6i->fib6_next)) {
 807                 bool matched = false;
 808                 struct fib6_nh *nh;
 809
 810                 if (cont && f6i->fib6_metric != metric) {
 811                         *cont = f6i;
 812                         return;
 813                 }
 814
 815                 if (fib6_check_expired(f6i))
 816                         continue;
 817
 818                 if (unlikely(f6i->nh)) {
 819                         struct fib6_nh_frl_arg arg = {
 820                                 .flags  = f6i->fib6_flags,
 821                                 .oif    = oif,
 822                                 .strict = strict,
 823                                 .mpri   = mpri,
 824                                 .do_rr  = do_rr
 825                         };
 826
 827                         if (nexthop_is_blackhole(f6i->nh)) {
 828                                 res->fib6_flags = RTF_REJECT;
 829                                 res->fib6_type = RTN_BLACKHOLE;
 830                                 res->f6i = f6i;
 831                                 res->nh = nexthop_fib6_nh(f6i->nh);
 832                                 return;
 833                         }
 834                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 835                                                      &arg)) {
 836                                 matched = true;
 837                                 nh = arg.nh;
 838                         }
 839                 } else {
 840                         nh = f6i->fib6_nh;
 841                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 842                                        mpri, do_rr))
 843                                 matched = true;
 844                 }
 845                 if (matched) {
 846                         res->f6i = f6i;
 847                         res->nh = nh;
 848                         res->fib6_flags = f6i->fib6_flags;
 849                         res->fib6_type = f6i->fib6_type;
 850                 }
 851         }
 852 }
 853
 854 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 855                          struct fib6_info *rr_head, int oif, int strict,
 856                          bool *do_rr, struct fib6_result *res)
 857 {
 858         u32 metric = rr_head->fib6_metric;
 859         struct fib6_info *cont = NULL;
 860         int mpri = -1;
 861
 862         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 863                        oif, strict, do_rr, &mpri);
 864
 865         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 866                        oif, strict, do_rr, &mpri);
 867
 868         if (res->f6i || !cont)
 869                 return;
 870
 871         __find_rr_leaf(cont, NULL, metric, res, NULL,
 872                        oif, strict, do_rr, &mpri);
 873 }
 874
 875 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 876                        struct fib6_result *res, int strict)
 877 {
 878         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 879         struct fib6_info *rt0;
 880         bool do_rr = false;
 881         int key_plen;
 882
 883         /* make sure this function or its helpers sets f6i */
 884         res->f6i = NULL;
 885
 886         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 887                 goto out;
 888
 889         rt0 = rcu_dereference(fn->rr_ptr);
 890         if (!rt0)
 891                 rt0 = leaf;
 892
 893         /* Double check to make sure fn is not an intermediate node
 894          * and fn->leaf does not points to its child's leaf
 895          * (This might happen if all routes under fn are deleted from
 896          * the tree and fib6_repair_tree() is called on the node.)
 897          */
 898         key_plen = rt0->fib6_dst.plen;
 899 #ifdef CONFIG_IPV6_SUBTREES
 900         if (rt0->fib6_src.plen)
 901                 key_plen = rt0->fib6_src.plen;
 902 #endif
 903         if (fn->fn_bit != key_plen)
 904                 goto out;
 905
 906         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 907         if (do_rr) {
 908                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 909
 910                 /* no entries matched; do round-robin */
 911                 if (!next || next->fib6_metric != rt0->fib6_metric)
 912                         next = leaf;
 913
 914                 if (next != rt0) {
 915                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 916                         /* make sure next is not being deleted from the tree */
 917                         if (next->fib6_node)
 918                                 rcu_assign_pointer(fn->rr_ptr, next);
 919                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 920                 }
 921         }
 922
 923 out:
 924         if (!res->f6i) {
 925                 res->f6i = net->ipv6.fib6_null_entry;
 926                 res->nh = res->f6i->fib6_nh;
 927                 res->fib6_flags = res->f6i->fib6_flags;
 928                 res->fib6_type = res->f6i->fib6_type;
 929         }
 930 }
 931
 932 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 933 {
 934         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 935                res->nh->fib_nh_gw_family;
 936 }
 937
 938 #ifdef CONFIG_IPV6_ROUTE_INFO
 939 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 940                   const struct in6_addr *gwaddr)
 941 {
 942         struct net *net = dev_net(dev);
 943         struct route_info *rinfo = (struct route_info *) opt;
 944         struct in6_addr prefix_buf, *prefix;
 945         unsigned int pref;
 946         unsigned long lifetime;
 947         struct fib6_info *rt;
 948
 949         if (len < sizeof(struct route_info)) {
 950                 return -EINVAL;
 951         }
 952
 953         /* Sanity check for prefix_len and length */
 954         if (rinfo->length > 3) {
 955                 return -EINVAL;
 956         } else if (rinfo->prefix_len > 128) {
 957                 return -EINVAL;
 958         } else if (rinfo->prefix_len > 64) {
 959                 if (rinfo->length < 2) {
 960                         return -EINVAL;
 961                 }
 962         } else if (rinfo->prefix_len > 0) {
 963                 if (rinfo->length < 1) {
 964                         return -EINVAL;
 965                 }
 966         }
 967
 968         pref = rinfo->route_pref;
 969         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 970                 return -EINVAL;
 971
 972         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 973
 974         if (rinfo->length == 3)
 975                 prefix = (struct in6_addr *)rinfo->prefix;
 976         else {
 977                 /* this function is safe */
 978                 ipv6_addr_prefix(&prefix_buf,
 979                                  (struct in6_addr *)rinfo->prefix,
 980                                  rinfo->prefix_len);
 981                 prefix = &prefix_buf;
 982         }
 983
 984         if (rinfo->prefix_len == 0)
 985                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 986         else
 987                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 988                                         gwaddr, dev);
 989
 990         if (rt && !lifetime) {
 991                 ip6_del_rt(net, rt, false);
 992                 rt = NULL;
 993         }
 994
 995         if (!rt && lifetime)
 996                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 997                                         dev, pref);
 998         else if (rt)
 999                 rt->fib6_flags = RTF_ROUTEINFO |
1000                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
1001
1002         if (rt) {
1003                 if (!addrconf_finite_timeout(lifetime))
1004                         fib6_clean_expires(rt);
1005                 else
1006                         fib6_set_expires(rt, jiffies + HZ * lifetime);
1007
1008                 fib6_info_release(rt);
1009         }
1010         return 0;
1011 }
1012 #endif
1013
1014 /*
1015  *      Misc support functions
1016  */
1017
1018 /* called with rcu_lock held */
1019 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1020 {
1021         struct net_device *dev = res->nh->fib_nh_dev;
1022
1023         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1024                 /* for copies of local routes, dst->dev needs to be the
1025                  * device if it is a master device, the master device if
1026                  * device is enslaved, and the loopback as the default
1027                  */
1028                 if (netif_is_l3_slave(dev) &&
1029                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1030                         dev = l3mdev_master_dev_rcu(dev);
1031                 else if (!netif_is_l3_master(dev))
1032                         dev = dev_net(dev)->loopback_dev;
1033                 /* last case is netif_is_l3_master(dev) is true in which
1034                  * case we want dev returned to be dev
1035                  */
1036         }
1037
1038         return dev;
1039 }
1040
1041 static const int fib6_prop[RTN_MAX + 1] = {
1042         [RTN_UNSPEC]    = 0,
1043         [RTN_UNICAST]   = 0,
1044         [RTN_LOCAL]     = 0,
1045         [RTN_BROADCAST] = 0,
1046         [RTN_ANYCAST]   = 0,
1047         [RTN_MULTICAST] = 0,
1048         [RTN_BLACKHOLE] = -EINVAL,
1049         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1050         [RTN_PROHIBIT]  = -EACCES,
1051         [RTN_THROW]     = -EAGAIN,
1052         [RTN_NAT]       = -EINVAL,
1053         [RTN_XRESOLVE]  = -EINVAL,
1054 };
1055
1056 static int ip6_rt_type_to_error(u8 fib6_type)
1057 {
1058         return fib6_prop[fib6_type];
1059 }
1060
1061 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1062 {
1063         unsigned short flags = 0;
1064
1065         if (rt->dst_nocount)
1066                 flags |= DST_NOCOUNT;
1067         if (rt->dst_nopolicy)
1068                 flags |= DST_NOPOLICY;
1069
1070         return flags;
1071 }
1072
1073 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1074 {
1075         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1076
1077         switch (fib6_type) {
1078         case RTN_BLACKHOLE:
1079                 rt->dst.output = dst_discard_out;
1080                 rt->dst.input = dst_discard;
1081                 break;
1082         case RTN_PROHIBIT:
1083                 rt->dst.output = ip6_pkt_prohibit_out;
1084                 rt->dst.input = ip6_pkt_prohibit;
1085                 break;
1086         case RTN_THROW:
1087         case RTN_UNREACHABLE:
1088         default:
1089                 rt->dst.output = ip6_pkt_discard_out;
1090                 rt->dst.input = ip6_pkt_discard;
1091                 break;
1092         }
1093 }
1094
1095 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1096 {
1097         struct fib6_info *f6i = res->f6i;
1098
1099         if (res->fib6_flags & RTF_REJECT) {
1100                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1101                 return;
1102         }
1103
1104         rt->dst.error = 0;
1105         rt->dst.output = ip6_output;
1106
1107         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1108                 rt->dst.input = ip6_input;
1109         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1110                 rt->dst.input = ip6_mc_input;
1111         } else {
1112                 rt->dst.input = ip6_forward;
1113         }
1114
1115         if (res->nh->fib_nh_lws) {
1116                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1117                 lwtunnel_set_redirect(&rt->dst);
1118         }
1119
1120         rt->dst.lastuse = jiffies;
1121 }
1122
1123 /* Caller must already hold reference to @from */
1124 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1125 {
1126         rt->rt6i_flags &= ~RTF_EXPIRES;
1127         rcu_assign_pointer(rt->from, from);
1128         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1129 }
1130
1131 /* Caller must already hold reference to f6i in result */
1132 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1133 {
1134         const struct fib6_nh *nh = res->nh;
1135         const struct net_device *dev = nh->fib_nh_dev;
1136         struct fib6_info *f6i = res->f6i;
1137
1138         ip6_rt_init_dst(rt, res);
1139
1140         rt->rt6i_dst = f6i->fib6_dst;
1141         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1142         rt->rt6i_flags = res->fib6_flags;
1143         if (nh->fib_nh_gw_family) {
1144                 rt->rt6i_gateway = nh->fib_nh_gw6;
1145                 rt->rt6i_flags |= RTF_GATEWAY;
1146         }
1147         rt6_set_from(rt, f6i);
1148 #ifdef CONFIG_IPV6_SUBTREES
1149         rt->rt6i_src = f6i->fib6_src;
1150 #endif
1151 }
1152
1153 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1154                                         struct in6_addr *saddr)
1155 {
1156         struct fib6_node *pn, *sn;
1157         while (1) {
1158                 if (fn->fn_flags & RTN_TL_ROOT)
1159                         return NULL;
1160                 pn = rcu_dereference(fn->parent);
1161                 sn = FIB6_SUBTREE(pn);
1162                 if (sn && sn != fn)
1163                         fn = fib6_node_lookup(sn, NULL, saddr);
1164                 else
1165                         fn = pn;
1166                 if (fn->fn_flags & RTN_RTINFO)
1167                         return fn;
1168         }
1169 }
1170
1171 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1172 {
1173         struct rt6_info *rt = *prt;
1174
1175         if (dst_hold_safe(&rt->dst))
1176                 return true;
1177         if (net) {
1178                 rt = net->ipv6.ip6_null_entry;
1179                 dst_hold(&rt->dst);
1180         } else {
1181                 rt = NULL;
1182         }
1183         *prt = rt;
1184         return false;
1185 }
1186
1187 /* called with rcu_lock held */
1188 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1189 {
1190         struct net_device *dev = res->nh->fib_nh_dev;
1191         struct fib6_info *f6i = res->f6i;
1192         unsigned short flags;
1193         struct rt6_info *nrt;
1194
1195         if (!fib6_info_hold_safe(f6i))
1196                 goto fallback;
1197
1198         flags = fib6_info_dst_flags(f6i);
1199         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1200         if (!nrt) {
1201                 fib6_info_release(f6i);
1202                 goto fallback;
1203         }
1204
1205         ip6_rt_copy_init(nrt, res);
1206         return nrt;
1207
1208 fallback:
1209         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1210         dst_hold(&nrt->dst);
1211         return nrt;
1212 }
1213
1214 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1215                                              struct fib6_table *table,
1216                                              struct flowi6 *fl6,
1217                                              const struct sk_buff *skb,
1218                                              int flags)
1219 {
1220         struct fib6_result res = {};
1221         struct fib6_node *fn;
1222         struct rt6_info *rt;
1223
1224         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1225                 flags &= ~RT6_LOOKUP_F_IFACE;
1226
1227         rcu_read_lock();
1228         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1229 restart:
1230         res.f6i = rcu_dereference(fn->leaf);
1231         if (!res.f6i)
1232                 res.f6i = net->ipv6.fib6_null_entry;
1233         else
1234                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1235                                  flags);
1236
1237         if (res.f6i == net->ipv6.fib6_null_entry) {
1238                 fn = fib6_backtrack(fn, &fl6->saddr);
1239                 if (fn)
1240                         goto restart;
1241
1242                 rt = net->ipv6.ip6_null_entry;
1243                 dst_hold(&rt->dst);
1244                 goto out;
1245         } else if (res.fib6_flags & RTF_REJECT) {
1246                 goto do_create;
1247         }
1248
1249         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1250                          fl6->flowi6_oif != 0, skb, flags);
1251
1252         /* Search through exception table */
1253         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1254         if (rt) {
1255                 if (ip6_hold_safe(net, &rt))
1256                         dst_use_noref(&rt->dst, jiffies);
1257         } else {
1258 do_create:
1259                 rt = ip6_create_rt_rcu(&res);
1260         }
1261
1262 out:
1263         trace_fib6_table_lookup(net, &res, table, fl6);
1264
1265         rcu_read_unlock();
1266
1267         return rt;
1268 }
1269
1270 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1271                                    const struct sk_buff *skb, int flags)
1272 {
1273         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1274 }
1275 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1276
1277 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1278                             const struct in6_addr *saddr, int oif,
1279                             const struct sk_buff *skb, int strict)
1280 {
1281         struct flowi6 fl6 = {
1282                 .flowi6_oif = oif,
1283                 .daddr = *daddr,
1284         };
1285         struct dst_entry *dst;
1286         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1287
1288         if (saddr) {
1289                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1290                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1291         }
1292
1293         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1294         if (dst->error == 0)
1295                 return (struct rt6_info *) dst;
1296
1297         dst_release(dst);
1298
1299         return NULL;
1300 }
1301 EXPORT_SYMBOL(rt6_lookup);
1302
1303 /* ip6_ins_rt is called with FREE table->tb6_lock.
1304  * It takes new route entry, the addition fails by any reason the
1305  * route is released.
1306  * Caller must hold dst before calling it.
1307  */
1308
1309 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1310                         struct netlink_ext_ack *extack)
1311 {
1312         int err;
1313         struct fib6_table *table;
1314
1315         table = rt->fib6_table;
1316         spin_lock_bh(&table->tb6_lock);
1317         err = fib6_add(&table->tb6_root, rt, info, extack);
1318         spin_unlock_bh(&table->tb6_lock);
1319
1320         return err;
1321 }
1322
1323 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1324 {
1325         struct nl_info info = { .nl_net = net, };
1326
1327         return __ip6_ins_rt(rt, &info, NULL);
1328 }
1329
1330 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1331                                            const struct in6_addr *daddr,
1332                                            const struct in6_addr *saddr)
1333 {
1334         struct fib6_info *f6i = res->f6i;
1335         struct net_device *dev;
1336         struct rt6_info *rt;
1337
1338         /*
1339          *      Clone the route.
1340          */
1341
1342         if (!fib6_info_hold_safe(f6i))
1343                 return NULL;
1344
1345         dev = ip6_rt_get_dev_rcu(res);
1346         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1347         if (!rt) {
1348                 fib6_info_release(f6i);
1349                 return NULL;
1350         }
1351
1352         ip6_rt_copy_init(rt, res);
1353         rt->rt6i_flags |= RTF_CACHE;
1354         rt->rt6i_dst.addr = *daddr;
1355         rt->rt6i_dst.plen = 128;
1356
1357         if (!rt6_is_gw_or_nonexthop(res)) {
1358                 if (f6i->fib6_dst.plen != 128 &&
1359                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1360                         rt->rt6i_flags |= RTF_ANYCAST;
1361 #ifdef CONFIG_IPV6_SUBTREES
1362                 if (rt->rt6i_src.plen && saddr) {
1363                         rt->rt6i_src.addr = *saddr;
1364                         rt->rt6i_src.plen = 128;
1365                 }
1366 #endif
1367         }
1368
1369         return rt;
1370 }
1371
1372 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1373 {
1374         struct fib6_info *f6i = res->f6i;
1375         unsigned short flags = fib6_info_dst_flags(f6i);
1376         struct net_device *dev;
1377         struct rt6_info *pcpu_rt;
1378
1379         if (!fib6_info_hold_safe(f6i))
1380                 return NULL;
1381
1382         rcu_read_lock();
1383         dev = ip6_rt_get_dev_rcu(res);
1384         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1385         rcu_read_unlock();
1386         if (!pcpu_rt) {
1387                 fib6_info_release(f6i);
1388                 return NULL;
1389         }
1390         ip6_rt_copy_init(pcpu_rt, res);
1391         pcpu_rt->rt6i_flags |= RTF_PCPU;
1392
1393         if (f6i->nh)
1394                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1395
1396         return pcpu_rt;
1397 }
1398
1399 static bool rt6_is_valid(const struct rt6_info *rt6)
1400 {
1401         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1402 }
1403
1404 /* It should be called with rcu_read_lock() acquired */
1405 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1406 {
1407         struct rt6_info *pcpu_rt;
1408
1409         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1410
1411         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1412                 struct rt6_info *prev, **p;
1413
1414                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1415                 prev = xchg(p, NULL);
1416                 if (prev) {
1417                         dst_dev_put(&prev->dst);
1418                         dst_release(&prev->dst);
1419                 }
1420
1421                 pcpu_rt = NULL;
1422         }
1423
1424         return pcpu_rt;
1425 }
1426
1427 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1428                                             const struct fib6_result *res)
1429 {
1430         struct rt6_info *pcpu_rt, *prev, **p;
1431
1432         pcpu_rt = ip6_rt_pcpu_alloc(res);
1433         if (!pcpu_rt)
1434                 return NULL;
1435
1436         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1437         prev = cmpxchg(p, NULL, pcpu_rt);
1438         BUG_ON(prev);
1439
1440         if (res->f6i->fib6_destroying) {
1441                 struct fib6_info *from;
1442
1443                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1444                 fib6_info_release(from);
1445         }
1446
1447         return pcpu_rt;
1448 }
1449
1450 /* exception hash table implementation
1451  */
1452 static DEFINE_SPINLOCK(rt6_exception_lock);
1453
1454 /* Remove rt6_ex from hash table and free the memory
1455  * Caller must hold rt6_exception_lock
1456  */
1457 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1458                                  struct rt6_exception *rt6_ex)
1459 {
1460         struct fib6_info *from;
1461         struct net *net;
1462
1463         if (!bucket || !rt6_ex)
1464                 return;
1465
1466         net = dev_net(rt6_ex->rt6i->dst.dev);
1467         net->ipv6.rt6_stats->fib_rt_cache--;
1468
1469         /* purge completely the exception to allow releasing the held resources:
1470          * some [sk] cache may keep the dst around for unlimited time
1471          */
1472         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1473         fib6_info_release(from);
1474         dst_dev_put(&rt6_ex->rt6i->dst);
1475
1476         hlist_del_rcu(&rt6_ex->hlist);
1477         dst_release(&rt6_ex->rt6i->dst);
1478         kfree_rcu(rt6_ex, rcu);
1479         WARN_ON_ONCE(!bucket->depth);
1480         bucket->depth--;
1481 }
1482
1483 /* Remove oldest rt6_ex in bucket and free the memory
1484  * Caller must hold rt6_exception_lock
1485  */
1486 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1487 {
1488         struct rt6_exception *rt6_ex, *oldest = NULL;
1489
1490         if (!bucket)
1491                 return;
1492
1493         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1494                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1495                         oldest = rt6_ex;
1496         }
1497         rt6_remove_exception(bucket, oldest);
1498 }
1499
1500 static u32 rt6_exception_hash(const struct in6_addr *dst,
1501                               const struct in6_addr *src)
1502 {
1503         static u32 seed __read_mostly;
1504         u32 val;
1505
1506         net_get_random_once(&seed, sizeof(seed));
1507         val = jhash2((const u32 *)dst, sizeof(*dst)/sizeof(u32), seed);
1508
1509 #ifdef CONFIG_IPV6_SUBTREES
1510         if (src)
1511                 val = jhash2((const u32 *)src, sizeof(*src)/sizeof(u32), val);
1512 #endif
1513         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1514 }
1515
1516 /* Helper function to find the cached rt in the hash table
1517  * and update bucket pointer to point to the bucket for this
1518  * (daddr, saddr) pair
1519  * Caller must hold rt6_exception_lock
1520  */
1521 static struct rt6_exception *
1522 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1523                               const struct in6_addr *daddr,
1524                               const struct in6_addr *saddr)
1525 {
1526         struct rt6_exception *rt6_ex;
1527         u32 hval;
1528
1529         if (!(*bucket) || !daddr)
1530                 return NULL;
1531
1532         hval = rt6_exception_hash(daddr, saddr);
1533         *bucket += hval;
1534
1535         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1536                 struct rt6_info *rt6 = rt6_ex->rt6i;
1537                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1538
1539 #ifdef CONFIG_IPV6_SUBTREES
1540                 if (matched && saddr)
1541                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1542 #endif
1543                 if (matched)
1544                         return rt6_ex;
1545         }
1546         return NULL;
1547 }
1548
1549 /* Helper function to find the cached rt in the hash table
1550  * and update bucket pointer to point to the bucket for this
1551  * (daddr, saddr) pair
1552  * Caller must hold rcu_read_lock()
1553  */
1554 static struct rt6_exception *
1555 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1556                          const struct in6_addr *daddr,
1557                          const struct in6_addr *saddr)
1558 {
1559         struct rt6_exception *rt6_ex;
1560         u32 hval;
1561
1562         WARN_ON_ONCE(!rcu_read_lock_held());
1563
1564         if (!(*bucket) || !daddr)
1565                 return NULL;
1566
1567         hval = rt6_exception_hash(daddr, saddr);
1568         *bucket += hval;
1569
1570         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1571                 struct rt6_info *rt6 = rt6_ex->rt6i;
1572                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1573
1574 #ifdef CONFIG_IPV6_SUBTREES
1575                 if (matched && saddr)
1576                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1577 #endif
1578                 if (matched)
1579                         return rt6_ex;
1580         }
1581         return NULL;
1582 }
1583
1584 static unsigned int fib6_mtu(const struct fib6_result *res)
1585 {
1586         const struct fib6_nh *nh = res->nh;
1587         unsigned int mtu;
1588
1589         if (res->f6i->fib6_pmtu) {
1590                 mtu = res->f6i->fib6_pmtu;
1591         } else {
1592                 struct net_device *dev = nh->fib_nh_dev;
1593                 struct inet6_dev *idev;
1594
1595                 rcu_read_lock();
1596                 idev = __in6_dev_get(dev);
1597                 mtu = idev->cnf.mtu6;
1598                 rcu_read_unlock();
1599         }
1600
1601         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1602
1603         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1604 }
1605
1606 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1607
1608 /* used when the flushed bit is not relevant, only access to the bucket
1609  * (ie., all bucket users except rt6_insert_exception);
1610  *
1611  * called under rcu lock; sometimes called with rt6_exception_lock held
1612  */
1613 static
1614 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1615                                                        spinlock_t *lock)
1616 {
1617         struct rt6_exception_bucket *bucket;
1618
1619         if (lock)
1620                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1621                                                    lockdep_is_held(lock));
1622         else
1623                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1624
1625         /* remove bucket flushed bit if set */
1626         if (bucket) {
1627                 unsigned long p = (unsigned long)bucket;
1628
1629                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1630                 bucket = (struct rt6_exception_bucket *)p;
1631         }
1632
1633         return bucket;
1634 }
1635
1636 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1637 {
1638         unsigned long p = (unsigned long)bucket;
1639
1640         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1641 }
1642
1643 /* called with rt6_exception_lock held */
1644 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1645                                               spinlock_t *lock)
1646 {
1647         struct rt6_exception_bucket *bucket;
1648         unsigned long p;
1649
1650         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1651                                            lockdep_is_held(lock));
1652
1653         p = (unsigned long)bucket;
1654         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1655         bucket = (struct rt6_exception_bucket *)p;
1656         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1657 }
1658
1659 static int rt6_insert_exception(struct rt6_info *nrt,
1660                                 const struct fib6_result *res)
1661 {
1662         struct net *net = dev_net(nrt->dst.dev);
1663         struct rt6_exception_bucket *bucket;
1664         struct fib6_info *f6i = res->f6i;
1665         struct in6_addr *src_key = NULL;
1666         struct rt6_exception *rt6_ex;
1667         struct fib6_nh *nh = res->nh;
1668         int err = 0;
1669
1670         spin_lock_bh(&rt6_exception_lock);
1671
1672         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1673                                           lockdep_is_held(&rt6_exception_lock));
1674         if (!bucket) {
1675                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1676                                  GFP_ATOMIC);
1677                 if (!bucket) {
1678                         err = -ENOMEM;
1679                         goto out;
1680                 }
1681                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1682         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1683                 err = -EINVAL;
1684                 goto out;
1685         }
1686
1687 #ifdef CONFIG_IPV6_SUBTREES
1688         /* fib6_src.plen != 0 indicates f6i is in subtree
1689          * and exception table is indexed by a hash of
1690          * both fib6_dst and fib6_src.
1691          * Otherwise, the exception table is indexed by
1692          * a hash of only fib6_dst.
1693          */
1694         if (f6i->fib6_src.plen)
1695                 src_key = &nrt->rt6i_src.addr;
1696 #endif
1697         /* rt6_mtu_change() might lower mtu on f6i.
1698          * Only insert this exception route if its mtu
1699          * is less than f6i's mtu value.
1700          */
1701         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1702                 err = -EINVAL;
1703                 goto out;
1704         }
1705
1706         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1707                                                src_key);
1708         if (rt6_ex)
1709                 rt6_remove_exception(bucket, rt6_ex);
1710
1711         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1712         if (!rt6_ex) {
1713                 err = -ENOMEM;
1714                 goto out;
1715         }
1716         rt6_ex->rt6i = nrt;
1717         rt6_ex->stamp = jiffies;
1718         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1719         bucket->depth++;
1720         net->ipv6.rt6_stats->fib_rt_cache++;
1721
1722         if (bucket->depth > FIB6_MAX_DEPTH)
1723                 rt6_exception_remove_oldest(bucket);
1724
1725 out:
1726         spin_unlock_bh(&rt6_exception_lock);
1727
1728         /* Update fn->fn_sernum to invalidate all cached dst */
1729         if (!err) {
1730                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1731                 fib6_update_sernum(net, f6i);
1732                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1733                 fib6_force_start_gc(net);
1734         }
1735
1736         return err;
1737 }
1738
1739 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1740 {
1741         struct rt6_exception_bucket *bucket;
1742         struct rt6_exception *rt6_ex;
1743         struct hlist_node *tmp;
1744         int i;
1745
1746         spin_lock_bh(&rt6_exception_lock);
1747
1748         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1749         if (!bucket)
1750                 goto out;
1751
1752         /* Prevent rt6_insert_exception() to recreate the bucket list */
1753         if (!from)
1754                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1755
1756         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1757                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1758                         if (!from ||
1759                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1760                                 rt6_remove_exception(bucket, rt6_ex);
1761                 }
1762                 WARN_ON_ONCE(!from && bucket->depth);
1763                 bucket++;
1764         }
1765 out:
1766         spin_unlock_bh(&rt6_exception_lock);
1767 }
1768
1769 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1770 {
1771         struct fib6_info *f6i = arg;
1772
1773         fib6_nh_flush_exceptions(nh, f6i);
1774
1775         return 0;
1776 }
1777
1778 void rt6_flush_exceptions(struct fib6_info *f6i)
1779 {
1780         if (f6i->nh)
1781                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1782                                          f6i);
1783         else
1784                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1785 }
1786
1787 /* Find cached rt in the hash table inside passed in rt
1788  * Caller has to hold rcu_read_lock()
1789  */
1790 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1791                                            const struct in6_addr *daddr,
1792                                            const struct in6_addr *saddr)
1793 {
1794         const struct in6_addr *src_key = NULL;
1795         struct rt6_exception_bucket *bucket;
1796         struct rt6_exception *rt6_ex;
1797         struct rt6_info *ret = NULL;
1798
1799 #ifdef CONFIG_IPV6_SUBTREES
1800         /* fib6i_src.plen != 0 indicates f6i is in subtree
1801          * and exception table is indexed by a hash of
1802          * both fib6_dst and fib6_src.
1803          * However, the src addr used to create the hash
1804          * might not be exactly the passed in saddr which
1805          * is a /128 addr from the flow.
1806          * So we need to use f6i->fib6_src to redo lookup
1807          * if the passed in saddr does not find anything.
1808          * (See the logic in ip6_rt_cache_alloc() on how
1809          * rt->rt6i_src is updated.)
1810          */
1811         if (res->f6i->fib6_src.plen)
1812                 src_key = saddr;
1813 find_ex:
1814 #endif
1815         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1816         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1817
1818         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1819                 ret = rt6_ex->rt6i;
1820
1821 #ifdef CONFIG_IPV6_SUBTREES
1822         /* Use fib6_src as src_key and redo lookup */
1823         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1824                 src_key = &res->f6i->fib6_src.addr;
1825                 goto find_ex;
1826         }
1827 #endif
1828
1829         return ret;
1830 }
1831
1832 /* Remove the passed in cached rt from the hash table that contains it */
1833 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1834                                     const struct rt6_info *rt)
1835 {
1836         const struct in6_addr *src_key = NULL;
1837         struct rt6_exception_bucket *bucket;
1838         struct rt6_exception *rt6_ex;
1839         int err;
1840
1841         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1842                 return -ENOENT;
1843
1844         spin_lock_bh(&rt6_exception_lock);
1845         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1846
1847 #ifdef CONFIG_IPV6_SUBTREES
1848         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1849          * and exception table is indexed by a hash of
1850          * both rt6i_dst and rt6i_src.
1851          * Otherwise, the exception table is indexed by
1852          * a hash of only rt6i_dst.
1853          */
1854         if (plen)
1855                 src_key = &rt->rt6i_src.addr;
1856 #endif
1857         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1858                                                &rt->rt6i_dst.addr,
1859                                                src_key);
1860         if (rt6_ex) {
1861                 rt6_remove_exception(bucket, rt6_ex);
1862                 err = 0;
1863         } else {
1864                 err = -ENOENT;
1865         }
1866
1867         spin_unlock_bh(&rt6_exception_lock);
1868         return err;
1869 }
1870
1871 struct fib6_nh_excptn_arg {
1872         struct rt6_info *rt;
1873         int             plen;
1874 };
1875
1876 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1877 {
1878         struct fib6_nh_excptn_arg *arg = _arg;
1879         int err;
1880
1881         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1882         if (err == 0)
1883                 return 1;
1884
1885         return 0;
1886 }
1887
1888 static int rt6_remove_exception_rt(struct rt6_info *rt)
1889 {
1890         struct fib6_info *from;
1891
1892         from = rcu_dereference(rt->from);
1893         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1894                 return -EINVAL;
1895
1896         if (from->nh) {
1897                 struct fib6_nh_excptn_arg arg = {
1898                         .rt = rt,
1899                         .plen = from->fib6_src.plen
1900                 };
1901                 int rc;
1902
1903                 /* rc = 1 means an entry was found */
1904                 rc = nexthop_for_each_fib6_nh(from->nh,
1905                                               rt6_nh_remove_exception_rt,
1906                                               &arg);
1907                 return rc ? 0 : -ENOENT;
1908         }
1909
1910         return fib6_nh_remove_exception(from->fib6_nh,
1911                                         from->fib6_src.plen, rt);
1912 }
1913
1914 /* Find rt6_ex which contains the passed in rt cache and
1915  * refresh its stamp
1916  */
1917 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1918                                      const struct rt6_info *rt)
1919 {
1920         const struct in6_addr *src_key = NULL;
1921         struct rt6_exception_bucket *bucket;
1922         struct rt6_exception *rt6_ex;
1923
1924         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1925 #ifdef CONFIG_IPV6_SUBTREES
1926         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1927          * and exception table is indexed by a hash of
1928          * both rt6i_dst and rt6i_src.
1929          * Otherwise, the exception table is indexed by
1930          * a hash of only rt6i_dst.
1931          */
1932         if (plen)
1933                 src_key = &rt->rt6i_src.addr;
1934 #endif
1935         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1936         if (rt6_ex)
1937                 rt6_ex->stamp = jiffies;
1938 }
1939
1940 struct fib6_nh_match_arg {
1941         const struct net_device *dev;
1942         const struct in6_addr   *gw;
1943         struct fib6_nh          *match;
1944 };
1945
1946 /* determine if fib6_nh has given device and gateway */
1947 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1948 {
1949         struct fib6_nh_match_arg *arg = _arg;
1950
1951         if (arg->dev != nh->fib_nh_dev ||
1952             (arg->gw && !nh->fib_nh_gw_family) ||
1953             (!arg->gw && nh->fib_nh_gw_family) ||
1954             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1955                 return 0;
1956
1957         arg->match = nh;
1958
1959         /* found a match, break the loop */
1960         return 1;
1961 }
1962
1963 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1964 {
1965         struct fib6_info *from;
1966         struct fib6_nh *fib6_nh;
1967
1968         rcu_read_lock();
1969
1970         from = rcu_dereference(rt->from);
1971         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1972                 goto unlock;
1973
1974         if (from->nh) {
1975                 struct fib6_nh_match_arg arg = {
1976                         .dev = rt->dst.dev,
1977                         .gw = &rt->rt6i_gateway,
1978                 };
1979
1980                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1981
1982                 if (!arg.match)
1983                         goto unlock;
1984                 fib6_nh = arg.match;
1985         } else {
1986                 fib6_nh = from->fib6_nh;
1987         }
1988         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1989 unlock:
1990         rcu_read_unlock();
1991 }
1992
1993 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1994                                          struct rt6_info *rt, int mtu)
1995 {
1996         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1997          * lowest MTU in the path: always allow updating the route PMTU to
1998          * reflect PMTU decreases.
1999          *
2000          * If the new MTU is higher, and the route PMTU is equal to the local
2001          * MTU, this means the old MTU is the lowest in the path, so allow
2002          * updating it: if other nodes now have lower MTUs, PMTU discovery will
2003          * handle this.
2004          */
2005
2006         if (dst_mtu(&rt->dst) >= mtu)
2007                 return true;
2008
2009         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2010                 return true;
2011
2012         return false;
2013 }
2014
2015 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2016                                        const struct fib6_nh *nh, int mtu)
2017 {
2018         struct rt6_exception_bucket *bucket;
2019         struct rt6_exception *rt6_ex;
2020         int i;
2021
2022         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2023         if (!bucket)
2024                 return;
2025
2026         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2027                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2028                         struct rt6_info *entry = rt6_ex->rt6i;
2029
2030                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2031                          * route), the metrics of its rt->from have already
2032                          * been updated.
2033                          */
2034                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2035                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2036                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2037                 }
2038                 bucket++;
2039         }
2040 }
2041
2042 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2043
2044 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2045                                             const struct in6_addr *gateway)
2046 {
2047         struct rt6_exception_bucket *bucket;
2048         struct rt6_exception *rt6_ex;
2049         struct hlist_node *tmp;
2050         int i;
2051
2052         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2053                 return;
2054
2055         spin_lock_bh(&rt6_exception_lock);
2056         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2057         if (bucket) {
2058                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2059                         hlist_for_each_entry_safe(rt6_ex, tmp,
2060                                                   &bucket->chain, hlist) {
2061                                 struct rt6_info *entry = rt6_ex->rt6i;
2062
2063                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2064                                     RTF_CACHE_GATEWAY &&
2065                                     ipv6_addr_equal(gateway,
2066                                                     &entry->rt6i_gateway)) {
2067                                         rt6_remove_exception(bucket, rt6_ex);
2068                                 }
2069                         }
2070                         bucket++;
2071                 }
2072         }
2073
2074         spin_unlock_bh(&rt6_exception_lock);
2075 }
2076
2077 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2078                                       struct rt6_exception *rt6_ex,
2079                                       struct fib6_gc_args *gc_args,
2080                                       unsigned long now)
2081 {
2082         struct rt6_info *rt = rt6_ex->rt6i;
2083
2084         /* we are pruning and obsoleting aged-out and non gateway exceptions
2085          * even if others have still references to them, so that on next
2086          * dst_check() such references can be dropped.
2087          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2088          * expired, independently from their aging, as per RFC 8201 section 4
2089          */
2090         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2091                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2092                         RT6_TRACE("aging clone %p\n", rt);
2093                         rt6_remove_exception(bucket, rt6_ex);
2094                         return;
2095                 }
2096         } else if (time_after(jiffies, rt->dst.expires)) {
2097                 RT6_TRACE("purging expired route %p\n", rt);
2098                 rt6_remove_exception(bucket, rt6_ex);
2099                 return;
2100         }
2101
2102         if (rt->rt6i_flags & RTF_GATEWAY) {
2103                 struct neighbour *neigh;
2104                 __u8 neigh_flags = 0;
2105
2106                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2107                 if (neigh)
2108                         neigh_flags = neigh->flags;
2109
2110                 if (!(neigh_flags & NTF_ROUTER)) {
2111                         RT6_TRACE("purging route %p via non-router but gateway\n",
2112                                   rt);
2113                         rt6_remove_exception(bucket, rt6_ex);
2114                         return;
2115                 }
2116         }
2117
2118         gc_args->more++;
2119 }
2120
2121 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2122                                    struct fib6_gc_args *gc_args,
2123                                    unsigned long now)
2124 {
2125         struct rt6_exception_bucket *bucket;
2126         struct rt6_exception *rt6_ex;
2127         struct hlist_node *tmp;
2128         int i;
2129
2130         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2131                 return;
2132
2133         rcu_read_lock_bh();
2134         spin_lock(&rt6_exception_lock);
2135         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2136         if (bucket) {
2137                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2138                         hlist_for_each_entry_safe(rt6_ex, tmp,
2139                                                   &bucket->chain, hlist) {
2140                                 rt6_age_examine_exception(bucket, rt6_ex,
2141                                                           gc_args, now);
2142                         }
2143                         bucket++;
2144                 }
2145         }
2146         spin_unlock(&rt6_exception_lock);
2147         rcu_read_unlock_bh();
2148 }
2149
2150 struct fib6_nh_age_excptn_arg {
2151         struct fib6_gc_args     *gc_args;
2152         unsigned long           now;
2153 };
2154
2155 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2156 {
2157         struct fib6_nh_age_excptn_arg *arg = _arg;
2158
2159         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2160         return 0;
2161 }
2162
2163 void rt6_age_exceptions(struct fib6_info *f6i,
2164                         struct fib6_gc_args *gc_args,
2165                         unsigned long now)
2166 {
2167         if (f6i->nh) {
2168                 struct fib6_nh_age_excptn_arg arg = {
2169                         .gc_args = gc_args,
2170                         .now = now
2171                 };
2172
2173                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2174                                          &arg);
2175         } else {
2176                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2177         }
2178 }
2179
2180 /* must be called with rcu lock held */
2181 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2182                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2183 {
2184         struct fib6_node *fn, *saved_fn;
2185
2186         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2187         saved_fn = fn;
2188
2189         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2190                 oif = 0;
2191
2192 redo_rt6_select:
2193         rt6_select(net, fn, oif, res, strict);
2194         if (res->f6i == net->ipv6.fib6_null_entry) {
2195                 fn = fib6_backtrack(fn, &fl6->saddr);
2196                 if (fn)
2197                         goto redo_rt6_select;
2198                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2199                         /* also consider unreachable route */
2200                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2201                         fn = saved_fn;
2202                         goto redo_rt6_select;
2203                 }
2204         }
2205
2206         trace_fib6_table_lookup(net, res, table, fl6);
2207
2208         return 0;
2209 }
2210
2211 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2212                                int oif, struct flowi6 *fl6,
2213                                const struct sk_buff *skb, int flags)
2214 {
2215         struct fib6_result res = {};
2216         struct rt6_info *rt = NULL;
2217         int strict = 0;
2218
2219         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2220                      !rcu_read_lock_held());
2221
2222         strict |= flags & RT6_LOOKUP_F_IFACE;
2223         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2224         if (net->ipv6.devconf_all->forwarding == 0)
2225                 strict |= RT6_LOOKUP_F_REACHABLE;
2226
2227         rcu_read_lock();
2228
2229         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2230         if (res.f6i == net->ipv6.fib6_null_entry)
2231                 goto out;
2232
2233         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2234
2235         /*Search through exception table */
2236         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2237         if (rt) {
2238                 goto out;
2239         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2240                             !res.nh->fib_nh_gw_family)) {
2241                 /* Create a RTF_CACHE clone which will not be
2242                  * owned by the fib6 tree.  It is for the special case where
2243                  * the daddr in the skb during the neighbor look-up is different
2244                  * from the fl6->daddr used to look-up route here.
2245                  */
2246                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2247
2248                 if (rt) {
2249                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2250                          * As rt6_uncached_list_add() does not consume refcnt,
2251                          * this refcnt is always returned to the caller even
2252                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2253                          */
2254                         rt6_uncached_list_add(rt);
2255                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2256                         rcu_read_unlock();
2257
2258                         return rt;
2259                 }
2260         } else {
2261                 /* Get a percpu copy */
2262                 local_bh_disable();
2263                 rt = rt6_get_pcpu_route(&res);
2264
2265                 if (!rt)
2266                         rt = rt6_make_pcpu_route(net, &res);
2267
2268                 local_bh_enable();
2269         }
2270 out:
2271         if (!rt)
2272                 rt = net->ipv6.ip6_null_entry;
2273         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2274                 ip6_hold_safe(net, &rt);
2275         rcu_read_unlock();
2276
2277         return rt;
2278 }
2279 EXPORT_SYMBOL_GPL(ip6_pol_route);
2280
2281 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2282                                             struct fib6_table *table,
2283                                             struct flowi6 *fl6,
2284                                             const struct sk_buff *skb,
2285                                             int flags)
2286 {
2287         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2288 }
2289
2290 struct dst_entry *ip6_route_input_lookup(struct net *net,
2291                                          struct net_device *dev,
2292                                          struct flowi6 *fl6,
2293                                          const struct sk_buff *skb,
2294                                          int flags)
2295 {
2296         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2297                 flags |= RT6_LOOKUP_F_IFACE;
2298
2299         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2300 }
2301 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2302
2303 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2304                                   struct flow_keys *keys,
2305                                   struct flow_keys *flkeys)
2306 {
2307         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2308         const struct ipv6hdr *key_iph = outer_iph;
2309         struct flow_keys *_flkeys = flkeys;
2310         const struct ipv6hdr *inner_iph;
2311         const struct icmp6hdr *icmph;
2312         struct ipv6hdr _inner_iph;
2313         struct icmp6hdr _icmph;
2314
2315         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2316                 goto out;
2317
2318         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2319                                    sizeof(_icmph), &_icmph);
2320         if (!icmph)
2321                 goto out;
2322
2323         if (!icmpv6_is_err(icmph->icmp6_type))
2324                 goto out;
2325
2326         inner_iph = skb_header_pointer(skb,
2327                                        skb_transport_offset(skb) + sizeof(*icmph),
2328                                        sizeof(_inner_iph), &_inner_iph);
2329         if (!inner_iph)
2330                 goto out;
2331
2332         key_iph = inner_iph;
2333         _flkeys = NULL;
2334 out:
2335         if (_flkeys) {
2336                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2337                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2338                 keys->tags.flow_label = _flkeys->tags.flow_label;
2339                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2340         } else {
2341                 keys->addrs.v6addrs.src = key_iph->saddr;
2342                 keys->addrs.v6addrs.dst = key_iph->daddr;
2343                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2344                 keys->basic.ip_proto = key_iph->nexthdr;
2345         }
2346 }
2347
2348 /* if skb is set it will be used and fl6 can be NULL */
2349 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2350                        const struct sk_buff *skb, struct flow_keys *flkeys)
2351 {
2352         struct flow_keys hash_keys;
2353         u32 mhash;
2354
2355         switch (ip6_multipath_hash_policy(net)) {
2356         case 0:
2357                 memset(&hash_keys, 0, sizeof(hash_keys));
2358                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2359                 if (skb) {
2360                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2361                 } else {
2362                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2363                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2364                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2365                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2366                 }
2367                 break;
2368         case 1:
2369                 if (skb) {
2370                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2371                         struct flow_keys keys;
2372
2373                         /* short-circuit if we already have L4 hash present */
2374                         if (skb->l4_hash)
2375                                 return skb_get_hash_raw(skb) >> 1;
2376
2377                         memset(&hash_keys, 0, sizeof(hash_keys));
2378
2379                         if (!flkeys) {
2380                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2381                                 flkeys = &keys;
2382                         }
2383                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2384                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2385                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2386                         hash_keys.ports.src = flkeys->ports.src;
2387                         hash_keys.ports.dst = flkeys->ports.dst;
2388                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2389                 } else {
2390                         memset(&hash_keys, 0, sizeof(hash_keys));
2391                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2392                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2393                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2394                         hash_keys.ports.src = fl6->fl6_sport;
2395                         hash_keys.ports.dst = fl6->fl6_dport;
2396                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2397                 }
2398                 break;
2399         case 2:
2400                 memset(&hash_keys, 0, sizeof(hash_keys));
2401                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2402                 if (skb) {
2403                         struct flow_keys keys;
2404
2405                         if (!flkeys) {
2406                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2407                                 flkeys = &keys;
2408                         }
2409
2410                         /* Inner can be v4 or v6 */
2411                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2412                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2413                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2414                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2415                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2416                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2417                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2418                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2419                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2420                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2421                         } else {
2422                                 /* Same as case 0 */
2423                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2424                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2425                         }
2426                 } else {
2427                         /* Same as case 0 */
2428                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2429                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2430                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2431                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2432                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2433                 }
2434                 break;
2435         }
2436         mhash = flow_hash_from_keys(&hash_keys);
2437
2438         return mhash >> 1;
2439 }
2440
2441 /* Called with rcu held */
2442 void ip6_route_input(struct sk_buff *skb)
2443 {
2444         const struct ipv6hdr *iph = ipv6_hdr(skb);
2445         struct net *net = dev_net(skb->dev);
2446         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2447         struct ip_tunnel_info *tun_info;
2448         struct flowi6 fl6 = {
2449                 .flowi6_iif = skb->dev->ifindex,
2450                 .daddr = iph->daddr,
2451                 .saddr = iph->saddr,
2452                 .flowlabel = ip6_flowinfo(iph),
2453                 .flowi6_mark = skb->mark,
2454                 .flowi6_proto = iph->nexthdr,
2455         };
2456         struct flow_keys *flkeys = NULL, _flkeys;
2457
2458         tun_info = skb_tunnel_info(skb);
2459         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2460                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2461
2462         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2463                 flkeys = &_flkeys;
2464
2465         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2466                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2467         skb_dst_drop(skb);
2468         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2469                                                       &fl6, skb, flags));
2470 }
2471
2472 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2473                                              struct fib6_table *table,
2474                                              struct flowi6 *fl6,
2475                                              const struct sk_buff *skb,
2476                                              int flags)
2477 {
2478         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2479 }
2480
2481 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2482                                                const struct sock *sk,
2483                                                struct flowi6 *fl6, int flags)
2484 {
2485         bool any_src;
2486
2487         if (ipv6_addr_type(&fl6->daddr) &
2488             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2489                 struct dst_entry *dst;
2490
2491                 /* This function does not take refcnt on the dst */
2492                 dst = l3mdev_link_scope_lookup(net, fl6);
2493                 if (dst)
2494                         return dst;
2495         }
2496
2497         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2498
2499         flags |= RT6_LOOKUP_F_DST_NOREF;
2500         any_src = ipv6_addr_any(&fl6->saddr);
2501         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2502             (fl6->flowi6_oif && any_src))
2503                 flags |= RT6_LOOKUP_F_IFACE;
2504
2505         if (!any_src)
2506                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2507         else if (sk)
2508                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2509
2510         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2511 }
2512 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2513
2514 struct dst_entry *ip6_route_output_flags(struct net *net,
2515                                          const struct sock *sk,
2516                                          struct flowi6 *fl6,
2517                                          int flags)
2518 {
2519         struct dst_entry *dst;
2520         struct rt6_info *rt6;
2521
2522         rcu_read_lock();
2523         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2524         rt6 = (struct rt6_info *)dst;
2525         /* For dst cached in uncached_list, refcnt is already taken. */
2526         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2527                 dst = &net->ipv6.ip6_null_entry->dst;
2528                 dst_hold(dst);
2529         }
2530         rcu_read_unlock();
2531
2532         return dst;
2533 }
2534 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2535
2536 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2537 {
2538         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2539         struct net_device *loopback_dev = net->loopback_dev;
2540         struct dst_entry *new = NULL;
2541
2542         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2543                        DST_OBSOLETE_DEAD, 0);
2544         if (rt) {
2545                 rt6_info_init(rt);
2546                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2547
2548                 new = &rt->dst;
2549                 new->__use = 1;
2550                 new->input = dst_discard;
2551                 new->output = dst_discard_out;
2552
2553                 dst_copy_metrics(new, &ort->dst);
2554
2555                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2556                 rt->rt6i_gateway = ort->rt6i_gateway;
2557                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2558
2559                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2560 #ifdef CONFIG_IPV6_SUBTREES
2561                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2562 #endif
2563         }
2564
2565         dst_release(dst_orig);
2566         return new ? new : ERR_PTR(-ENOMEM);
2567 }
2568
2569 /*
2570  *      Destination cache support functions
2571  */
2572
2573 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2574 {
2575         u32 rt_cookie = 0;
2576
2577         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2578                 return false;
2579
2580         if (fib6_check_expired(f6i))
2581                 return false;
2582
2583         return true;
2584 }
2585
2586 static struct dst_entry *rt6_check(struct rt6_info *rt,
2587                                    struct fib6_info *from,
2588                                    u32 cookie)
2589 {
2590         u32 rt_cookie = 0;
2591
2592         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2593             rt_cookie != cookie)
2594                 return NULL;
2595
2596         if (rt6_check_expired(rt))
2597                 return NULL;
2598
2599         return &rt->dst;
2600 }
2601
2602 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2603                                             struct fib6_info *from,
2604                                             u32 cookie)
2605 {
2606         if (!__rt6_check_expired(rt) &&
2607             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2608             fib6_check(from, cookie))
2609                 return &rt->dst;
2610         else
2611                 return NULL;
2612 }
2613
2614 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2615 {
2616         struct dst_entry *dst_ret;
2617         struct fib6_info *from;
2618         struct rt6_info *rt;
2619
2620         rt = container_of(dst, struct rt6_info, dst);
2621
2622         if (rt->sernum)
2623                 return rt6_is_valid(rt) ? dst : NULL;
2624
2625         rcu_read_lock();
2626
2627         /* All IPV6 dsts are created with ->obsolete set to the value
2628          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2629          * into this function always.
2630          */
2631
2632         from = rcu_dereference(rt->from);
2633
2634         if (from && (rt->rt6i_flags & RTF_PCPU ||
2635             unlikely(!list_empty(&rt->rt6i_uncached))))
2636                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2637         else
2638                 dst_ret = rt6_check(rt, from, cookie);
2639
2640         rcu_read_unlock();
2641
2642         return dst_ret;
2643 }
2644
2645 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2646 {
2647         struct rt6_info *rt = (struct rt6_info *) dst;
2648
2649         if (rt) {
2650                 if (rt->rt6i_flags & RTF_CACHE) {
2651                         rcu_read_lock();
2652                         if (rt6_check_expired(rt)) {
2653                                 rt6_remove_exception_rt(rt);
2654                                 dst = NULL;
2655                         }
2656                         rcu_read_unlock();
2657                 } else {
2658                         dst_release(dst);
2659                         dst = NULL;
2660                 }
2661         }
2662         return dst;
2663 }
2664
2665 static void ip6_link_failure(struct sk_buff *skb)
2666 {
2667         struct rt6_info *rt;
2668
2669         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2670
2671         rt = (struct rt6_info *) skb_dst(skb);
2672         if (rt) {
2673                 rcu_read_lock();
2674                 if (rt->rt6i_flags & RTF_CACHE) {
2675                         rt6_remove_exception_rt(rt);
2676                 } else {
2677                         struct fib6_info *from;
2678                         struct fib6_node *fn;
2679
2680                         from = rcu_dereference(rt->from);
2681                         if (from) {
2682                                 fn = rcu_dereference(from->fib6_node);
2683                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2684                                         fn->fn_sernum = -1;
2685                         }
2686                 }
2687                 rcu_read_unlock();
2688         }
2689 }
2690
2691 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2692 {
2693         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2694                 struct fib6_info *from;
2695
2696                 rcu_read_lock();
2697                 from = rcu_dereference(rt0->from);
2698                 if (from)
2699                         rt0->dst.expires = from->expires;
2700                 rcu_read_unlock();
2701         }
2702
2703         dst_set_expires(&rt0->dst, timeout);
2704         rt0->rt6i_flags |= RTF_EXPIRES;
2705 }
2706
2707 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2708 {
2709         struct net *net = dev_net(rt->dst.dev);
2710
2711         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2712         rt->rt6i_flags |= RTF_MODIFIED;
2713         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2714 }
2715
2716 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2717 {
2718         return !(rt->rt6i_flags & RTF_CACHE) &&
2719                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2720 }
2721
2722 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2723                                  const struct ipv6hdr *iph, u32 mtu,
2724                                  bool confirm_neigh)
2725 {
2726         const struct in6_addr *daddr, *saddr;
2727         struct rt6_info *rt6 = (struct rt6_info *)dst;
2728
2729         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2730          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2731          * [see also comment in rt6_mtu_change_route()]
2732          */
2733
2734         if (iph) {
2735                 daddr = &iph->daddr;
2736                 saddr = &iph->saddr;
2737         } else if (sk) {
2738                 daddr = &sk->sk_v6_daddr;
2739                 saddr = &inet6_sk(sk)->saddr;
2740         } else {
2741                 daddr = NULL;
2742                 saddr = NULL;
2743         }
2744
2745         if (confirm_neigh)
2746                 dst_confirm_neigh(dst, daddr);
2747
2748         mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2749         if (mtu >= dst_mtu(dst))
2750                 return;
2751
2752         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2753                 rt6_do_update_pmtu(rt6, mtu);
2754                 /* update rt6_ex->stamp for cache */
2755                 if (rt6->rt6i_flags & RTF_CACHE)
2756                         rt6_update_exception_stamp_rt(rt6);
2757         } else if (daddr) {
2758                 struct fib6_result res = {};
2759                 struct rt6_info *nrt6;
2760
2761                 rcu_read_lock();
2762                 res.f6i = rcu_dereference(rt6->from);
2763                 if (!res.f6i)
2764                         goto out_unlock;
2765
2766                 res.fib6_flags = res.f6i->fib6_flags;
2767                 res.fib6_type = res.f6i->fib6_type;
2768
2769                 if (res.f6i->nh) {
2770                         struct fib6_nh_match_arg arg = {
2771                                 .dev = dst->dev,
2772                                 .gw = &rt6->rt6i_gateway,
2773                         };
2774
2775                         nexthop_for_each_fib6_nh(res.f6i->nh,
2776                                                  fib6_nh_find_match, &arg);
2777
2778                         /* fib6_info uses a nexthop that does not have fib6_nh
2779                          * using the dst->dev + gw. Should be impossible.
2780                          */
2781                         if (!arg.match)
2782                                 goto out_unlock;
2783
2784                         res.nh = arg.match;
2785                 } else {
2786                         res.nh = res.f6i->fib6_nh;
2787                 }
2788
2789                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2790                 if (nrt6) {
2791                         rt6_do_update_pmtu(nrt6, mtu);
2792                         if (rt6_insert_exception(nrt6, &res))
2793                                 dst_release_immediate(&nrt6->dst);
2794                 }
2795 out_unlock:
2796                 rcu_read_unlock();
2797         }
2798 }
2799
2800 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2801                                struct sk_buff *skb, u32 mtu,
2802                                bool confirm_neigh)
2803 {
2804         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2805                              confirm_neigh);
2806 }
2807
2808 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2809                      int oif, u32 mark, kuid_t uid)
2810 {
2811         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2812         struct dst_entry *dst;
2813         struct flowi6 fl6 = {
2814                 .flowi6_oif = oif,
2815                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2816                 .daddr = iph->daddr,
2817                 .saddr = iph->saddr,
2818                 .flowlabel = ip6_flowinfo(iph),
2819                 .flowi6_uid = uid,
2820         };
2821
2822         dst = ip6_route_output(net, NULL, &fl6);
2823         if (!dst->error)
2824                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2825         dst_release(dst);
2826 }
2827 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2828
2829 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2830 {
2831         int oif = sk->sk_bound_dev_if;
2832         struct dst_entry *dst;
2833
2834         if (!oif && skb->dev)
2835                 oif = l3mdev_master_ifindex(skb->dev);
2836
2837         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2838
2839         dst = __sk_dst_get(sk);
2840         if (!dst || !dst->obsolete ||
2841             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2842                 return;
2843
2844         bh_lock_sock(sk);
2845         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2846                 ip6_datagram_dst_update(sk, false);
2847         bh_unlock_sock(sk);
2848 }
2849 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2850
2851 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2852                            const struct flowi6 *fl6)
2853 {
2854 #ifdef CONFIG_IPV6_SUBTREES
2855         struct ipv6_pinfo *np = inet6_sk(sk);
2856 #endif
2857
2858         ip6_dst_store(sk, dst,
2859                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2860                       &sk->sk_v6_daddr : NULL,
2861 #ifdef CONFIG_IPV6_SUBTREES
2862                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2863                       &np->saddr :
2864 #endif
2865                       NULL);
2866 }
2867
2868 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2869                                   struct flowi6 *fl6,
2870                                   const struct in6_addr *gw,
2871                                   struct rt6_info **ret)
2872 {
2873         const struct fib6_nh *nh = res->nh;
2874
2875         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2876             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2877                 return false;
2878
2879         /* rt_cache's gateway might be different from its 'parent'
2880          * in the case of an ip redirect.
2881          * So we keep searching in the exception table if the gateway
2882          * is different.
2883          */
2884         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2885                 struct rt6_info *rt_cache;
2886
2887                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2888                 if (rt_cache &&
2889                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2890                         *ret = rt_cache;
2891                         return true;
2892                 }
2893                 return false;
2894         }
2895         return true;
2896 }
2897
2898 struct fib6_nh_rd_arg {
2899         struct fib6_result      *res;
2900         struct flowi6           *fl6;
2901         const struct in6_addr   *gw;
2902         struct rt6_info         **ret;
2903 };
2904
2905 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2906 {
2907         struct fib6_nh_rd_arg *arg = _arg;
2908
2909         arg->res->nh = nh;
2910         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2911 }
2912
2913 /* Handle redirects */
2914 struct ip6rd_flowi {
2915         struct flowi6 fl6;
2916         struct in6_addr gateway;
2917 };
2918
2919 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
2920                                              struct fib6_table *table,
2921                                              struct flowi6 *fl6,
2922                                              const struct sk_buff *skb,
2923                                              int flags)
2924 {
2925         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2926         struct rt6_info *ret = NULL;
2927         struct fib6_result res = {};
2928         struct fib6_nh_rd_arg arg = {
2929                 .res = &res,
2930                 .fl6 = fl6,
2931                 .gw  = &rdfl->gateway,
2932                 .ret = &ret
2933         };
2934         struct fib6_info *rt;
2935         struct fib6_node *fn;
2936
2937         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2938          * this case we must match on the real ingress device, so reset it
2939          */
2940         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2941                 fl6->flowi6_oif = skb->dev->ifindex;
2942
2943         /* Get the "current" route for this destination and
2944          * check if the redirect has come from appropriate router.
2945          *
2946          * RFC 4861 specifies that redirects should only be
2947          * accepted if they come from the nexthop to the target.
2948          * Due to the way the routes are chosen, this notion
2949          * is a bit fuzzy and one might need to check all possible
2950          * routes.
2951          */
2952
2953         rcu_read_lock();
2954         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2955 restart:
2956         for_each_fib6_node_rt_rcu(fn) {
2957                 res.f6i = rt;
2958                 if (fib6_check_expired(rt))
2959                         continue;
2960                 if (rt->fib6_flags & RTF_REJECT)
2961                         break;
2962                 if (unlikely(rt->nh)) {
2963                         if (nexthop_is_blackhole(rt->nh))
2964                                 continue;
2965                         /* on match, res->nh is filled in and potentially ret */
2966                         if (nexthop_for_each_fib6_nh(rt->nh,
2967                                                      fib6_nh_redirect_match,
2968                                                      &arg))
2969                                 goto out;
2970                 } else {
2971                         res.nh = rt->fib6_nh;
2972                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2973                                                   &ret))
2974                                 goto out;
2975                 }
2976         }
2977
2978         if (!rt)
2979                 rt = net->ipv6.fib6_null_entry;
2980         else if (rt->fib6_flags & RTF_REJECT) {
2981                 ret = net->ipv6.ip6_null_entry;
2982                 goto out;
2983         }
2984
2985         if (rt == net->ipv6.fib6_null_entry) {
2986                 fn = fib6_backtrack(fn, &fl6->saddr);
2987                 if (fn)
2988                         goto restart;
2989         }
2990
2991         res.f6i = rt;
2992         res.nh = rt->fib6_nh;
2993 out:
2994         if (ret) {
2995                 ip6_hold_safe(net, &ret);
2996         } else {
2997                 res.fib6_flags = res.f6i->fib6_flags;
2998                 res.fib6_type = res.f6i->fib6_type;
2999                 ret = ip6_create_rt_rcu(&res);
3000         }
3001
3002         rcu_read_unlock();
3003
3004         trace_fib6_table_lookup(net, &res, table, fl6);
3005         return ret;
3006 };
3007
3008 static struct dst_entry *ip6_route_redirect(struct net *net,
3009                                             const struct flowi6 *fl6,
3010                                             const struct sk_buff *skb,
3011                                             const struct in6_addr *gateway)
3012 {
3013         int flags = RT6_LOOKUP_F_HAS_SADDR;
3014         struct ip6rd_flowi rdfl;
3015
3016         rdfl.fl6 = *fl6;
3017         rdfl.gateway = *gateway;
3018
3019         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3020                                 flags, __ip6_route_redirect);
3021 }
3022
3023 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3024                   kuid_t uid)
3025 {
3026         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3027         struct dst_entry *dst;
3028         struct flowi6 fl6 = {
3029                 .flowi6_iif = LOOPBACK_IFINDEX,
3030                 .flowi6_oif = oif,
3031                 .flowi6_mark = mark,
3032                 .daddr = iph->daddr,
3033                 .saddr = iph->saddr,
3034                 .flowlabel = ip6_flowinfo(iph),
3035                 .flowi6_uid = uid,
3036         };
3037
3038         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3039         rt6_do_redirect(dst, NULL, skb);
3040         dst_release(dst);
3041 }
3042 EXPORT_SYMBOL_GPL(ip6_redirect);
3043
3044 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3045 {
3046         const struct ipv6hdr *iph = ipv6_hdr(skb);
3047         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3048         struct dst_entry *dst;
3049         struct flowi6 fl6 = {
3050                 .flowi6_iif = LOOPBACK_IFINDEX,
3051                 .flowi6_oif = oif,
3052                 .daddr = msg->dest,
3053                 .saddr = iph->daddr,
3054                 .flowi6_uid = sock_net_uid(net, NULL),
3055         };
3056
3057         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3058         rt6_do_redirect(dst, NULL, skb);
3059         dst_release(dst);
3060 }
3061
3062 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3063 {
3064         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3065                      sk->sk_uid);
3066 }
3067 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3068
3069 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3070 {
3071         struct net_device *dev = dst->dev;
3072         unsigned int mtu = dst_mtu(dst);
3073         struct net *net = dev_net(dev);
3074
3075         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3076
3077         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3078                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3079
3080         /*
3081          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3082          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3083          * IPV6_MAXPLEN is also valid and means: "any MSS,
3084          * rely only on pmtu discovery"
3085          */
3086         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3087                 mtu = IPV6_MAXPLEN;
3088         return mtu;
3089 }
3090
3091 static unsigned int ip6_mtu(const struct dst_entry *dst)
3092 {
3093         struct inet6_dev *idev;
3094         unsigned int mtu;
3095
3096         mtu = dst_metric_raw(dst, RTAX_MTU);
3097         if (mtu)
3098                 goto out;
3099
3100         mtu = IPV6_MIN_MTU;
3101
3102         rcu_read_lock();
3103         idev = __in6_dev_get(dst->dev);
3104         if (idev)
3105                 mtu = idev->cnf.mtu6;
3106         rcu_read_unlock();
3107
3108 out:
3109         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3110
3111         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3112 }
3113
3114 /* MTU selection:
3115  * 1. mtu on route is locked - use it
3116  * 2. mtu from nexthop exception
3117  * 3. mtu from egress device
3118  *
3119  * based on ip6_dst_mtu_forward and exception logic of
3120  * rt6_find_cached_rt; called with rcu_read_lock
3121  */
3122 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3123                       const struct in6_addr *daddr,
3124                       const struct in6_addr *saddr)
3125 {
3126         const struct fib6_nh *nh = res->nh;
3127         struct fib6_info *f6i = res->f6i;
3128         struct inet6_dev *idev;
3129         struct rt6_info *rt;
3130         u32 mtu = 0;
3131
3132         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3133                 mtu = f6i->fib6_pmtu;
3134                 if (mtu)
3135                         goto out;
3136         }
3137
3138         rt = rt6_find_cached_rt(res, daddr, saddr);
3139         if (unlikely(rt)) {
3140                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3141         } else {
3142                 struct net_device *dev = nh->fib_nh_dev;
3143
3144                 mtu = IPV6_MIN_MTU;
3145                 idev = __in6_dev_get(dev);
3146                 if (idev && idev->cnf.mtu6 > mtu)
3147                         mtu = idev->cnf.mtu6;
3148         }
3149
3150         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3151 out:
3152         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3153 }
3154
3155 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3156                                   struct flowi6 *fl6)
3157 {
3158         struct dst_entry *dst;
3159         struct rt6_info *rt;
3160         struct inet6_dev *idev = in6_dev_get(dev);
3161         struct net *net = dev_net(dev);
3162
3163         if (unlikely(!idev))
3164                 return ERR_PTR(-ENODEV);
3165
3166         rt = ip6_dst_alloc(net, dev, 0);
3167         if (unlikely(!rt)) {
3168                 in6_dev_put(idev);
3169                 dst = ERR_PTR(-ENOMEM);
3170                 goto out;
3171         }
3172
3173         rt->dst.input = ip6_input;
3174         rt->dst.output  = ip6_output;
3175         rt->rt6i_gateway  = fl6->daddr;
3176         rt->rt6i_dst.addr = fl6->daddr;
3177         rt->rt6i_dst.plen = 128;
3178         rt->rt6i_idev     = idev;
3179         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3180
3181         /* Add this dst into uncached_list so that rt6_disable_ip() can
3182          * do proper release of the net_device
3183          */
3184         rt6_uncached_list_add(rt);
3185         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3186
3187         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3188
3189 out:
3190         return dst;
3191 }
3192
3193 static int ip6_dst_gc(struct dst_ops *ops)
3194 {
3195         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3196         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3197         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3198         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3199         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3200         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3201         int entries;
3202
3203         entries = dst_entries_get_fast(ops);
3204         if (entries > rt_max_size)
3205                 entries = dst_entries_get_slow(ops);
3206
3207         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3208             entries <= rt_max_size)
3209                 goto out;
3210
3211         net->ipv6.ip6_rt_gc_expire++;
3212         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3213         entries = dst_entries_get_slow(ops);
3214         if (entries < ops->gc_thresh)
3215                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3216 out:
3217         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3218         return entries > rt_max_size;
3219 }
3220
3221 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3222                                const struct in6_addr *gw_addr, u32 tbid,
3223                                int flags, struct fib6_result *res)
3224 {
3225         struct flowi6 fl6 = {
3226                 .flowi6_oif = cfg->fc_ifindex,
3227                 .daddr = *gw_addr,
3228                 .saddr = cfg->fc_prefsrc,
3229         };
3230         struct fib6_table *table;
3231         int err;
3232
3233         table = fib6_get_table(net, tbid);
3234         if (!table)
3235                 return -EINVAL;
3236
3237         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3238                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3239
3240         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3241
3242         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3243         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3244                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3245                                  cfg->fc_ifindex != 0, NULL, flags);
3246
3247         return err;
3248 }
3249
3250 static int ip6_route_check_nh_onlink(struct net *net,
3251                                      struct fib6_config *cfg,
3252                                      const struct net_device *dev,
3253                                      struct netlink_ext_ack *extack)
3254 {
3255         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3256         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3257         struct fib6_result res = {};
3258         int err;
3259
3260         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3261         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3262             /* ignore match if it is the default route */
3263             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3264             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3265                 NL_SET_ERR_MSG(extack,
3266                                "Nexthop has invalid gateway or device mismatch");
3267                 err = -EINVAL;
3268         }
3269
3270         return err;
3271 }
3272
3273 static int ip6_route_check_nh(struct net *net,
3274                               struct fib6_config *cfg,
3275                               struct net_device **_dev,
3276                               struct inet6_dev **idev)
3277 {
3278         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3279         struct net_device *dev = _dev ? *_dev : NULL;
3280         int flags = RT6_LOOKUP_F_IFACE;
3281         struct fib6_result res = {};
3282         int err = -EHOSTUNREACH;
3283
3284         if (cfg->fc_table) {
3285                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3286                                           cfg->fc_table, flags, &res);
3287                 /* gw_addr can not require a gateway or resolve to a reject
3288                  * route. If a device is given, it must match the result.
3289                  */
3290                 if (err || res.fib6_flags & RTF_REJECT ||
3291                     res.nh->fib_nh_gw_family ||
3292                     (dev && dev != res.nh->fib_nh_dev))
3293                         err = -EHOSTUNREACH;
3294         }
3295
3296         if (err < 0) {
3297                 struct flowi6 fl6 = {
3298                         .flowi6_oif = cfg->fc_ifindex,
3299                         .daddr = *gw_addr,
3300                 };
3301
3302                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3303                 if (err || res.fib6_flags & RTF_REJECT ||
3304                     res.nh->fib_nh_gw_family)
3305                         err = -EHOSTUNREACH;
3306
3307                 if (err)
3308                         return err;
3309
3310                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3311                                  cfg->fc_ifindex != 0, NULL, flags);
3312         }
3313
3314         err = 0;
3315         if (dev) {
3316                 if (dev != res.nh->fib_nh_dev)
3317                         err = -EHOSTUNREACH;
3318         } else {
3319                 *_dev = dev = res.nh->fib_nh_dev;
3320                 dev_hold(dev);
3321                 *idev = in6_dev_get(dev);
3322         }
3323
3324         return err;
3325 }
3326
3327 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3328                            struct net_device **_dev, struct inet6_dev **idev,
3329                            struct netlink_ext_ack *extack)
3330 {
3331         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3332         int gwa_type = ipv6_addr_type(gw_addr);
3333         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3334         const struct net_device *dev = *_dev;
3335         bool need_addr_check = !dev;
3336         int err = -EINVAL;
3337
3338         /* if gw_addr is local we will fail to detect this in case
3339          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3340          * will return already-added prefix route via interface that
3341          * prefix route was assigned to, which might be non-loopback.
3342          */
3343         if (dev &&
3344             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3345                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3346                 goto out;
3347         }
3348
3349         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3350                 /* IPv6 strictly inhibits using not link-local
3351                  * addresses as nexthop address.
3352                  * Otherwise, router will not able to send redirects.
3353                  * It is very good, but in some (rare!) circumstances
3354                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3355                  * some exceptions. --ANK
3356                  * We allow IPv4-mapped nexthops to support RFC4798-type
3357                  * addressing
3358                  */
3359                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3360                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3361                         goto out;
3362                 }
3363
3364                 rcu_read_lock();
3365
3366                 if (cfg->fc_flags & RTNH_F_ONLINK)
3367                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3368                 else
3369                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3370
3371                 rcu_read_unlock();
3372
3373                 if (err)
3374                         goto out;
3375         }
3376
3377         /* reload in case device was changed */
3378         dev = *_dev;
3379
3380         err = -EINVAL;
3381         if (!dev) {
3382                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3383                 goto out;
3384         } else if (dev->flags & IFF_LOOPBACK) {
3385                 NL_SET_ERR_MSG(extack,
3386                                "Egress device can not be loopback device for this route");
3387                 goto out;
3388         }
3389
3390         /* if we did not check gw_addr above, do so now that the
3391          * egress device has been resolved.
3392          */
3393         if (need_addr_check &&
3394             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3395                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3396                 goto out;
3397         }
3398
3399         err = 0;
3400 out:
3401         return err;
3402 }
3403
3404 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3405 {
3406         if ((flags & RTF_REJECT) ||
3407             (dev && (dev->flags & IFF_LOOPBACK) &&
3408              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3409              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3410                 return true;
3411
3412         return false;
3413 }
3414
3415 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3416                  struct fib6_config *cfg, gfp_t gfp_flags,
3417                  struct netlink_ext_ack *extack)
3418 {
3419         struct net_device *dev = NULL;
3420         struct inet6_dev *idev = NULL;
3421         int addr_type;
3422         int err;
3423
3424         fib6_nh->fib_nh_family = AF_INET6;
3425 #ifdef CONFIG_IPV6_ROUTER_PREF
3426         fib6_nh->last_probe = jiffies;
3427 #endif
3428         if (cfg->fc_is_fdb) {
3429                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3430                 fib6_nh->fib_nh_gw_family = AF_INET6;
3431                 return 0;
3432         }
3433
3434         err = -ENODEV;
3435         if (cfg->fc_ifindex) {
3436                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3437                 if (!dev)
3438                         goto out;
3439                 idev = in6_dev_get(dev);
3440                 if (!idev)
3441                         goto out;
3442         }
3443
3444         if (cfg->fc_flags & RTNH_F_ONLINK) {
3445                 if (!dev) {
3446                         NL_SET_ERR_MSG(extack,
3447                                        "Nexthop device required for onlink");
3448                         goto out;
3449                 }
3450
3451                 if (!(dev->flags & IFF_UP)) {
3452                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3453                         err = -ENETDOWN;
3454                         goto out;
3455                 }
3456
3457                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3458         }
3459
3460         fib6_nh->fib_nh_weight = 1;
3461
3462         /* We cannot add true routes via loopback here,
3463          * they would result in kernel looping; promote them to reject routes
3464          */
3465         addr_type = ipv6_addr_type(&cfg->fc_dst);
3466         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3467                 /* hold loopback dev/idev if we haven't done so. */
3468                 if (dev != net->loopback_dev) {
3469                         if (dev) {
3470                                 dev_put(dev);
3471                                 in6_dev_put(idev);
3472                         }
3473                         dev = net->loopback_dev;
3474                         dev_hold(dev);
3475                         idev = in6_dev_get(dev);
3476                         if (!idev) {
3477                                 err = -ENODEV;
3478                                 goto out;
3479                         }
3480                 }
3481                 goto pcpu_alloc;
3482         }
3483
3484         if (cfg->fc_flags & RTF_GATEWAY) {
3485                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3486                 if (err)
3487                         goto out;
3488
3489                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3490                 fib6_nh->fib_nh_gw_family = AF_INET6;
3491         }
3492
3493         err = -ENODEV;
3494         if (!dev)
3495                 goto out;
3496
3497         if (idev->cnf.disable_ipv6) {
3498                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3499                 err = -EACCES;
3500                 goto out;
3501         }
3502
3503         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3504                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3505                 err = -ENETDOWN;
3506                 goto out;
3507         }
3508
3509         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3510             !netif_carrier_ok(dev))
3511                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3512
3513         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3514                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3515         if (err)
3516                 goto out;
3517
3518 pcpu_alloc:
3519         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3520         if (!fib6_nh->rt6i_pcpu) {
3521                 err = -ENOMEM;
3522                 goto out;
3523         }
3524
3525         fib6_nh->fib_nh_dev = dev;
3526         fib6_nh->fib_nh_oif = dev->ifindex;
3527         err = 0;
3528 out:
3529         if (idev)
3530                 in6_dev_put(idev);
3531
3532         if (err) {
3533                 lwtstate_put(fib6_nh->fib_nh_lws);
3534                 fib6_nh->fib_nh_lws = NULL;
3535                 if (dev)
3536                         dev_put(dev);
3537         }
3538
3539         return err;
3540 }
3541
3542 void fib6_nh_release(struct fib6_nh *fib6_nh)
3543 {
3544         struct rt6_exception_bucket *bucket;
3545
3546         rcu_read_lock();
3547
3548         fib6_nh_flush_exceptions(fib6_nh, NULL);
3549         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3550         if (bucket) {
3551                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3552                 kfree(bucket);
3553         }
3554
3555         rcu_read_unlock();
3556
3557         if (fib6_nh->rt6i_pcpu) {
3558                 int cpu;
3559
3560                 for_each_possible_cpu(cpu) {
3561                         struct rt6_info **ppcpu_rt;
3562                         struct rt6_info *pcpu_rt;
3563
3564                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3565                         pcpu_rt = *ppcpu_rt;
3566                         if (pcpu_rt) {
3567                                 dst_dev_put(&pcpu_rt->dst);
3568                                 dst_release(&pcpu_rt->dst);
3569                                 *ppcpu_rt = NULL;
3570                         }
3571                 }
3572
3573                 free_percpu(fib6_nh->rt6i_pcpu);
3574         }
3575
3576         fib_nh_common_release(&fib6_nh->nh_common);
3577 }
3578
3579 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3580                                               gfp_t gfp_flags,
3581                                               struct netlink_ext_ack *extack)
3582 {
3583         struct net *net = cfg->fc_nlinfo.nl_net;
3584         struct fib6_info *rt = NULL;
3585         struct nexthop *nh = NULL;
3586         struct fib6_table *table;
3587         struct fib6_nh *fib6_nh;
3588         int err = -EINVAL;
3589         int addr_type;
3590
3591         /* RTF_PCPU is an internal flag; can not be set by userspace */
3592         if (cfg->fc_flags & RTF_PCPU) {
3593                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3594                 goto out;
3595         }
3596
3597         /* RTF_CACHE is an internal flag; can not be set by userspace */
3598         if (cfg->fc_flags & RTF_CACHE) {
3599                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3600                 goto out;
3601         }
3602
3603         if (cfg->fc_type > RTN_MAX) {
3604                 NL_SET_ERR_MSG(extack, "Invalid route type");
3605                 goto out;
3606         }
3607
3608         if (cfg->fc_dst_len > 128) {
3609                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3610                 goto out;
3611         }
3612         if (cfg->fc_src_len > 128) {
3613                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3614                 goto out;
3615         }
3616 #ifndef CONFIG_IPV6_SUBTREES
3617         if (cfg->fc_src_len) {
3618                 NL_SET_ERR_MSG(extack,
3619                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3620                 goto out;
3621         }
3622 #endif
3623         if (cfg->fc_nh_id) {
3624                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3625                 if (!nh) {
3626                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3627                         goto out;
3628                 }
3629                 err = fib6_check_nexthop(nh, cfg, extack);
3630                 if (err)
3631                         goto out;
3632         }
3633
3634         err = -ENOBUFS;
3635         if (cfg->fc_nlinfo.nlh &&
3636             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3637                 table = fib6_get_table(net, cfg->fc_table);
3638                 if (!table) {
3639                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3640                         table = fib6_new_table(net, cfg->fc_table);
3641                 }
3642         } else {
3643                 table = fib6_new_table(net, cfg->fc_table);
3644         }
3645
3646         if (!table)
3647                 goto out;
3648
3649         err = -ENOMEM;
3650         rt = fib6_info_alloc(gfp_flags, !nh);
3651         if (!rt)
3652                 goto out;
3653
3654         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3655                                                extack);
3656         if (IS_ERR(rt->fib6_metrics)) {
3657                 err = PTR_ERR(rt->fib6_metrics);
3658                 /* Do not leave garbage there. */
3659                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3660                 goto out;
3661         }
3662
3663         if (cfg->fc_flags & RTF_ADDRCONF)
3664                 rt->dst_nocount = true;
3665
3666         if (cfg->fc_flags & RTF_EXPIRES)
3667                 fib6_set_expires(rt, jiffies +
3668                                 clock_t_to_jiffies(cfg->fc_expires));
3669         else
3670                 fib6_clean_expires(rt);
3671
3672         if (cfg->fc_protocol == RTPROT_UNSPEC)
3673                 cfg->fc_protocol = RTPROT_BOOT;
3674         rt->fib6_protocol = cfg->fc_protocol;
3675
3676         rt->fib6_table = table;
3677         rt->fib6_metric = cfg->fc_metric;
3678         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3679         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3680
3681         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3682         rt->fib6_dst.plen = cfg->fc_dst_len;
3683
3684 #ifdef CONFIG_IPV6_SUBTREES
3685         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3686         rt->fib6_src.plen = cfg->fc_src_len;
3687 #endif
3688         if (nh) {
3689                 if (rt->fib6_src.plen) {
3690                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3691                         goto out;
3692                 }
3693                 if (!nexthop_get(nh)) {
3694                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3695                         goto out;
3696                 }
3697                 rt->nh = nh;
3698                 fib6_nh = nexthop_fib6_nh(rt->nh);
3699         } else {
3700                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3701                 if (err)
3702                         goto out;
3703
3704                 fib6_nh = rt->fib6_nh;
3705
3706                 /* We cannot add true routes via loopback here, they would
3707                  * result in kernel looping; promote them to reject routes
3708                  */
3709                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3710                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3711                                    addr_type))
3712                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3713         }
3714
3715         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3716                 struct net_device *dev = fib6_nh->fib_nh_dev;
3717
3718                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3719                         NL_SET_ERR_MSG(extack, "Invalid source address");
3720                         err = -EINVAL;
3721                         goto out;
3722                 }
3723                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3724                 rt->fib6_prefsrc.plen = 128;
3725         } else
3726                 rt->fib6_prefsrc.plen = 0;
3727
3728         return rt;
3729 out:
3730         fib6_info_release(rt);
3731         return ERR_PTR(err);
3732 }
3733
3734 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3735                   struct netlink_ext_ack *extack)
3736 {
3737         struct fib6_info *rt;
3738         int err;
3739
3740         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3741         if (IS_ERR(rt))
3742                 return PTR_ERR(rt);
3743
3744         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3745         fib6_info_release(rt);
3746
3747         return err;
3748 }
3749
3750 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3751 {
3752         struct net *net = info->nl_net;
3753         struct fib6_table *table;
3754         int err;
3755
3756         if (rt == net->ipv6.fib6_null_entry) {
3757                 err = -ENOENT;
3758                 goto out;
3759         }
3760
3761         table = rt->fib6_table;
3762         spin_lock_bh(&table->tb6_lock);
3763         err = fib6_del(rt, info);
3764         spin_unlock_bh(&table->tb6_lock);
3765
3766 out:
3767         fib6_info_release(rt);
3768         return err;
3769 }
3770
3771 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3772 {
3773         struct nl_info info = {
3774                 .nl_net = net,
3775                 .skip_notify = skip_notify
3776         };
3777
3778         return __ip6_del_rt(rt, &info);
3779 }
3780
3781 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3782 {
3783         struct nl_info *info = &cfg->fc_nlinfo;
3784         struct net *net = info->nl_net;
3785         struct sk_buff *skb = NULL;
3786         struct fib6_table *table;
3787         int err = -ENOENT;
3788
3789         if (rt == net->ipv6.fib6_null_entry)
3790                 goto out_put;
3791         table = rt->fib6_table;
3792         spin_lock_bh(&table->tb6_lock);
3793
3794         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3795                 struct fib6_info *sibling, *next_sibling;
3796                 struct fib6_node *fn;
3797
3798                 /* prefer to send a single notification with all hops */
3799                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3800                 if (skb) {
3801                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3802
3803                         if (rt6_fill_node(net, skb, rt, NULL,
3804                                           NULL, NULL, 0, RTM_DELROUTE,
3805                                           info->portid, seq, 0) < 0) {
3806                                 kfree_skb(skb);
3807                                 skb = NULL;
3808                         } else
3809                                 info->skip_notify = 1;
3810                 }
3811
3812                 /* 'rt' points to the first sibling route. If it is not the
3813                  * leaf, then we do not need to send a notification. Otherwise,
3814                  * we need to check if the last sibling has a next route or not
3815                  * and emit a replace or delete notification, respectively.
3816                  */
3817                 info->skip_notify_kernel = 1;
3818                 fn = rcu_dereference_protected(rt->fib6_node,
3819                                             lockdep_is_held(&table->tb6_lock));
3820                 if (rcu_access_pointer(fn->leaf) == rt) {
3821                         struct fib6_info *last_sibling, *replace_rt;
3822
3823                         last_sibling = list_last_entry(&rt->fib6_siblings,
3824                                                        struct fib6_info,
3825                                                        fib6_siblings);
3826                         replace_rt = rcu_dereference_protected(
3827                                             last_sibling->fib6_next,
3828                                             lockdep_is_held(&table->tb6_lock));
3829                         if (replace_rt)
3830                                 call_fib6_entry_notifiers_replace(net,
3831                                                                   replace_rt);
3832                         else
3833                                 call_fib6_multipath_entry_notifiers(net,
3834                                                        FIB_EVENT_ENTRY_DEL,
3835                                                        rt, rt->fib6_nsiblings,
3836                                                        NULL);
3837                 }
3838                 list_for_each_entry_safe(sibling, next_sibling,
3839                                          &rt->fib6_siblings,
3840                                          fib6_siblings) {
3841                         err = fib6_del(sibling, info);
3842                         if (err)
3843                                 goto out_unlock;
3844                 }
3845         }
3846
3847         err = fib6_del(rt, info);
3848 out_unlock:
3849         spin_unlock_bh(&table->tb6_lock);
3850 out_put:
3851         fib6_info_release(rt);
3852
3853         if (skb) {
3854                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3855                             info->nlh, gfp_any());
3856         }
3857         return err;
3858 }
3859
3860 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3861 {
3862         int rc = -ESRCH;
3863
3864         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3865                 goto out;
3866
3867         if (cfg->fc_flags & RTF_GATEWAY &&
3868             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3869                 goto out;
3870
3871         rc = rt6_remove_exception_rt(rt);
3872 out:
3873         return rc;
3874 }
3875
3876 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3877                              struct fib6_nh *nh)
3878 {
3879         struct fib6_result res = {
3880                 .f6i = rt,
3881                 .nh = nh,
3882         };
3883         struct rt6_info *rt_cache;
3884
3885         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3886         if (rt_cache)
3887                 return __ip6_del_cached_rt(rt_cache, cfg);
3888
3889         return 0;
3890 }
3891
3892 struct fib6_nh_del_cached_rt_arg {
3893         struct fib6_config *cfg;
3894         struct fib6_info *f6i;
3895 };
3896
3897 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3898 {
3899         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3900         int rc;
3901
3902         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3903         return rc != -ESRCH ? rc : 0;
3904 }
3905
3906 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3907 {
3908         struct fib6_nh_del_cached_rt_arg arg = {
3909                 .cfg = cfg,
3910                 .f6i = f6i
3911         };
3912
3913         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3914 }
3915
3916 static int ip6_route_del(struct fib6_config *cfg,
3917                          struct netlink_ext_ack *extack)
3918 {
3919         struct fib6_table *table;
3920         struct fib6_info *rt;
3921         struct fib6_node *fn;
3922         int err = -ESRCH;
3923
3924         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3925         if (!table) {
3926                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3927                 return err;
3928         }
3929
3930         rcu_read_lock();
3931
3932         fn = fib6_locate(&table->tb6_root,
3933                          &cfg->fc_dst, cfg->fc_dst_len,
3934                          &cfg->fc_src, cfg->fc_src_len,
3935                          !(cfg->fc_flags & RTF_CACHE));
3936
3937         if (fn) {
3938                 for_each_fib6_node_rt_rcu(fn) {
3939                         struct fib6_nh *nh;
3940
3941                         if (rt->nh && cfg->fc_nh_id &&
3942                             rt->nh->id != cfg->fc_nh_id)
3943                                 continue;
3944
3945                         if (cfg->fc_flags & RTF_CACHE) {
3946                                 int rc = 0;
3947
3948                                 if (rt->nh) {
3949                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3950                                 } else if (cfg->fc_nh_id) {
3951                                         continue;
3952                                 } else {
3953                                         nh = rt->fib6_nh;
3954                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3955                                 }
3956                                 if (rc != -ESRCH) {
3957                                         rcu_read_unlock();
3958                                         return rc;
3959                                 }
3960                                 continue;
3961                         }
3962
3963                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3964                                 continue;
3965                         if (cfg->fc_protocol &&
3966                             cfg->fc_protocol != rt->fib6_protocol)
3967                                 continue;
3968
3969                         if (rt->nh) {
3970                                 if (!fib6_info_hold_safe(rt))
3971                                         continue;
3972                                 rcu_read_unlock();
3973
3974                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3975                         }
3976                         if (cfg->fc_nh_id)
3977                                 continue;
3978
3979                         nh = rt->fib6_nh;
3980                         if (cfg->fc_ifindex &&
3981                             (!nh->fib_nh_dev ||
3982                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3983                                 continue;
3984                         if (cfg->fc_flags & RTF_GATEWAY &&
3985                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3986                                 continue;
3987                         if (!fib6_info_hold_safe(rt))
3988                                 continue;
3989                         rcu_read_unlock();
3990
3991                         /* if gateway was specified only delete the one hop */
3992                         if (cfg->fc_flags & RTF_GATEWAY)
3993                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3994
3995                         return __ip6_del_rt_siblings(rt, cfg);
3996                 }
3997         }
3998         rcu_read_unlock();
3999
4000         return err;
4001 }
4002
4003 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4004 {
4005         struct netevent_redirect netevent;
4006         struct rt6_info *rt, *nrt = NULL;
4007         struct fib6_result res = {};
4008         struct ndisc_options ndopts;
4009         struct inet6_dev *in6_dev;
4010         struct neighbour *neigh;
4011         struct rd_msg *msg;
4012         int optlen, on_link;
4013         u8 *lladdr;
4014
4015         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4016         optlen -= sizeof(*msg);
4017
4018         if (optlen < 0) {
4019                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4020                 return;
4021         }
4022
4023         msg = (struct rd_msg *)icmp6_hdr(skb);
4024
4025         if (ipv6_addr_is_multicast(&msg->dest)) {
4026                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4027                 return;
4028         }
4029
4030         on_link = 0;
4031         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4032                 on_link = 1;
4033         } else if (ipv6_addr_type(&msg->target) !=
4034                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4035                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4036                 return;
4037         }
4038
4039         in6_dev = __in6_dev_get(skb->dev);
4040         if (!in6_dev)
4041                 return;
4042         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4043                 return;
4044
4045         /* RFC2461 8.1:
4046          *      The IP source address of the Redirect MUST be the same as the current
4047          *      first-hop router for the specified ICMP Destination Address.
4048          */
4049
4050         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4051                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4052                 return;
4053         }
4054
4055         lladdr = NULL;
4056         if (ndopts.nd_opts_tgt_lladdr) {
4057                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4058                                              skb->dev);
4059                 if (!lladdr) {
4060                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4061                         return;
4062                 }
4063         }
4064
4065         rt = (struct rt6_info *) dst;
4066         if (rt->rt6i_flags & RTF_REJECT) {
4067                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4068                 return;
4069         }
4070
4071         /* Redirect received -> path was valid.
4072          * Look, redirects are sent only in response to data packets,
4073          * so that this nexthop apparently is reachable. --ANK
4074          */
4075         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4076
4077         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4078         if (!neigh)
4079                 return;
4080
4081         /*
4082          *      We have finally decided to accept it.
4083          */
4084
4085         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4086                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4087                      NEIGH_UPDATE_F_OVERRIDE|
4088                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4089                                      NEIGH_UPDATE_F_ISROUTER)),
4090                      NDISC_REDIRECT, &ndopts);
4091
4092         rcu_read_lock();
4093         res.f6i = rcu_dereference(rt->from);
4094         if (!res.f6i)
4095                 goto out;
4096
4097         if (res.f6i->nh) {
4098                 struct fib6_nh_match_arg arg = {
4099                         .dev = dst->dev,
4100                         .gw = &rt->rt6i_gateway,
4101                 };
4102
4103                 nexthop_for_each_fib6_nh(res.f6i->nh,
4104                                          fib6_nh_find_match, &arg);
4105
4106                 /* fib6_info uses a nexthop that does not have fib6_nh
4107                  * using the dst->dev. Should be impossible
4108                  */
4109                 if (!arg.match)
4110                         goto out;
4111                 res.nh = arg.match;
4112         } else {
4113                 res.nh = res.f6i->fib6_nh;
4114         }
4115
4116         res.fib6_flags = res.f6i->fib6_flags;
4117         res.fib6_type = res.f6i->fib6_type;
4118         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4119         if (!nrt)
4120                 goto out;
4121
4122         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4123         if (on_link)
4124                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4125
4126         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4127
4128         /* rt6_insert_exception() will take care of duplicated exceptions */
4129         if (rt6_insert_exception(nrt, &res)) {
4130                 dst_release_immediate(&nrt->dst);
4131                 goto out;
4132         }
4133
4134         netevent.old = &rt->dst;
4135         netevent.new = &nrt->dst;
4136         netevent.daddr = &msg->dest;
4137         netevent.neigh = neigh;
4138         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4139
4140 out:
4141         rcu_read_unlock();
4142         neigh_release(neigh);
4143 }
4144
4145 #ifdef CONFIG_IPV6_ROUTE_INFO
4146 static struct fib6_info *rt6_get_route_info(struct net *net,
4147                                            const struct in6_addr *prefix, int prefixlen,
4148                                            const struct in6_addr *gwaddr,
4149                                            struct net_device *dev)
4150 {
4151         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4152         int ifindex = dev->ifindex;
4153         struct fib6_node *fn;
4154         struct fib6_info *rt = NULL;
4155         struct fib6_table *table;
4156
4157         table = fib6_get_table(net, tb_id);
4158         if (!table)
4159                 return NULL;
4160
4161         rcu_read_lock();
4162         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4163         if (!fn)
4164                 goto out;
4165
4166         for_each_fib6_node_rt_rcu(fn) {
4167                 /* these routes do not use nexthops */
4168                 if (rt->nh)
4169                         continue;
4170                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4171                         continue;
4172                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4173                     !rt->fib6_nh->fib_nh_gw_family)
4174                         continue;
4175                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4176                         continue;
4177                 if (!fib6_info_hold_safe(rt))
4178                         continue;
4179                 break;
4180         }
4181 out:
4182         rcu_read_unlock();
4183         return rt;
4184 }
4185
4186 static struct fib6_info *rt6_add_route_info(struct net *net,
4187                                            const struct in6_addr *prefix, int prefixlen,
4188                                            const struct in6_addr *gwaddr,
4189                                            struct net_device *dev,
4190                                            unsigned int pref)
4191 {
4192         struct fib6_config cfg = {
4193                 .fc_metric      = IP6_RT_PRIO_USER,
4194                 .fc_ifindex     = dev->ifindex,
4195                 .fc_dst_len     = prefixlen,
4196                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4197                                   RTF_UP | RTF_PREF(pref),
4198                 .fc_protocol = RTPROT_RA,
4199                 .fc_type = RTN_UNICAST,
4200                 .fc_nlinfo.portid = 0,
4201                 .fc_nlinfo.nlh = NULL,
4202                 .fc_nlinfo.nl_net = net,
4203         };
4204
4205         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4206         cfg.fc_dst = *prefix;
4207         cfg.fc_gateway = *gwaddr;
4208
4209         /* We should treat it as a default route if prefix length is 0. */
4210         if (!prefixlen)
4211                 cfg.fc_flags |= RTF_DEFAULT;
4212
4213         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4214
4215         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4216 }
4217 #endif
4218
4219 struct fib6_info *rt6_get_dflt_router(struct net *net,
4220                                      const struct in6_addr *addr,
4221                                      struct net_device *dev)
4222 {
4223         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4224         struct fib6_info *rt;
4225         struct fib6_table *table;
4226
4227         table = fib6_get_table(net, tb_id);
4228         if (!table)
4229                 return NULL;
4230
4231         rcu_read_lock();
4232         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4233                 struct fib6_nh *nh;
4234
4235                 /* RA routes do not use nexthops */
4236                 if (rt->nh)
4237                         continue;
4238
4239                 nh = rt->fib6_nh;
4240                 if (dev == nh->fib_nh_dev &&
4241                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4242                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4243                         break;
4244         }
4245         if (rt && !fib6_info_hold_safe(rt))
4246                 rt = NULL;
4247         rcu_read_unlock();
4248         return rt;
4249 }
4250
4251 struct fib6_info *rt6_add_dflt_router(struct net *net,
4252                                      const struct in6_addr *gwaddr,
4253                                      struct net_device *dev,
4254                                      unsigned int pref)
4255 {
4256         struct fib6_config cfg = {
4257                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4258                 .fc_metric      = IP6_RT_PRIO_USER,
4259                 .fc_ifindex     = dev->ifindex,
4260                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4261                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4262                 .fc_protocol = RTPROT_RA,
4263                 .fc_type = RTN_UNICAST,
4264                 .fc_nlinfo.portid = 0,
4265                 .fc_nlinfo.nlh = NULL,
4266                 .fc_nlinfo.nl_net = net,
4267         };
4268
4269         cfg.fc_gateway = *gwaddr;
4270
4271         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4272                 struct fib6_table *table;
4273
4274                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4275                 if (table)
4276                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4277         }
4278
4279         return rt6_get_dflt_router(net, gwaddr, dev);
4280 }
4281
4282 static void __rt6_purge_dflt_routers(struct net *net,
4283                                      struct fib6_table *table)
4284 {
4285         struct fib6_info *rt;
4286
4287 restart:
4288         rcu_read_lock();
4289         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4290                 struct net_device *dev = fib6_info_nh_dev(rt);
4291                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4292
4293                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4294                     (!idev || idev->cnf.accept_ra != 2) &&
4295                     fib6_info_hold_safe(rt)) {
4296                         rcu_read_unlock();
4297                         ip6_del_rt(net, rt, false);
4298                         goto restart;
4299                 }
4300         }
4301         rcu_read_unlock();
4302
4303         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4304 }
4305
4306 void rt6_purge_dflt_routers(struct net *net)
4307 {
4308         struct fib6_table *table;
4309         struct hlist_head *head;
4310         unsigned int h;
4311
4312         rcu_read_lock();
4313
4314         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4315                 head = &net->ipv6.fib_table_hash[h];
4316                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4317                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4318                                 __rt6_purge_dflt_routers(net, table);
4319                 }
4320         }
4321
4322         rcu_read_unlock();
4323 }
4324
4325 static void rtmsg_to_fib6_config(struct net *net,
4326                                  struct in6_rtmsg *rtmsg,
4327                                  struct fib6_config *cfg)
4328 {
4329         *cfg = (struct fib6_config){
4330                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4331                          : RT6_TABLE_MAIN,
4332                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4333                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4334                 .fc_expires = rtmsg->rtmsg_info,
4335                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4336                 .fc_src_len = rtmsg->rtmsg_src_len,
4337                 .fc_flags = rtmsg->rtmsg_flags,
4338                 .fc_type = rtmsg->rtmsg_type,
4339
4340                 .fc_nlinfo.nl_net = net,
4341
4342                 .fc_dst = rtmsg->rtmsg_dst,
4343                 .fc_src = rtmsg->rtmsg_src,
4344                 .fc_gateway = rtmsg->rtmsg_gateway,
4345         };
4346 }
4347
4348 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4349 {
4350         struct fib6_config cfg;
4351         int err;
4352
4353         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4354                 return -EINVAL;
4355         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4356                 return -EPERM;
4357
4358         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4359
4360         rtnl_lock();
4361         switch (cmd) {
4362         case SIOCADDRT:
4363                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4364                 break;
4365         case SIOCDELRT:
4366                 err = ip6_route_del(&cfg, NULL);
4367                 break;
4368         }
4369         rtnl_unlock();
4370         return err;
4371 }
4372
4373 /*
4374  *      Drop the packet on the floor
4375  */
4376
4377 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4378 {
4379         struct dst_entry *dst = skb_dst(skb);
4380         struct net *net = dev_net(dst->dev);
4381         struct inet6_dev *idev;
4382         int type;
4383
4384         if (netif_is_l3_master(skb->dev) &&
4385             dst->dev == net->loopback_dev)
4386                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4387         else
4388                 idev = ip6_dst_idev(dst);
4389
4390         switch (ipstats_mib_noroutes) {
4391         case IPSTATS_MIB_INNOROUTES:
4392                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4393                 if (type == IPV6_ADDR_ANY) {
4394                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4395                         break;
4396                 }
4397                 fallthrough;
4398         case IPSTATS_MIB_OUTNOROUTES:
4399                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4400                 break;
4401         }
4402
4403         /* Start over by dropping the dst for l3mdev case */
4404         if (netif_is_l3_master(skb->dev))
4405                 skb_dst_drop(skb);
4406
4407         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4408         kfree_skb(skb);
4409         return 0;
4410 }
4411
4412 static int ip6_pkt_discard(struct sk_buff *skb)
4413 {
4414         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4415 }
4416
4417 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4418 {
4419         skb->dev = skb_dst(skb)->dev;
4420         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4421 }
4422
4423 static int ip6_pkt_prohibit(struct sk_buff *skb)
4424 {
4425         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4426 }
4427
4428 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4429 {
4430         skb->dev = skb_dst(skb)->dev;
4431         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4432 }
4433
4434 /*
4435  *      Allocate a dst for local (unicast / anycast) address.
4436  */
4437
4438 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4439                                      struct inet6_dev *idev,
4440                                      const struct in6_addr *addr,
4441                                      bool anycast, gfp_t gfp_flags)
4442 {
4443         struct fib6_config cfg = {
4444                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4445                 .fc_ifindex = idev->dev->ifindex,
4446                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4447                 .fc_dst = *addr,
4448                 .fc_dst_len = 128,
4449                 .fc_protocol = RTPROT_KERNEL,
4450                 .fc_nlinfo.nl_net = net,
4451                 .fc_ignore_dev_down = true,
4452         };
4453         struct fib6_info *f6i;
4454
4455         if (anycast) {
4456                 cfg.fc_type = RTN_ANYCAST;
4457                 cfg.fc_flags |= RTF_ANYCAST;
4458         } else {
4459                 cfg.fc_type = RTN_LOCAL;
4460                 cfg.fc_flags |= RTF_LOCAL;
4461         }
4462
4463         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4464         if (!IS_ERR(f6i))
4465                 f6i->dst_nocount = true;
4466         return f6i;
4467 }
4468
4469 /* remove deleted ip from prefsrc entries */
4470 struct arg_dev_net_ip {
4471         struct net_device *dev;
4472         struct net *net;
4473         struct in6_addr *addr;
4474 };
4475
4476 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4477 {
4478         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4479         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4480         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4481
4482         if (!rt->nh &&
4483             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4484             rt != net->ipv6.fib6_null_entry &&
4485             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4486                 spin_lock_bh(&rt6_exception_lock);
4487                 /* remove prefsrc entry */
4488                 rt->fib6_prefsrc.plen = 0;
4489                 spin_unlock_bh(&rt6_exception_lock);
4490         }
4491         return 0;
4492 }
4493
4494 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4495 {
4496         struct net *net = dev_net(ifp->idev->dev);
4497         struct arg_dev_net_ip adni = {
4498                 .dev = ifp->idev->dev,
4499                 .net = net,
4500                 .addr = &ifp->addr,
4501         };
4502         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4503 }
4504
4505 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4506
4507 /* Remove routers and update dst entries when gateway turn into host. */
4508 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4509 {
4510         struct in6_addr *gateway = (struct in6_addr *)arg;
4511         struct fib6_nh *nh;
4512
4513         /* RA routes do not use nexthops */
4514         if (rt->nh)
4515                 return 0;
4516
4517         nh = rt->fib6_nh;
4518         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4519             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4520                 return -1;
4521
4522         /* Further clean up cached routes in exception table.
4523          * This is needed because cached route may have a different
4524          * gateway than its 'parent' in the case of an ip redirect.
4525          */
4526         fib6_nh_exceptions_clean_tohost(nh, gateway);
4527
4528         return 0;
4529 }
4530
4531 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4532 {
4533         fib6_clean_all(net, fib6_clean_tohost, gateway);
4534 }
4535
4536 struct arg_netdev_event {
4537         const struct net_device *dev;
4538         union {
4539                 unsigned char nh_flags;
4540                 unsigned long event;
4541         };
4542 };
4543
4544 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4545 {
4546         struct fib6_info *iter;
4547         struct fib6_node *fn;
4548
4549         fn = rcu_dereference_protected(rt->fib6_node,
4550                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4551         iter = rcu_dereference_protected(fn->leaf,
4552                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4553         while (iter) {
4554                 if (iter->fib6_metric == rt->fib6_metric &&
4555                     rt6_qualify_for_ecmp(iter))
4556                         return iter;
4557                 iter = rcu_dereference_protected(iter->fib6_next,
4558                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4559         }
4560
4561         return NULL;
4562 }
4563
4564 /* only called for fib entries with builtin fib6_nh */
4565 static bool rt6_is_dead(const struct fib6_info *rt)
4566 {
4567         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4568             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4569              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4570                 return true;
4571
4572         return false;
4573 }
4574
4575 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4576 {
4577         struct fib6_info *iter;
4578         int total = 0;
4579
4580         if (!rt6_is_dead(rt))
4581                 total += rt->fib6_nh->fib_nh_weight;
4582
4583         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4584                 if (!rt6_is_dead(iter))
4585                         total += iter->fib6_nh->fib_nh_weight;
4586         }
4587
4588         return total;
4589 }
4590
4591 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4592 {
4593         int upper_bound = -1;
4594
4595         if (!rt6_is_dead(rt)) {
4596                 *weight += rt->fib6_nh->fib_nh_weight;
4597                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4598                                                     total) - 1;
4599         }
4600         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4601 }
4602
4603 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4604 {
4605         struct fib6_info *iter;
4606         int weight = 0;
4607
4608         rt6_upper_bound_set(rt, &weight, total);
4609
4610         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4611                 rt6_upper_bound_set(iter, &weight, total);
4612 }
4613
4614 void rt6_multipath_rebalance(struct fib6_info *rt)
4615 {
4616         struct fib6_info *first;
4617         int total;
4618
4619         /* In case the entire multipath route was marked for flushing,
4620          * then there is no need to rebalance upon the removal of every
4621          * sibling route.
4622          */
4623         if (!rt->fib6_nsiblings || rt->should_flush)
4624                 return;
4625
4626         /* During lookup routes are evaluated in order, so we need to
4627          * make sure upper bounds are assigned from the first sibling
4628          * onwards.
4629          */
4630         first = rt6_multipath_first_sibling(rt);
4631         if (WARN_ON_ONCE(!first))
4632                 return;
4633
4634         total = rt6_multipath_total_weight(first);
4635         rt6_multipath_upper_bound_set(first, total);
4636 }
4637
4638 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4639 {
4640         const struct arg_netdev_event *arg = p_arg;
4641         struct net *net = dev_net(arg->dev);
4642
4643         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4644             rt->fib6_nh->fib_nh_dev == arg->dev) {
4645                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4646                 fib6_update_sernum_upto_root(net, rt);
4647                 rt6_multipath_rebalance(rt);
4648         }
4649
4650         return 0;
4651 }
4652
4653 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4654 {
4655         struct arg_netdev_event arg = {
4656                 .dev = dev,
4657                 {
4658                         .nh_flags = nh_flags,
4659                 },
4660         };
4661
4662         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4663                 arg.nh_flags |= RTNH_F_LINKDOWN;
4664
4665         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4666 }
4667
4668 /* only called for fib entries with inline fib6_nh */
4669 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4670                                    const struct net_device *dev)
4671 {
4672         struct fib6_info *iter;
4673
4674         if (rt->fib6_nh->fib_nh_dev == dev)
4675                 return true;
4676         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4677                 if (iter->fib6_nh->fib_nh_dev == dev)
4678                         return true;
4679
4680         return false;
4681 }
4682
4683 static void rt6_multipath_flush(struct fib6_info *rt)
4684 {
4685         struct fib6_info *iter;
4686
4687         rt->should_flush = 1;
4688         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4689                 iter->should_flush = 1;
4690 }
4691
4692 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4693                                              const struct net_device *down_dev)
4694 {
4695         struct fib6_info *iter;
4696         unsigned int dead = 0;
4697
4698         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4699             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4700                 dead++;
4701         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4702                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4703                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4704                         dead++;
4705
4706         return dead;
4707 }
4708
4709 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4710                                        const struct net_device *dev,
4711                                        unsigned char nh_flags)
4712 {
4713         struct fib6_info *iter;
4714
4715         if (rt->fib6_nh->fib_nh_dev == dev)
4716                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4717         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4718                 if (iter->fib6_nh->fib_nh_dev == dev)
4719                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4720 }
4721
4722 /* called with write lock held for table with rt */
4723 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4724 {
4725         const struct arg_netdev_event *arg = p_arg;
4726         const struct net_device *dev = arg->dev;
4727         struct net *net = dev_net(dev);
4728
4729         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4730                 return 0;
4731
4732         switch (arg->event) {
4733         case NETDEV_UNREGISTER:
4734                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4735         case NETDEV_DOWN:
4736                 if (rt->should_flush)
4737                         return -1;
4738                 if (!rt->fib6_nsiblings)
4739                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4740                 if (rt6_multipath_uses_dev(rt, dev)) {
4741                         unsigned int count;
4742
4743                         count = rt6_multipath_dead_count(rt, dev);
4744                         if (rt->fib6_nsiblings + 1 == count) {
4745                                 rt6_multipath_flush(rt);
4746                                 return -1;
4747                         }
4748                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4749                                                    RTNH_F_LINKDOWN);
4750                         fib6_update_sernum(net, rt);
4751                         rt6_multipath_rebalance(rt);
4752                 }
4753                 return -2;
4754         case NETDEV_CHANGE:
4755                 if (rt->fib6_nh->fib_nh_dev != dev ||
4756                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4757                         break;
4758                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4759                 rt6_multipath_rebalance(rt);
4760                 break;
4761         }
4762
4763         return 0;
4764 }
4765
4766 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4767 {
4768         struct arg_netdev_event arg = {
4769                 .dev = dev,
4770                 {
4771                         .event = event,
4772                 },
4773         };
4774         struct net *net = dev_net(dev);
4775
4776         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4777                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4778         else
4779                 fib6_clean_all(net, fib6_ifdown, &arg);
4780 }
4781
4782 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4783 {
4784         rt6_sync_down_dev(dev, event);
4785         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4786         neigh_ifdown(&nd_tbl, dev);
4787 }
4788
4789 struct rt6_mtu_change_arg {
4790         struct net_device *dev;
4791         unsigned int mtu;
4792         struct fib6_info *f6i;
4793 };
4794
4795 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4796 {
4797         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4798         struct fib6_info *f6i = arg->f6i;
4799
4800         /* For administrative MTU increase, there is no way to discover
4801          * IPv6 PMTU increase, so PMTU increase should be updated here.
4802          * Since RFC 1981 doesn't include administrative MTU increase
4803          * update PMTU increase is a MUST. (i.e. jumbo frame)
4804          */
4805         if (nh->fib_nh_dev == arg->dev) {
4806                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4807                 u32 mtu = f6i->fib6_pmtu;
4808
4809                 if (mtu >= arg->mtu ||
4810                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4811                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4812
4813                 spin_lock_bh(&rt6_exception_lock);
4814                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4815                 spin_unlock_bh(&rt6_exception_lock);
4816         }
4817
4818         return 0;
4819 }
4820
4821 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4822 {
4823         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4824         struct inet6_dev *idev;
4825
4826         /* In IPv6 pmtu discovery is not optional,
4827            so that RTAX_MTU lock cannot disable it.
4828            We still use this lock to block changes
4829            caused by addrconf/ndisc.
4830         */
4831
4832         idev = __in6_dev_get(arg->dev);
4833         if (!idev)
4834                 return 0;
4835
4836         if (fib6_metric_locked(f6i, RTAX_MTU))
4837                 return 0;
4838
4839         arg->f6i = f6i;
4840         if (f6i->nh) {
4841                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4842                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4843                                                 arg);
4844         }
4845
4846         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4847 }
4848
4849 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4850 {
4851         struct rt6_mtu_change_arg arg = {
4852                 .dev = dev,
4853                 .mtu = mtu,
4854         };
4855
4856         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4857 }
4858
4859 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4860         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4861         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4862         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4863         [RTA_OIF]               = { .type = NLA_U32 },
4864         [RTA_IIF]               = { .type = NLA_U32 },
4865         [RTA_PRIORITY]          = { .type = NLA_U32 },
4866         [RTA_METRICS]           = { .type = NLA_NESTED },
4867         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4868         [RTA_PREF]              = { .type = NLA_U8 },
4869         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4870         [RTA_ENCAP]             = { .type = NLA_NESTED },
4871         [RTA_EXPIRES]           = { .type = NLA_U32 },
4872         [RTA_UID]               = { .type = NLA_U32 },
4873         [RTA_MARK]              = { .type = NLA_U32 },
4874         [RTA_TABLE]             = { .type = NLA_U32 },
4875         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4876         [RTA_SPORT]             = { .type = NLA_U16 },
4877         [RTA_DPORT]             = { .type = NLA_U16 },
4878         [RTA_NH_ID]             = { .type = NLA_U32 },
4879 };
4880
4881 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4882                               struct fib6_config *cfg,
4883                               struct netlink_ext_ack *extack)
4884 {
4885         struct rtmsg *rtm;
4886         struct nlattr *tb[RTA_MAX+1];
4887         unsigned int pref;
4888         int err;
4889
4890         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4891                                      rtm_ipv6_policy, extack);
4892         if (err < 0)
4893                 goto errout;
4894
4895         err = -EINVAL;
4896         rtm = nlmsg_data(nlh);
4897
4898         *cfg = (struct fib6_config){
4899                 .fc_table = rtm->rtm_table,
4900                 .fc_dst_len = rtm->rtm_dst_len,
4901                 .fc_src_len = rtm->rtm_src_len,
4902                 .fc_flags = RTF_UP,
4903                 .fc_protocol = rtm->rtm_protocol,
4904                 .fc_type = rtm->rtm_type,
4905
4906                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4907                 .fc_nlinfo.nlh = nlh,
4908                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4909         };
4910
4911         if (rtm->rtm_type == RTN_UNREACHABLE ||
4912             rtm->rtm_type == RTN_BLACKHOLE ||
4913             rtm->rtm_type == RTN_PROHIBIT ||
4914             rtm->rtm_type == RTN_THROW)
4915                 cfg->fc_flags |= RTF_REJECT;
4916
4917         if (rtm->rtm_type == RTN_LOCAL)
4918                 cfg->fc_flags |= RTF_LOCAL;
4919
4920         if (rtm->rtm_flags & RTM_F_CLONED)
4921                 cfg->fc_flags |= RTF_CACHE;
4922
4923         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4924
4925         if (tb[RTA_NH_ID]) {
4926                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4927                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4928                         NL_SET_ERR_MSG(extack,
4929                                        "Nexthop specification and nexthop id are mutually exclusive");
4930                         goto errout;
4931                 }
4932                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4933         }
4934
4935         if (tb[RTA_GATEWAY]) {
4936                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4937                 cfg->fc_flags |= RTF_GATEWAY;
4938         }
4939         if (tb[RTA_VIA]) {
4940                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4941                 goto errout;
4942         }
4943
4944         if (tb[RTA_DST]) {
4945                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4946
4947                 if (nla_len(tb[RTA_DST]) < plen)
4948                         goto errout;
4949
4950                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4951         }
4952
4953         if (tb[RTA_SRC]) {
4954                 int plen = (rtm->rtm_src_len + 7) >> 3;
4955
4956                 if (nla_len(tb[RTA_SRC]) < plen)
4957                         goto errout;
4958
4959                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4960         }
4961
4962         if (tb[RTA_PREFSRC])
4963                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4964
4965         if (tb[RTA_OIF])
4966                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4967
4968         if (tb[RTA_PRIORITY])
4969                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4970
4971         if (tb[RTA_METRICS]) {
4972                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4973                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4974         }
4975
4976         if (tb[RTA_TABLE])
4977                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4978
4979         if (tb[RTA_MULTIPATH]) {
4980                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4981                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4982
4983                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4984                                                      cfg->fc_mp_len, extack);
4985                 if (err < 0)
4986                         goto errout;
4987         }
4988
4989         if (tb[RTA_PREF]) {
4990                 pref = nla_get_u8(tb[RTA_PREF]);
4991                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4992                     pref != ICMPV6_ROUTER_PREF_HIGH)
4993                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4994                 cfg->fc_flags |= RTF_PREF(pref);
4995         }
4996
4997         if (tb[RTA_ENCAP])
4998                 cfg->fc_encap = tb[RTA_ENCAP];
4999
5000         if (tb[RTA_ENCAP_TYPE]) {
5001                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5002
5003                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5004                 if (err < 0)
5005                         goto errout;
5006         }
5007
5008         if (tb[RTA_EXPIRES]) {
5009                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5010
5011                 if (addrconf_finite_timeout(timeout)) {
5012                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5013                         cfg->fc_flags |= RTF_EXPIRES;
5014                 }
5015         }
5016
5017         err = 0;
5018 errout:
5019         return err;
5020 }
5021
5022 struct rt6_nh {
5023         struct fib6_info *fib6_info;
5024         struct fib6_config r_cfg;
5025         struct list_head next;
5026 };
5027
5028 static int ip6_route_info_append(struct net *net,
5029                                  struct list_head *rt6_nh_list,
5030                                  struct fib6_info *rt,
5031                                  struct fib6_config *r_cfg)
5032 {
5033         struct rt6_nh *nh;
5034         int err = -EEXIST;
5035
5036         list_for_each_entry(nh, rt6_nh_list, next) {
5037                 /* check if fib6_info already exists */
5038                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5039                         return err;
5040         }
5041
5042         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5043         if (!nh)
5044                 return -ENOMEM;
5045         nh->fib6_info = rt;
5046         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5047         list_add_tail(&nh->next, rt6_nh_list);
5048
5049         return 0;
5050 }
5051
5052 static void ip6_route_mpath_notify(struct fib6_info *rt,
5053                                    struct fib6_info *rt_last,
5054                                    struct nl_info *info,
5055                                    __u16 nlflags)
5056 {
5057         /* if this is an APPEND route, then rt points to the first route
5058          * inserted and rt_last points to last route inserted. Userspace
5059          * wants a consistent dump of the route which starts at the first
5060          * nexthop. Since sibling routes are always added at the end of
5061          * the list, find the first sibling of the last route appended
5062          */
5063         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5064                 rt = list_first_entry(&rt_last->fib6_siblings,
5065                                       struct fib6_info,
5066                                       fib6_siblings);
5067         }
5068
5069         if (rt)
5070                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5071 }
5072
5073 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5074 {
5075         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5076         bool should_notify = false;
5077         struct fib6_info *leaf;
5078         struct fib6_node *fn;
5079
5080         rcu_read_lock();
5081         fn = rcu_dereference(rt->fib6_node);
5082         if (!fn)
5083                 goto out;
5084
5085         leaf = rcu_dereference(fn->leaf);
5086         if (!leaf)
5087                 goto out;
5088
5089         if (rt == leaf ||
5090             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5091              rt6_qualify_for_ecmp(leaf)))
5092                 should_notify = true;
5093 out:
5094         rcu_read_unlock();
5095
5096         return should_notify;
5097 }
5098
5099 static int ip6_route_multipath_add(struct fib6_config *cfg,
5100                                    struct netlink_ext_ack *extack)
5101 {
5102         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5103         struct nl_info *info = &cfg->fc_nlinfo;
5104         struct fib6_config r_cfg;
5105         struct rtnexthop *rtnh;
5106         struct fib6_info *rt;
5107         struct rt6_nh *err_nh;
5108         struct rt6_nh *nh, *nh_safe;
5109         __u16 nlflags;
5110         int remaining;
5111         int attrlen;
5112         int err = 1;
5113         int nhn = 0;
5114         int replace = (cfg->fc_nlinfo.nlh &&
5115                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5116         LIST_HEAD(rt6_nh_list);
5117
5118         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5119         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5120                 nlflags |= NLM_F_APPEND;
5121
5122         remaining = cfg->fc_mp_len;
5123         rtnh = (struct rtnexthop *)cfg->fc_mp;
5124
5125         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5126          * fib6_info structs per nexthop
5127          */
5128         while (rtnh_ok(rtnh, remaining)) {
5129                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5130                 if (rtnh->rtnh_ifindex)
5131                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5132
5133                 attrlen = rtnh_attrlen(rtnh);
5134                 if (attrlen > 0) {
5135                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5136
5137                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5138                         if (nla) {
5139                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5140                                 r_cfg.fc_flags |= RTF_GATEWAY;
5141                         }
5142                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5143                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5144                         if (nla)
5145                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5146                 }
5147
5148                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5149                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5150                 if (IS_ERR(rt)) {
5151                         err = PTR_ERR(rt);
5152                         rt = NULL;
5153                         goto cleanup;
5154                 }
5155                 if (!rt6_qualify_for_ecmp(rt)) {
5156                         err = -EINVAL;
5157                         NL_SET_ERR_MSG(extack,
5158                                        "Device only routes can not be added for IPv6 using the multipath API.");
5159                         fib6_info_release(rt);
5160                         goto cleanup;
5161                 }
5162
5163                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5164
5165                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5166                                             rt, &r_cfg);
5167                 if (err) {
5168                         fib6_info_release(rt);
5169                         goto cleanup;
5170                 }
5171
5172                 rtnh = rtnh_next(rtnh, &remaining);
5173         }
5174
5175         if (list_empty(&rt6_nh_list)) {
5176                 NL_SET_ERR_MSG(extack,
5177                                "Invalid nexthop configuration - no valid nexthops");
5178                 return -EINVAL;
5179         }
5180
5181         /* for add and replace send one notification with all nexthops.
5182          * Skip the notification in fib6_add_rt2node and send one with
5183          * the full route when done
5184          */
5185         info->skip_notify = 1;
5186
5187         /* For add and replace, send one notification with all nexthops. For
5188          * append, send one notification with all appended nexthops.
5189          */
5190         info->skip_notify_kernel = 1;
5191
5192         err_nh = NULL;
5193         list_for_each_entry(nh, &rt6_nh_list, next) {
5194                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5195                 fib6_info_release(nh->fib6_info);
5196
5197                 if (!err) {
5198                         /* save reference to last route successfully inserted */
5199                         rt_last = nh->fib6_info;
5200
5201                         /* save reference to first route for notification */
5202                         if (!rt_notif)
5203                                 rt_notif = nh->fib6_info;
5204                 }
5205
5206                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5207                 nh->fib6_info = NULL;
5208                 if (err) {
5209                         if (replace && nhn)
5210                                 NL_SET_ERR_MSG_MOD(extack,
5211                                                    "multipath route replace failed (check consistency of installed routes)");
5212                         err_nh = nh;
5213                         goto add_errout;
5214                 }
5215
5216                 /* Because each route is added like a single route we remove
5217                  * these flags after the first nexthop: if there is a collision,
5218                  * we have already failed to add the first nexthop:
5219                  * fib6_add_rt2node() has rejected it; when replacing, old
5220                  * nexthops have been replaced by first new, the rest should
5221                  * be added to it.
5222                  */
5223                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5224                                                      NLM_F_REPLACE);
5225                 cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5226                 nhn++;
5227         }
5228
5229         /* An in-kernel notification should only be sent in case the new
5230          * multipath route is added as the first route in the node, or if
5231          * it was appended to it. We pass 'rt_notif' since it is the first
5232          * sibling and might allow us to skip some checks in the replace case.
5233          */
5234         if (ip6_route_mpath_should_notify(rt_notif)) {
5235                 enum fib_event_type fib_event;
5236
5237                 if (rt_notif->fib6_nsiblings != nhn - 1)
5238                         fib_event = FIB_EVENT_ENTRY_APPEND;
5239                 else
5240                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5241
5242                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5243                                                           fib_event, rt_notif,
5244                                                           nhn - 1, extack);
5245                 if (err) {
5246                         /* Delete all the siblings that were just added */
5247                         err_nh = NULL;
5248                         goto add_errout;
5249                 }
5250         }
5251
5252         /* success ... tell user about new route */
5253         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5254         goto cleanup;
5255
5256 add_errout:
5257         /* send notification for routes that were added so that
5258          * the delete notifications sent by ip6_route_del are
5259          * coherent
5260          */
5261         if (rt_notif)
5262                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5263
5264         /* Delete routes that were already added */
5265         list_for_each_entry(nh, &rt6_nh_list, next) {
5266                 if (err_nh == nh)
5267                         break;
5268                 ip6_route_del(&nh->r_cfg, extack);
5269         }
5270
5271 cleanup:
5272         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5273                 if (nh->fib6_info)
5274                         fib6_info_release(nh->fib6_info);
5275                 list_del(&nh->next);
5276                 kfree(nh);
5277         }
5278
5279         return err;
5280 }
5281
5282 static int ip6_route_multipath_del(struct fib6_config *cfg,
5283                                    struct netlink_ext_ack *extack)
5284 {
5285         struct fib6_config r_cfg;
5286         struct rtnexthop *rtnh;
5287         int remaining;
5288         int attrlen;
5289         int err = 1, last_err = 0;
5290
5291         remaining = cfg->fc_mp_len;
5292         rtnh = (struct rtnexthop *)cfg->fc_mp;
5293
5294         /* Parse a Multipath Entry */
5295         while (rtnh_ok(rtnh, remaining)) {
5296                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5297                 if (rtnh->rtnh_ifindex)
5298                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5299
5300                 attrlen = rtnh_attrlen(rtnh);
5301                 if (attrlen > 0) {
5302                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5303
5304                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5305                         if (nla) {
5306                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5307                                 r_cfg.fc_flags |= RTF_GATEWAY;
5308                         }
5309                 }
5310                 err = ip6_route_del(&r_cfg, extack);
5311                 if (err)
5312                         last_err = err;
5313
5314                 rtnh = rtnh_next(rtnh, &remaining);
5315         }
5316
5317         return last_err;
5318 }
5319
5320 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5321                               struct netlink_ext_ack *extack)
5322 {
5323         struct fib6_config cfg;
5324         int err;
5325
5326         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5327         if (err < 0)
5328                 return err;
5329
5330         if (cfg.fc_nh_id &&
5331             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5332                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5333                 return -EINVAL;
5334         }
5335
5336         if (cfg.fc_mp)
5337                 return ip6_route_multipath_del(&cfg, extack);
5338         else {
5339                 cfg.fc_delete_all_nh = 1;
5340                 return ip6_route_del(&cfg, extack);
5341         }
5342 }
5343
5344 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5345                               struct netlink_ext_ack *extack)
5346 {
5347         struct fib6_config cfg;
5348         int err;
5349
5350         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5351         if (err < 0)
5352                 return err;
5353
5354         if (cfg.fc_metric == 0)
5355                 cfg.fc_metric = IP6_RT_PRIO_USER;
5356
5357         if (cfg.fc_mp)
5358                 return ip6_route_multipath_add(&cfg, extack);
5359         else
5360                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5361 }
5362
5363 /* add the overhead of this fib6_nh to nexthop_len */
5364 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5365 {
5366         int *nexthop_len = arg;
5367
5368         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5369                      + NLA_ALIGN(sizeof(struct rtnexthop))
5370                      + nla_total_size(16); /* RTA_GATEWAY */
5371
5372         if (nh->fib_nh_lws) {
5373                 /* RTA_ENCAP_TYPE */
5374                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5375                 /* RTA_ENCAP */
5376                 *nexthop_len += nla_total_size(2);
5377         }
5378
5379         return 0;
5380 }
5381
5382 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5383 {
5384         int nexthop_len;
5385
5386         if (f6i->nh) {
5387                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5388                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5389                                          &nexthop_len);
5390         } else {
5391                 struct fib6_nh *nh = f6i->fib6_nh;
5392
5393                 nexthop_len = 0;
5394                 if (f6i->fib6_nsiblings) {
5395                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5396                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5397                                     + nla_total_size(16) /* RTA_GATEWAY */
5398                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5399
5400                         nexthop_len *= f6i->fib6_nsiblings;
5401                 }
5402                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5403         }
5404
5405         return NLMSG_ALIGN(sizeof(struct rtmsg))
5406                + nla_total_size(16) /* RTA_SRC */
5407                + nla_total_size(16) /* RTA_DST */
5408                + nla_total_size(16) /* RTA_GATEWAY */
5409                + nla_total_size(16) /* RTA_PREFSRC */
5410                + nla_total_size(4) /* RTA_TABLE */
5411                + nla_total_size(4) /* RTA_IIF */
5412                + nla_total_size(4) /* RTA_OIF */
5413                + nla_total_size(4) /* RTA_PRIORITY */
5414                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5415                + nla_total_size(sizeof(struct rta_cacheinfo))
5416                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5417                + nla_total_size(1) /* RTA_PREF */
5418                + nexthop_len;
5419 }
5420
5421 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5422                                  unsigned char *flags)
5423 {
5424         if (nexthop_is_multipath(nh)) {
5425                 struct nlattr *mp;
5426
5427                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5428                 if (!mp)
5429                         goto nla_put_failure;
5430
5431                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5432                         goto nla_put_failure;
5433
5434                 nla_nest_end(skb, mp);
5435         } else {
5436                 struct fib6_nh *fib6_nh;
5437
5438                 fib6_nh = nexthop_fib6_nh(nh);
5439                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5440                                      flags, false) < 0)
5441                         goto nla_put_failure;
5442         }
5443
5444         return 0;
5445
5446 nla_put_failure:
5447         return -EMSGSIZE;
5448 }
5449
5450 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5451                          struct fib6_info *rt, struct dst_entry *dst,
5452                          struct in6_addr *dest, struct in6_addr *src,
5453                          int iif, int type, u32 portid, u32 seq,
5454                          unsigned int flags)
5455 {
5456         struct rt6_info *rt6 = (struct rt6_info *)dst;
5457         struct rt6key *rt6_dst, *rt6_src;
5458         u32 *pmetrics, table, rt6_flags;
5459         unsigned char nh_flags = 0;
5460         struct nlmsghdr *nlh;
5461         struct rtmsg *rtm;
5462         long expires = 0;
5463
5464         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5465         if (!nlh)
5466                 return -EMSGSIZE;
5467
5468         if (rt6) {
5469                 rt6_dst = &rt6->rt6i_dst;
5470                 rt6_src = &rt6->rt6i_src;
5471                 rt6_flags = rt6->rt6i_flags;
5472         } else {
5473                 rt6_dst = &rt->fib6_dst;
5474                 rt6_src = &rt->fib6_src;
5475                 rt6_flags = rt->fib6_flags;
5476         }
5477
5478         rtm = nlmsg_data(nlh);
5479         rtm->rtm_family = AF_INET6;
5480         rtm->rtm_dst_len = rt6_dst->plen;
5481         rtm->rtm_src_len = rt6_src->plen;
5482         rtm->rtm_tos = 0;
5483         if (rt->fib6_table)
5484                 table = rt->fib6_table->tb6_id;
5485         else
5486                 table = RT6_TABLE_UNSPEC;
5487         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5488         if (nla_put_u32(skb, RTA_TABLE, table))
5489                 goto nla_put_failure;
5490
5491         rtm->rtm_type = rt->fib6_type;
5492         rtm->rtm_flags = 0;
5493         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5494         rtm->rtm_protocol = rt->fib6_protocol;
5495
5496         if (rt6_flags & RTF_CACHE)
5497                 rtm->rtm_flags |= RTM_F_CLONED;
5498
5499         if (dest) {
5500                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5501                         goto nla_put_failure;
5502                 rtm->rtm_dst_len = 128;
5503         } else if (rtm->rtm_dst_len)
5504                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5505                         goto nla_put_failure;
5506 #ifdef CONFIG_IPV6_SUBTREES
5507         if (src) {
5508                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5509                         goto nla_put_failure;
5510                 rtm->rtm_src_len = 128;
5511         } else if (rtm->rtm_src_len &&
5512                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5513                 goto nla_put_failure;
5514 #endif
5515         if (iif) {
5516 #ifdef CONFIG_IPV6_MROUTE
5517                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5518                         int err = ip6mr_get_route(net, skb, rtm, portid);
5519
5520                         if (err == 0)
5521                                 return 0;
5522                         if (err < 0)
5523                                 goto nla_put_failure;
5524                 } else
5525 #endif
5526                         if (nla_put_u32(skb, RTA_IIF, iif))
5527                                 goto nla_put_failure;
5528         } else if (dest) {
5529                 struct in6_addr saddr_buf;
5530                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5531                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5532                         goto nla_put_failure;
5533         }
5534
5535         if (rt->fib6_prefsrc.plen) {
5536                 struct in6_addr saddr_buf;
5537                 saddr_buf = rt->fib6_prefsrc.addr;
5538                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5539                         goto nla_put_failure;
5540         }
5541
5542         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5543         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5544                 goto nla_put_failure;
5545
5546         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5547                 goto nla_put_failure;
5548
5549         /* For multipath routes, walk the siblings list and add
5550          * each as a nexthop within RTA_MULTIPATH.
5551          */
5552         if (rt6) {
5553                 if (rt6_flags & RTF_GATEWAY &&
5554                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5555                         goto nla_put_failure;
5556
5557                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5558                         goto nla_put_failure;
5559         } else if (rt->fib6_nsiblings) {
5560                 struct fib6_info *sibling, *next_sibling;
5561                 struct nlattr *mp;
5562
5563                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5564                 if (!mp)
5565                         goto nla_put_failure;
5566
5567                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5568                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5569                         goto nla_put_failure;
5570
5571                 list_for_each_entry_safe(sibling, next_sibling,
5572                                          &rt->fib6_siblings, fib6_siblings) {
5573                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5574                                             sibling->fib6_nh->fib_nh_weight,
5575                                             AF_INET6) < 0)
5576                                 goto nla_put_failure;
5577                 }
5578
5579                 nla_nest_end(skb, mp);
5580         } else if (rt->nh) {
5581                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5582                         goto nla_put_failure;
5583
5584                 if (nexthop_is_blackhole(rt->nh))
5585                         rtm->rtm_type = RTN_BLACKHOLE;
5586
5587                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5588                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5589                         goto nla_put_failure;
5590
5591                 rtm->rtm_flags |= nh_flags;
5592         } else {
5593                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5594                                      &nh_flags, false) < 0)
5595                         goto nla_put_failure;
5596
5597                 rtm->rtm_flags |= nh_flags;
5598         }
5599
5600         if (rt6_flags & RTF_EXPIRES) {
5601                 expires = dst ? dst->expires : rt->expires;
5602                 expires -= jiffies;
5603         }
5604
5605         if (!dst) {
5606                 if (rt->offload)
5607                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5608                 if (rt->trap)
5609                         rtm->rtm_flags |= RTM_F_TRAP;
5610         }
5611
5612         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5613                 goto nla_put_failure;
5614
5615         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5616                 goto nla_put_failure;
5617
5618
5619         nlmsg_end(skb, nlh);
5620         return 0;
5621
5622 nla_put_failure:
5623         nlmsg_cancel(skb, nlh);
5624         return -EMSGSIZE;
5625 }
5626
5627 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5628 {
5629         const struct net_device *dev = arg;
5630
5631         if (nh->fib_nh_dev == dev)
5632                 return 1;
5633
5634         return 0;
5635 }
5636
5637 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5638                                const struct net_device *dev)
5639 {
5640         if (f6i->nh) {
5641                 struct net_device *_dev = (struct net_device *)dev;
5642
5643                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5644                                                   fib6_info_nh_uses_dev,
5645                                                   _dev);
5646         }
5647
5648         if (f6i->fib6_nh->fib_nh_dev == dev)
5649                 return true;
5650
5651         if (f6i->fib6_nsiblings) {
5652                 struct fib6_info *sibling, *next_sibling;
5653
5654                 list_for_each_entry_safe(sibling, next_sibling,
5655                                          &f6i->fib6_siblings, fib6_siblings) {
5656                         if (sibling->fib6_nh->fib_nh_dev == dev)
5657                                 return true;
5658                 }
5659         }
5660
5661         return false;
5662 }
5663
5664 struct fib6_nh_exception_dump_walker {
5665         struct rt6_rtnl_dump_arg *dump;
5666         struct fib6_info *rt;
5667         unsigned int flags;
5668         unsigned int skip;
5669         unsigned int count;
5670 };
5671
5672 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5673 {
5674         struct fib6_nh_exception_dump_walker *w = arg;
5675         struct rt6_rtnl_dump_arg *dump = w->dump;
5676         struct rt6_exception_bucket *bucket;
5677         struct rt6_exception *rt6_ex;
5678         int i, err;
5679
5680         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5681         if (!bucket)
5682                 return 0;
5683
5684         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5685                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5686                         if (w->skip) {
5687                                 w->skip--;
5688                                 continue;
5689                         }
5690
5691                         /* Expiration of entries doesn't bump sernum, insertion
5692                          * does. Removal is triggered by insertion, so we can
5693                          * rely on the fact that if entries change between two
5694                          * partial dumps, this node is scanned again completely,
5695                          * see rt6_insert_exception() and fib6_dump_table().
5696                          *
5697                          * Count expired entries we go through as handled
5698                          * entries that we'll skip next time, in case of partial
5699                          * node dump. Otherwise, if entries expire meanwhile,
5700                          * we'll skip the wrong amount.
5701                          */
5702                         if (rt6_check_expired(rt6_ex->rt6i)) {
5703                                 w->count++;
5704                                 continue;
5705                         }
5706
5707                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5708                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5709                                             RTM_NEWROUTE,
5710                                             NETLINK_CB(dump->cb->skb).portid,
5711                                             dump->cb->nlh->nlmsg_seq, w->flags);
5712                         if (err)
5713                                 return err;
5714
5715                         w->count++;
5716                 }
5717                 bucket++;
5718         }
5719
5720         return 0;
5721 }
5722
5723 /* Return -1 if done with node, number of handled routes on partial dump */
5724 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5725 {
5726         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5727         struct fib_dump_filter *filter = &arg->filter;
5728         unsigned int flags = NLM_F_MULTI;
5729         struct net *net = arg->net;
5730         int count = 0;
5731
5732         if (rt == net->ipv6.fib6_null_entry)
5733                 return -1;
5734
5735         if ((filter->flags & RTM_F_PREFIX) &&
5736             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5737                 /* success since this is not a prefix route */
5738                 return -1;
5739         }
5740         if (filter->filter_set &&
5741             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5742              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5743              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5744                 return -1;
5745         }
5746
5747         if (filter->filter_set ||
5748             !filter->dump_routes || !filter->dump_exceptions) {
5749                 flags |= NLM_F_DUMP_FILTERED;
5750         }
5751
5752         if (filter->dump_routes) {
5753                 if (skip) {
5754                         skip--;
5755                 } else {
5756                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5757                                           0, RTM_NEWROUTE,
5758                                           NETLINK_CB(arg->cb->skb).portid,
5759                                           arg->cb->nlh->nlmsg_seq, flags)) {
5760                                 return 0;
5761                         }
5762                         count++;
5763                 }
5764         }
5765
5766         if (filter->dump_exceptions) {
5767                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5768                                                            .rt = rt,
5769                                                            .flags = flags,
5770                                                            .skip = skip,
5771                                                            .count = 0 };
5772                 int err;
5773
5774                 rcu_read_lock();
5775                 if (rt->nh) {
5776                         err = nexthop_for_each_fib6_nh(rt->nh,
5777                                                        rt6_nh_dump_exceptions,
5778                                                        &w);
5779                 } else {
5780                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5781                 }
5782                 rcu_read_unlock();
5783
5784                 if (err)
5785                         return count += w.count;
5786         }
5787
5788         return -1;
5789 }
5790
5791 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5792                                         const struct nlmsghdr *nlh,
5793                                         struct nlattr **tb,
5794                                         struct netlink_ext_ack *extack)
5795 {
5796         struct rtmsg *rtm;
5797         int i, err;
5798
5799         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5800                 NL_SET_ERR_MSG_MOD(extack,
5801                                    "Invalid header for get route request");
5802                 return -EINVAL;
5803         }
5804
5805         if (!netlink_strict_get_check(skb))
5806                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5807                                               rtm_ipv6_policy, extack);
5808
5809         rtm = nlmsg_data(nlh);
5810         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5811             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5812             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5813             rtm->rtm_type) {
5814                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5815                 return -EINVAL;
5816         }
5817         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5818                 NL_SET_ERR_MSG_MOD(extack,
5819                                    "Invalid flags for get route request");
5820                 return -EINVAL;
5821         }
5822
5823         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5824                                             rtm_ipv6_policy, extack);
5825         if (err)
5826                 return err;
5827
5828         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5829             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5830                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5831                 return -EINVAL;
5832         }
5833
5834         for (i = 0; i <= RTA_MAX; i++) {
5835                 if (!tb[i])
5836                         continue;
5837
5838                 switch (i) {
5839                 case RTA_SRC:
5840                 case RTA_DST:
5841                 case RTA_IIF:
5842                 case RTA_OIF:
5843                 case RTA_MARK:
5844                 case RTA_UID:
5845                 case RTA_SPORT:
5846                 case RTA_DPORT:
5847                 case RTA_IP_PROTO:
5848                         break;
5849                 default:
5850                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5851                         return -EINVAL;
5852                 }
5853         }
5854
5855         return 0;
5856 }
5857
5858 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5859                               struct netlink_ext_ack *extack)
5860 {
5861         struct net *net = sock_net(in_skb->sk);
5862         struct nlattr *tb[RTA_MAX+1];
5863         int err, iif = 0, oif = 0;
5864         struct fib6_info *from;
5865         struct dst_entry *dst;
5866         struct rt6_info *rt;
5867         struct sk_buff *skb;
5868         struct rtmsg *rtm;
5869         struct flowi6 fl6 = {};
5870         bool fibmatch;
5871
5872         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5873         if (err < 0)
5874                 goto errout;
5875
5876         err = -EINVAL;
5877         rtm = nlmsg_data(nlh);
5878         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5879         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5880
5881         if (tb[RTA_SRC]) {
5882                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5883                         goto errout;
5884
5885                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5886         }
5887
5888         if (tb[RTA_DST]) {
5889                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5890                         goto errout;
5891
5892                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5893         }
5894
5895         if (tb[RTA_IIF])
5896                 iif = nla_get_u32(tb[RTA_IIF]);
5897
5898         if (tb[RTA_OIF])
5899                 oif = nla_get_u32(tb[RTA_OIF]);
5900
5901         if (tb[RTA_MARK])
5902                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5903
5904         if (tb[RTA_UID])
5905                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5906                                            nla_get_u32(tb[RTA_UID]));
5907         else
5908                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5909
5910         if (tb[RTA_SPORT])
5911                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5912
5913         if (tb[RTA_DPORT])
5914                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5915
5916         if (tb[RTA_IP_PROTO]) {
5917                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5918                                                   &fl6.flowi6_proto, AF_INET6,
5919                                                   extack);
5920                 if (err)
5921                         goto errout;
5922         }
5923
5924         if (iif) {
5925                 struct net_device *dev;
5926                 int flags = 0;
5927
5928                 rcu_read_lock();
5929
5930                 dev = dev_get_by_index_rcu(net, iif);
5931                 if (!dev) {
5932                         rcu_read_unlock();
5933                         err = -ENODEV;
5934                         goto errout;
5935                 }
5936
5937                 fl6.flowi6_iif = iif;
5938
5939                 if (!ipv6_addr_any(&fl6.saddr))
5940                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5941
5942                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5943
5944                 rcu_read_unlock();
5945         } else {
5946                 fl6.flowi6_oif = oif;
5947
5948                 dst = ip6_route_output(net, NULL, &fl6);
5949         }
5950
5951
5952         rt = container_of(dst, struct rt6_info, dst);
5953         if (rt->dst.error) {
5954                 err = rt->dst.error;
5955                 ip6_rt_put(rt);
5956                 goto errout;
5957         }
5958
5959         if (rt == net->ipv6.ip6_null_entry) {
5960                 err = rt->dst.error;
5961                 ip6_rt_put(rt);
5962                 goto errout;
5963         }
5964
5965         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5966         if (!skb) {
5967                 ip6_rt_put(rt);
5968                 err = -ENOBUFS;
5969                 goto errout;
5970         }
5971
5972         skb_dst_set(skb, &rt->dst);
5973
5974         rcu_read_lock();
5975         from = rcu_dereference(rt->from);
5976         if (from) {
5977                 if (fibmatch)
5978                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5979                                             iif, RTM_NEWROUTE,
5980                                             NETLINK_CB(in_skb).portid,
5981                                             nlh->nlmsg_seq, 0);
5982                 else
5983                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5984                                             &fl6.saddr, iif, RTM_NEWROUTE,
5985                                             NETLINK_CB(in_skb).portid,
5986                                             nlh->nlmsg_seq, 0);
5987         } else {
5988                 err = -ENETUNREACH;
5989         }
5990         rcu_read_unlock();
5991
5992         if (err < 0) {
5993                 kfree_skb(skb);
5994                 goto errout;
5995         }
5996
5997         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5998 errout:
5999         return err;
6000 }
6001
6002 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6003                      unsigned int nlm_flags)
6004 {
6005         struct sk_buff *skb;
6006         struct net *net = info->nl_net;
6007         u32 seq;
6008         int err;
6009
6010         err = -ENOBUFS;
6011         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6012
6013         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6014         if (!skb)
6015                 goto errout;
6016
6017         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6018                             event, info->portid, seq, nlm_flags);
6019         if (err < 0) {
6020                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6021                 WARN_ON(err == -EMSGSIZE);
6022                 kfree_skb(skb);
6023                 goto errout;
6024         }
6025         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6026                     info->nlh, gfp_any());
6027         return;
6028 errout:
6029         if (err < 0)
6030                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6031 }
6032
6033 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6034                     struct nl_info *info)
6035 {
6036         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6037         struct sk_buff *skb;
6038         int err = -ENOBUFS;
6039
6040         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
6041          * is implemented and supported for nexthop objects
6042          */
6043         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
6044
6045         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6046         if (!skb)
6047                 goto errout;
6048
6049         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6050                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6051         if (err < 0) {
6052                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6053                 WARN_ON(err == -EMSGSIZE);
6054                 kfree_skb(skb);
6055                 goto errout;
6056         }
6057         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6058                     info->nlh, gfp_any());
6059         return;
6060 errout:
6061         if (err < 0)
6062                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6063 }
6064
6065 static int ip6_route_dev_notify(struct notifier_block *this,
6066                                 unsigned long event, void *ptr)
6067 {
6068         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6069         struct net *net = dev_net(dev);
6070
6071         if (!(dev->flags & IFF_LOOPBACK))
6072                 return NOTIFY_OK;
6073
6074         if (event == NETDEV_REGISTER) {
6075                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6076                 net->ipv6.ip6_null_entry->dst.dev = dev;
6077                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6078 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6079                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6080                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6081                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6082                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6083 #endif
6084          } else if (event == NETDEV_UNREGISTER &&
6085                     dev->reg_state != NETREG_UNREGISTERED) {
6086                 /* NETDEV_UNREGISTER could be fired for multiple times by
6087                  * netdev_wait_allrefs(). Make sure we only call this once.
6088                  */
6089                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6090 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6091                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6092                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6093 #endif
6094         }
6095
6096         return NOTIFY_OK;
6097 }
6098
6099 /*
6100  *      /proc
6101  */
6102
6103 #ifdef CONFIG_PROC_FS
6104 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6105 {
6106         struct net *net = (struct net *)seq->private;
6107         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6108                    net->ipv6.rt6_stats->fib_nodes,
6109                    net->ipv6.rt6_stats->fib_route_nodes,
6110                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6111                    net->ipv6.rt6_stats->fib_rt_entries,
6112                    net->ipv6.rt6_stats->fib_rt_cache,
6113                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6114                    net->ipv6.rt6_stats->fib_discarded_routes);
6115
6116         return 0;
6117 }
6118 #endif  /* CONFIG_PROC_FS */
6119
6120 #ifdef CONFIG_SYSCTL
6121
6122 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6123                               void *buffer, size_t *lenp, loff_t *ppos)
6124 {
6125         struct net *net;
6126         int delay;
6127         int ret;
6128         if (!write)
6129                 return -EINVAL;
6130
6131         net = (struct net *)ctl->extra1;
6132         delay = net->ipv6.sysctl.flush_delay;
6133         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6134         if (ret)
6135                 return ret;
6136
6137         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6138         return 0;
6139 }
6140
6141 static struct ctl_table ipv6_route_table_template[] = {
6142         {
6143                 .procname       =       "flush",
6144                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6145                 .maxlen         =       sizeof(int),
6146                 .mode           =       0200,
6147                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6148         },
6149         {
6150                 .procname       =       "gc_thresh",
6151                 .data           =       &ip6_dst_ops_template.gc_thresh,
6152                 .maxlen         =       sizeof(int),
6153                 .mode           =       0644,
6154                 .proc_handler   =       proc_dointvec,
6155         },
6156         {
6157                 .procname       =       "max_size",
6158                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6159                 .maxlen         =       sizeof(int),
6160                 .mode           =       0644,
6161                 .proc_handler   =       proc_dointvec,
6162         },
6163         {
6164                 .procname       =       "gc_min_interval",
6165                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6166                 .maxlen         =       sizeof(int),
6167                 .mode           =       0644,
6168                 .proc_handler   =       proc_dointvec_jiffies,
6169         },
6170         {
6171                 .procname       =       "gc_timeout",
6172                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6173                 .maxlen         =       sizeof(int),
6174                 .mode           =       0644,
6175                 .proc_handler   =       proc_dointvec_jiffies,
6176         },
6177         {
6178                 .procname       =       "gc_interval",
6179                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6180                 .maxlen         =       sizeof(int),
6181                 .mode           =       0644,
6182                 .proc_handler   =       proc_dointvec_jiffies,
6183         },
6184         {
6185                 .procname       =       "gc_elasticity",
6186                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6187                 .maxlen         =       sizeof(int),
6188                 .mode           =       0644,
6189                 .proc_handler   =       proc_dointvec,
6190         },
6191         {
6192                 .procname       =       "mtu_expires",
6193                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6194                 .maxlen         =       sizeof(int),
6195                 .mode           =       0644,
6196                 .proc_handler   =       proc_dointvec_jiffies,
6197         },
6198         {
6199                 .procname       =       "min_adv_mss",
6200                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6201                 .maxlen         =       sizeof(int),
6202                 .mode           =       0644,
6203                 .proc_handler   =       proc_dointvec,
6204         },
6205         {
6206                 .procname       =       "gc_min_interval_ms",
6207                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6208                 .maxlen         =       sizeof(int),
6209                 .mode           =       0644,
6210                 .proc_handler   =       proc_dointvec_ms_jiffies,
6211         },
6212         {
6213                 .procname       =       "skip_notify_on_dev_down",
6214                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6215                 .maxlen         =       sizeof(int),
6216                 .mode           =       0644,
6217                 .proc_handler   =       proc_dointvec_minmax,
6218                 .extra1         =       SYSCTL_ZERO,
6219                 .extra2         =       SYSCTL_ONE,
6220         },
6221         { }
6222 };
6223
6224 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6225 {
6226         struct ctl_table *table;
6227
6228         table = kmemdup(ipv6_route_table_template,
6229                         sizeof(ipv6_route_table_template),
6230                         GFP_KERNEL);
6231
6232         if (table) {
6233                 table[0].data = &net->ipv6.sysctl.flush_delay;
6234                 table[0].extra1 = net;
6235                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6236                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6237                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6238                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6239                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6240                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6241                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6242                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6243                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6244                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6245
6246                 /* Don't export sysctls to unprivileged users */
6247                 if (net->user_ns != &init_user_ns)
6248                         table[0].procname = NULL;
6249         }
6250
6251         return table;
6252 }
6253 #endif
6254
6255 static int __net_init ip6_route_net_init(struct net *net)
6256 {
6257         int ret = -ENOMEM;
6258
6259         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6260                sizeof(net->ipv6.ip6_dst_ops));
6261
6262         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6263                 goto out_ip6_dst_ops;
6264
6265         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6266         if (!net->ipv6.fib6_null_entry)
6267                 goto out_ip6_dst_entries;
6268         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6269                sizeof(*net->ipv6.fib6_null_entry));
6270
6271         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6272                                            sizeof(*net->ipv6.ip6_null_entry),
6273                                            GFP_KERNEL);
6274         if (!net->ipv6.ip6_null_entry)
6275                 goto out_fib6_null_entry;
6276         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6277         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6278                          ip6_template_metrics, true);
6279         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6280
6281 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6282         net->ipv6.fib6_has_custom_rules = false;
6283         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6284                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6285                                                GFP_KERNEL);
6286         if (!net->ipv6.ip6_prohibit_entry)
6287                 goto out_ip6_null_entry;
6288         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6289         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6290                          ip6_template_metrics, true);
6291         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6292
6293         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6294                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6295                                                GFP_KERNEL);
6296         if (!net->ipv6.ip6_blk_hole_entry)
6297                 goto out_ip6_prohibit_entry;
6298         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6299         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6300                          ip6_template_metrics, true);
6301         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6302 #ifdef CONFIG_IPV6_SUBTREES
6303         net->ipv6.fib6_routes_require_src = 0;
6304 #endif
6305 #endif
6306
6307         net->ipv6.sysctl.flush_delay = 0;
6308         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6309         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6310         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6311         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6312         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6313         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6314         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6315         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6316
6317         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6318
6319         ret = 0;
6320 out:
6321         return ret;
6322
6323 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6324 out_ip6_prohibit_entry:
6325         kfree(net->ipv6.ip6_prohibit_entry);
6326 out_ip6_null_entry:
6327         kfree(net->ipv6.ip6_null_entry);
6328 #endif
6329 out_fib6_null_entry:
6330         kfree(net->ipv6.fib6_null_entry);
6331 out_ip6_dst_entries:
6332         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6333 out_ip6_dst_ops:
6334         goto out;
6335 }
6336
6337 static void __net_exit ip6_route_net_exit(struct net *net)
6338 {
6339         kfree(net->ipv6.fib6_null_entry);
6340         kfree(net->ipv6.ip6_null_entry);
6341 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6342         kfree(net->ipv6.ip6_prohibit_entry);
6343         kfree(net->ipv6.ip6_blk_hole_entry);
6344 #endif
6345         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6346 }
6347
6348 static int __net_init ip6_route_net_init_late(struct net *net)
6349 {
6350 #ifdef CONFIG_PROC_FS
6351         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6352                         sizeof(struct ipv6_route_iter));
6353         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6354                         rt6_stats_seq_show, NULL);
6355 #endif
6356         return 0;
6357 }
6358
6359 static void __net_exit ip6_route_net_exit_late(struct net *net)
6360 {
6361 #ifdef CONFIG_PROC_FS
6362         remove_proc_entry("ipv6_route", net->proc_net);
6363         remove_proc_entry("rt6_stats", net->proc_net);
6364 #endif
6365 }
6366
6367 static struct pernet_operations ip6_route_net_ops = {
6368         .init = ip6_route_net_init,
6369         .exit = ip6_route_net_exit,
6370 };
6371
6372 static int __net_init ipv6_inetpeer_init(struct net *net)
6373 {
6374         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6375
6376         if (!bp)
6377                 return -ENOMEM;
6378         inet_peer_base_init(bp);
6379         net->ipv6.peers = bp;
6380         return 0;
6381 }
6382
6383 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6384 {
6385         struct inet_peer_base *bp = net->ipv6.peers;
6386
6387         net->ipv6.peers = NULL;
6388         inetpeer_invalidate_tree(bp);
6389         kfree(bp);
6390 }
6391
6392 static struct pernet_operations ipv6_inetpeer_ops = {
6393         .init   =       ipv6_inetpeer_init,
6394         .exit   =       ipv6_inetpeer_exit,
6395 };
6396
6397 static struct pernet_operations ip6_route_net_late_ops = {
6398         .init = ip6_route_net_init_late,
6399         .exit = ip6_route_net_exit_late,
6400 };
6401
6402 static struct notifier_block ip6_route_dev_notifier = {
6403         .notifier_call = ip6_route_dev_notify,
6404         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6405 };
6406
6407 void __init ip6_route_init_special_entries(void)
6408 {
6409         /* Registering of the loopback is done before this portion of code,
6410          * the loopback reference in rt6_info will not be taken, do it
6411          * manually for init_net */
6412         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6413         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6414         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6415   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6416         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6417         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6418         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6419         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6420   #endif
6421 }
6422
6423 #if IS_BUILTIN(CONFIG_IPV6)
6424 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6425 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6426
6427 BTF_ID_LIST(btf_fib6_info_id)
6428 BTF_ID(struct, fib6_info)
6429
6430 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6431         .seq_ops                = &ipv6_route_seq_ops,
6432         .init_seq_private       = bpf_iter_init_seq_net,
6433         .fini_seq_private       = bpf_iter_fini_seq_net,
6434         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6435 };
6436
6437 static struct bpf_iter_reg ipv6_route_reg_info = {
6438         .target                 = "ipv6_route",
6439         .ctx_arg_info_size      = 1,
6440         .ctx_arg_info           = {
6441                 { offsetof(struct bpf_iter__ipv6_route, rt),
6442                   PTR_TO_BTF_ID_OR_NULL },
6443         },
6444         .seq_info               = &ipv6_route_seq_info,
6445 };
6446
6447 static int __init bpf_iter_register(void)
6448 {
6449         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6450         return bpf_iter_reg_target(&ipv6_route_reg_info);
6451 }
6452
6453 static void bpf_iter_unregister(void)
6454 {
6455         bpf_iter_unreg_target(&ipv6_route_reg_info);
6456 }
6457 #endif
6458 #endif
6459
6460 int __init ip6_route_init(void)
6461 {
6462         int ret;
6463         int cpu;
6464
6465         ret = -ENOMEM;
6466         ip6_dst_ops_template.kmem_cachep =
6467                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6468                                   SLAB_HWCACHE_ALIGN, NULL);
6469         if (!ip6_dst_ops_template.kmem_cachep)
6470                 goto out;
6471
6472         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6473         if (ret)
6474                 goto out_kmem_cache;
6475
6476         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6477         if (ret)
6478                 goto out_dst_entries;
6479
6480         ret = register_pernet_subsys(&ip6_route_net_ops);
6481         if (ret)
6482                 goto out_register_inetpeer;
6483
6484         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6485
6486         ret = fib6_init();
6487         if (ret)
6488                 goto out_register_subsys;
6489
6490         ret = xfrm6_init();
6491         if (ret)
6492                 goto out_fib6_init;
6493
6494         ret = fib6_rules_init();
6495         if (ret)
6496                 goto xfrm6_init;
6497
6498         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6499         if (ret)
6500                 goto fib6_rules_init;
6501
6502         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6503                                    inet6_rtm_newroute, NULL, 0);
6504         if (ret < 0)
6505                 goto out_register_late_subsys;
6506
6507         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6508                                    inet6_rtm_delroute, NULL, 0);
6509         if (ret < 0)
6510                 goto out_register_late_subsys;
6511
6512         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6513                                    inet6_rtm_getroute, NULL,
6514                                    RTNL_FLAG_DOIT_UNLOCKED);
6515         if (ret < 0)
6516                 goto out_register_late_subsys;
6517
6518         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6519         if (ret)
6520                 goto out_register_late_subsys;
6521
6522 #if IS_BUILTIN(CONFIG_IPV6)
6523 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6524         ret = bpf_iter_register();
6525         if (ret)
6526                 goto out_register_late_subsys;
6527 #endif
6528 #endif
6529
6530         for_each_possible_cpu(cpu) {
6531                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6532
6533                 INIT_LIST_HEAD(&ul->head);
6534                 spin_lock_init(&ul->lock);
6535         }
6536
6537 out:
6538         return ret;
6539
6540 out_register_late_subsys:
6541         rtnl_unregister_all(PF_INET6);
6542         unregister_pernet_subsys(&ip6_route_net_late_ops);
6543 fib6_rules_init:
6544         fib6_rules_cleanup();
6545 xfrm6_init:
6546         xfrm6_fini();
6547 out_fib6_init:
6548         fib6_gc_cleanup();
6549 out_register_subsys:
6550         unregister_pernet_subsys(&ip6_route_net_ops);
6551 out_register_inetpeer:
6552         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6553 out_dst_entries:
6554         dst_entries_destroy(&ip6_dst_blackhole_ops);
6555 out_kmem_cache:
6556         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6557         goto out;
6558 }
6559
6560 void ip6_route_cleanup(void)
6561 {
6562 #if IS_BUILTIN(CONFIG_IPV6)
6563 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6564         bpf_iter_unregister();
6565 #endif
6566 #endif
6567         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6568         unregister_pernet_subsys(&ip6_route_net_late_ops);
6569         fib6_rules_cleanup();
6570         xfrm6_fini();
6571         fib6_gc_cleanup();
6572         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6573         unregister_pernet_subsys(&ip6_route_net_ops);
6574         dst_entries_destroy(&ip6_dst_blackhole_ops);
6575         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6576 }