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 <linux/siphash.h>
  45 #include <net/net_namespace.h>
  46 #include <net/snmp.h>
  47 #include <net/ipv6.h>
  48 #include <net/ip6_fib.h>
  49 #include <net/ip6_route.h>
  50 #include <net/ndisc.h>
  51 #include <net/addrconf.h>
  52 #include <net/tcp.h>
  53 #include <linux/rtnetlink.h>
  54 #include <net/dst.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/xfrm.h>
  57 #include <net/netevent.h>
  58 #include <net/netlink.h>
  59 #include <net/rtnh.h>
  60 #include <net/lwtunnel.h>
  61 #include <net/ip_tunnels.h>
  62 #include <net/l3mdev.h>
  63 #include <net/ip.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/btf_ids.h>
  66
  67 #ifdef CONFIG_SYSCTL
  68 #include <linux/sysctl.h>
  69 #endif
  70
  71 static int ip6_rt_type_to_error(u8 fib6_type);
  72
  73 #define CREATE_TRACE_POINTS
  74 #include <trace/events/fib6.h>
  75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  76 #undef CREATE_TRACE_POINTS
  77
  78 enum rt6_nud_state {
  79         RT6_NUD_FAIL_HARD = -3,
  80         RT6_NUD_FAIL_PROBE = -2,
  81         RT6_NUD_FAIL_DO_RR = -1,
  82         RT6_NUD_SUCCEED = 1
  83 };
  84
  85 INDIRECT_CALLABLE_SCOPE
  86 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  87 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  88 INDIRECT_CALLABLE_SCOPE
  89 unsigned int            ip6_mtu(const struct dst_entry *dst);
  90 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  91 static void             ip6_dst_destroy(struct dst_entry *);
  92 static void             ip6_dst_ifdown(struct dst_entry *,
  93                                        struct net_device *dev, int how);
  94 static int               ip6_dst_gc(struct dst_ops *ops);
  95
  96 static int              ip6_pkt_discard(struct sk_buff *skb);
  97 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  98 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  99 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
 100 static void             ip6_link_failure(struct sk_buff *skb);
 101 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 102                                            struct sk_buff *skb, u32 mtu,
 103                                            bool confirm_neigh);
 104 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 105                                         struct sk_buff *skb);
 106 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 107                            int strict);
 108 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 109 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 110                          struct fib6_info *rt, struct dst_entry *dst,
 111                          struct in6_addr *dest, struct in6_addr *src,
 112                          int iif, int type, u32 portid, u32 seq,
 113                          unsigned int flags);
 114 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 115                                            const struct in6_addr *daddr,
 116                                            const struct in6_addr *saddr);
 117
 118 #ifdef CONFIG_IPV6_ROUTE_INFO
 119 static struct fib6_info *rt6_add_route_info(struct net *net,
 120                                            const struct in6_addr *prefix, int prefixlen,
 121                                            const struct in6_addr *gwaddr,
 122                                            struct net_device *dev,
 123                                            unsigned int pref);
 124 static struct fib6_info *rt6_get_route_info(struct net *net,
 125                                            const struct in6_addr *prefix, int prefixlen,
 126                                            const struct in6_addr *gwaddr,
 127                                            struct net_device *dev);
 128 #endif
 129
 130 struct uncached_list {
 131         spinlock_t              lock;
 132         struct list_head        head;
 133 };
 134
 135 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 136
 137 void rt6_uncached_list_add(struct rt6_info *rt)
 138 {
 139         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 140
 141         rt->rt6i_uncached_list = ul;
 142
 143         spin_lock_bh(&ul->lock);
 144         list_add_tail(&rt->rt6i_uncached, &ul->head);
 145         spin_unlock_bh(&ul->lock);
 146 }
 147
 148 void rt6_uncached_list_del(struct rt6_info *rt)
 149 {
 150         if (!list_empty(&rt->rt6i_uncached)) {
 151                 struct uncached_list *ul = rt->rt6i_uncached_list;
 152                 struct net *net = dev_net(rt->dst.dev);
 153
 154                 spin_lock_bh(&ul->lock);
 155                 list_del(&rt->rt6i_uncached);
 156                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 157                 spin_unlock_bh(&ul->lock);
 158         }
 159 }
 160
 161 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 162 {
 163         struct net_device *loopback_dev = net->loopback_dev;
 164         int cpu;
 165
 166         if (dev == loopback_dev)
 167                 return;
 168
 169         for_each_possible_cpu(cpu) {
 170                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 171                 struct rt6_info *rt;
 172
 173                 spin_lock_bh(&ul->lock);
 174                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 175                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 176                         struct net_device *rt_dev = rt->dst.dev;
 177
 178                         if (rt_idev->dev == dev) {
 179                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 180                                 in6_dev_put(rt_idev);
 181                         }
 182
 183                         if (rt_dev == dev) {
 184                                 rt->dst.dev = blackhole_netdev;
 185                                 dev_hold(rt->dst.dev);
 186                                 dev_put(rt_dev);
 187                         }
 188                 }
 189                 spin_unlock_bh(&ul->lock);
 190         }
 191 }
 192
 193 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 194                                              struct sk_buff *skb,
 195                                              const void *daddr)
 196 {
 197         if (!ipv6_addr_any(p))
 198                 return (const void *) p;
 199         else if (skb)
 200                 return &ipv6_hdr(skb)->daddr;
 201         return daddr;
 202 }
 203
 204 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 205                                    struct net_device *dev,
 206                                    struct sk_buff *skb,
 207                                    const void *daddr)
 208 {
 209         struct neighbour *n;
 210
 211         daddr = choose_neigh_daddr(gw, skb, daddr);
 212         n = __ipv6_neigh_lookup(dev, daddr);
 213         if (n)
 214                 return n;
 215
 216         n = neigh_create(&nd_tbl, daddr, dev);
 217         return IS_ERR(n) ? NULL : n;
 218 }
 219
 220 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 221                                               struct sk_buff *skb,
 222                                               const void *daddr)
 223 {
 224         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 225
 226         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 227                                 dst->dev, skb, daddr);
 228 }
 229
 230 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 231 {
 232         struct net_device *dev = dst->dev;
 233         struct rt6_info *rt = (struct rt6_info *)dst;
 234
 235         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 236         if (!daddr)
 237                 return;
 238         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 239                 return;
 240         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 241                 return;
 242         __ipv6_confirm_neigh(dev, daddr);
 243 }
 244
 245 static struct dst_ops ip6_dst_ops_template = {
 246         .family                 =       AF_INET6,
 247         .gc                     =       ip6_dst_gc,
 248         .gc_thresh              =       1024,
 249         .check                  =       ip6_dst_check,
 250         .default_advmss         =       ip6_default_advmss,
 251         .mtu                    =       ip6_mtu,
 252         .cow_metrics            =       dst_cow_metrics_generic,
 253         .destroy                =       ip6_dst_destroy,
 254         .ifdown                 =       ip6_dst_ifdown,
 255         .negative_advice        =       ip6_negative_advice,
 256         .link_failure           =       ip6_link_failure,
 257         .update_pmtu            =       ip6_rt_update_pmtu,
 258         .redirect               =       rt6_do_redirect,
 259         .local_out              =       __ip6_local_out,
 260         .neigh_lookup           =       ip6_dst_neigh_lookup,
 261         .confirm_neigh          =       ip6_confirm_neigh,
 262 };
 263
 264 static struct dst_ops ip6_dst_blackhole_ops = {
 265         .family                 = AF_INET6,
 266         .default_advmss         = ip6_default_advmss,
 267         .neigh_lookup           = ip6_dst_neigh_lookup,
 268         .check                  = ip6_dst_check,
 269         .destroy                = ip6_dst_destroy,
 270         .cow_metrics            = dst_cow_metrics_generic,
 271         .update_pmtu            = dst_blackhole_update_pmtu,
 272         .redirect               = dst_blackhole_redirect,
 273         .mtu                    = dst_blackhole_mtu,
 274 };
 275
 276 static const u32 ip6_template_metrics[RTAX_MAX] = {
 277         [RTAX_HOPLIMIT - 1] = 0,
 278 };
 279
 280 static const struct fib6_info fib6_null_entry_template = {
 281         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 282         .fib6_protocol  = RTPROT_KERNEL,
 283         .fib6_metric    = ~(u32)0,
 284         .fib6_ref       = REFCOUNT_INIT(1),
 285         .fib6_type      = RTN_UNREACHABLE,
 286         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 287 };
 288
 289 static const struct rt6_info ip6_null_entry_template = {
 290         .dst = {
 291                 .__refcnt       = ATOMIC_INIT(1),
 292                 .__use          = 1,
 293                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 294                 .error          = -ENETUNREACH,
 295                 .input          = ip6_pkt_discard,
 296                 .output         = ip6_pkt_discard_out,
 297         },
 298         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 299 };
 300
 301 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 302
 303 static const struct rt6_info ip6_prohibit_entry_template = {
 304         .dst = {
 305                 .__refcnt       = ATOMIC_INIT(1),
 306                 .__use          = 1,
 307                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 308                 .error          = -EACCES,
 309                 .input          = ip6_pkt_prohibit,
 310                 .output         = ip6_pkt_prohibit_out,
 311         },
 312         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 313 };
 314
 315 static const struct rt6_info ip6_blk_hole_entry_template = {
 316         .dst = {
 317                 .__refcnt       = ATOMIC_INIT(1),
 318                 .__use          = 1,
 319                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 320                 .error          = -EINVAL,
 321                 .input          = dst_discard,
 322                 .output         = dst_discard_out,
 323         },
 324         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 325 };
 326
 327 #endif
 328
 329 static void rt6_info_init(struct rt6_info *rt)
 330 {
 331         struct dst_entry *dst = &rt->dst;
 332
 333         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 334         INIT_LIST_HEAD(&rt->rt6i_uncached);
 335 }
 336
 337 /* allocate dst with ip6_dst_ops */
 338 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 339                                int flags)
 340 {
 341         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 342                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 343
 344         if (rt) {
 345                 rt6_info_init(rt);
 346                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 347         }
 348
 349         return rt;
 350 }
 351 EXPORT_SYMBOL(ip6_dst_alloc);
 352
 353 static void ip6_dst_destroy(struct dst_entry *dst)
 354 {
 355         struct rt6_info *rt = (struct rt6_info *)dst;
 356         struct fib6_info *from;
 357         struct inet6_dev *idev;
 358
 359         ip_dst_metrics_put(dst);
 360         rt6_uncached_list_del(rt);
 361
 362         idev = rt->rt6i_idev;
 363         if (idev) {
 364                 rt->rt6i_idev = NULL;
 365                 in6_dev_put(idev);
 366         }
 367
 368         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 369         fib6_info_release(from);
 370 }
 371
 372 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 373                            int how)
 374 {
 375         struct rt6_info *rt = (struct rt6_info *)dst;
 376         struct inet6_dev *idev = rt->rt6i_idev;
 377         struct net_device *loopback_dev =
 378                 dev_net(dev)->loopback_dev;
 379
 380         if (idev && idev->dev != loopback_dev) {
 381                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 382                 if (loopback_idev) {
 383                         rt->rt6i_idev = loopback_idev;
 384                         in6_dev_put(idev);
 385                 }
 386         }
 387 }
 388
 389 static bool __rt6_check_expired(const struct rt6_info *rt)
 390 {
 391         if (rt->rt6i_flags & RTF_EXPIRES)
 392                 return time_after(jiffies, rt->dst.expires);
 393         else
 394                 return false;
 395 }
 396
 397 static bool rt6_check_expired(const struct rt6_info *rt)
 398 {
 399         struct fib6_info *from;
 400
 401         from = rcu_dereference(rt->from);
 402
 403         if (rt->rt6i_flags & RTF_EXPIRES) {
 404                 if (time_after(jiffies, rt->dst.expires))
 405                         return true;
 406         } else if (from) {
 407                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 408                         fib6_check_expired(from);
 409         }
 410         return false;
 411 }
 412
 413 void fib6_select_path(const struct net *net, struct fib6_result *res,
 414                       struct flowi6 *fl6, int oif, bool have_oif_match,
 415                       const struct sk_buff *skb, int strict)
 416 {
 417         struct fib6_info *sibling, *next_sibling;
 418         struct fib6_info *match = res->f6i;
 419
 420         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 421                 goto out;
 422
 423         if (match->nh && have_oif_match && res->nh)
 424                 return;
 425
 426         /* We might have already computed the hash for ICMPv6 errors. In such
 427          * case it will always be non-zero. Otherwise now is the time to do it.
 428          */
 429         if (!fl6->mp_hash &&
 430             (!match->nh || nexthop_is_multipath(match->nh)))
 431                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 432
 433         if (unlikely(match->nh)) {
 434                 nexthop_path_fib6_result(res, fl6->mp_hash);
 435                 return;
 436         }
 437
 438         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 439                 goto out;
 440
 441         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 442                                  fib6_siblings) {
 443                 const struct fib6_nh *nh = sibling->fib6_nh;
 444                 int nh_upper_bound;
 445
 446                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 447                 if (fl6->mp_hash > nh_upper_bound)
 448                         continue;
 449                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 450                         break;
 451                 match = sibling;
 452                 break;
 453         }
 454
 455 out:
 456         res->f6i = match;
 457         res->nh = match->fib6_nh;
 458 }
 459
 460 /*
 461  *      Route lookup. rcu_read_lock() should be held.
 462  */
 463
 464 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 465                                const struct in6_addr *saddr, int oif, int flags)
 466 {
 467         const struct net_device *dev;
 468
 469         if (nh->fib_nh_flags & RTNH_F_DEAD)
 470                 return false;
 471
 472         dev = nh->fib_nh_dev;
 473         if (oif) {
 474                 if (dev->ifindex == oif)
 475                         return true;
 476         } else {
 477                 if (ipv6_chk_addr(net, saddr, dev,
 478                                   flags & RT6_LOOKUP_F_IFACE))
 479                         return true;
 480         }
 481
 482         return false;
 483 }
 484
 485 struct fib6_nh_dm_arg {
 486         struct net              *net;
 487         const struct in6_addr   *saddr;
 488         int                     oif;
 489         int                     flags;
 490         struct fib6_nh          *nh;
 491 };
 492
 493 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 494 {
 495         struct fib6_nh_dm_arg *arg = _arg;
 496
 497         arg->nh = nh;
 498         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 499                                   arg->flags);
 500 }
 501
 502 /* returns fib6_nh from nexthop or NULL */
 503 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 504                                         struct fib6_result *res,
 505                                         const struct in6_addr *saddr,
 506                                         int oif, int flags)
 507 {
 508         struct fib6_nh_dm_arg arg = {
 509                 .net   = net,
 510                 .saddr = saddr,
 511                 .oif   = oif,
 512                 .flags = flags,
 513         };
 514
 515         if (nexthop_is_blackhole(nh))
 516                 return NULL;
 517
 518         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 519                 return arg.nh;
 520
 521         return NULL;
 522 }
 523
 524 static void rt6_device_match(struct net *net, struct fib6_result *res,
 525                              const struct in6_addr *saddr, int oif, int flags)
 526 {
 527         struct fib6_info *f6i = res->f6i;
 528         struct fib6_info *spf6i;
 529         struct fib6_nh *nh;
 530
 531         if (!oif && ipv6_addr_any(saddr)) {
 532                 if (unlikely(f6i->nh)) {
 533                         nh = nexthop_fib6_nh(f6i->nh);
 534                         if (nexthop_is_blackhole(f6i->nh))
 535                                 goto out_blackhole;
 536                 } else {
 537                         nh = f6i->fib6_nh;
 538                 }
 539                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 540                         goto out;
 541         }
 542
 543         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 544                 bool matched = false;
 545
 546                 if (unlikely(spf6i->nh)) {
 547                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 548                                               oif, flags);
 549                         if (nh)
 550                                 matched = true;
 551                 } else {
 552                         nh = spf6i->fib6_nh;
 553                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 554                                 matched = true;
 555                 }
 556                 if (matched) {
 557                         res->f6i = spf6i;
 558                         goto out;
 559                 }
 560         }
 561
 562         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 563                 res->f6i = net->ipv6.fib6_null_entry;
 564                 nh = res->f6i->fib6_nh;
 565                 goto out;
 566         }
 567
 568         if (unlikely(f6i->nh)) {
 569                 nh = nexthop_fib6_nh(f6i->nh);
 570                 if (nexthop_is_blackhole(f6i->nh))
 571                         goto out_blackhole;
 572         } else {
 573                 nh = f6i->fib6_nh;
 574         }
 575
 576         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 577                 res->f6i = net->ipv6.fib6_null_entry;
 578                 nh = res->f6i->fib6_nh;
 579         }
 580 out:
 581         res->nh = nh;
 582         res->fib6_type = res->f6i->fib6_type;
 583         res->fib6_flags = res->f6i->fib6_flags;
 584         return;
 585
 586 out_blackhole:
 587         res->fib6_flags |= RTF_REJECT;
 588         res->fib6_type = RTN_BLACKHOLE;
 589         res->nh = nh;
 590 }
 591
 592 #ifdef CONFIG_IPV6_ROUTER_PREF
 593 struct __rt6_probe_work {
 594         struct work_struct work;
 595         struct in6_addr target;
 596         struct net_device *dev;
 597 };
 598
 599 static void rt6_probe_deferred(struct work_struct *w)
 600 {
 601         struct in6_addr mcaddr;
 602         struct __rt6_probe_work *work =
 603                 container_of(w, struct __rt6_probe_work, work);
 604
 605         addrconf_addr_solict_mult(&work->target, &mcaddr);
 606         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 607         dev_put(work->dev);
 608         kfree(work);
 609 }
 610
 611 static void rt6_probe(struct fib6_nh *fib6_nh)
 612 {
 613         struct __rt6_probe_work *work = NULL;
 614         const struct in6_addr *nh_gw;
 615         unsigned long last_probe;
 616         struct neighbour *neigh;
 617         struct net_device *dev;
 618         struct inet6_dev *idev;
 619
 620         /*
 621          * Okay, this does not seem to be appropriate
 622          * for now, however, we need to check if it
 623          * is really so; aka Router Reachability Probing.
 624          *
 625          * Router Reachability Probe MUST be rate-limited
 626          * to no more than one per minute.
 627          */
 628         if (!fib6_nh->fib_nh_gw_family)
 629                 return;
 630
 631         nh_gw = &fib6_nh->fib_nh_gw6;
 632         dev = fib6_nh->fib_nh_dev;
 633         rcu_read_lock_bh();
 634         last_probe = READ_ONCE(fib6_nh->last_probe);
 635         idev = __in6_dev_get(dev);
 636         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 637         if (neigh) {
 638                 if (neigh->nud_state & NUD_VALID)
 639                         goto out;
 640
 641                 write_lock(&neigh->lock);
 642                 if (!(neigh->nud_state & NUD_VALID) &&
 643                     time_after(jiffies,
 644                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 645                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 646                         if (work)
 647                                 __neigh_set_probe_once(neigh);
 648                 }
 649                 write_unlock(&neigh->lock);
 650         } else if (time_after(jiffies, last_probe +
 651                                        idev->cnf.rtr_probe_interval)) {
 652                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 653         }
 654
 655         if (!work || cmpxchg(&fib6_nh->last_probe,
 656                              last_probe, jiffies) != last_probe) {
 657                 kfree(work);
 658         } else {
 659                 INIT_WORK(&work->work, rt6_probe_deferred);
 660                 work->target = *nh_gw;
 661                 dev_hold(dev);
 662                 work->dev = dev;
 663                 schedule_work(&work->work);
 664         }
 665
 666 out:
 667         rcu_read_unlock_bh();
 668 }
 669 #else
 670 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 671 {
 672 }
 673 #endif
 674
 675 /*
 676  * Default Router Selection (RFC 2461 6.3.6)
 677  */
 678 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 679 {
 680         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 681         struct neighbour *neigh;
 682
 683         rcu_read_lock_bh();
 684         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 685                                           &fib6_nh->fib_nh_gw6);
 686         if (neigh) {
 687                 read_lock(&neigh->lock);
 688                 if (neigh->nud_state & NUD_VALID)
 689                         ret = RT6_NUD_SUCCEED;
 690 #ifdef CONFIG_IPV6_ROUTER_PREF
 691                 else if (!(neigh->nud_state & NUD_FAILED))
 692                         ret = RT6_NUD_SUCCEED;
 693                 else
 694                         ret = RT6_NUD_FAIL_PROBE;
 695 #endif
 696                 read_unlock(&neigh->lock);
 697         } else {
 698                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 699                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 700         }
 701         rcu_read_unlock_bh();
 702
 703         return ret;
 704 }
 705
 706 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 707                            int strict)
 708 {
 709         int m = 0;
 710
 711         if (!oif || nh->fib_nh_dev->ifindex == oif)
 712                 m = 2;
 713
 714         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 715                 return RT6_NUD_FAIL_HARD;
 716 #ifdef CONFIG_IPV6_ROUTER_PREF
 717         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 718 #endif
 719         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 720             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 721                 int n = rt6_check_neigh(nh);
 722                 if (n < 0)
 723                         return n;
 724         }
 725         return m;
 726 }
 727
 728 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 729                        int oif, int strict, int *mpri, bool *do_rr)
 730 {
 731         bool match_do_rr = false;
 732         bool rc = false;
 733         int m;
 734
 735         if (nh->fib_nh_flags & RTNH_F_DEAD)
 736                 goto out;
 737
 738         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 739             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 740             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 741                 goto out;
 742
 743         m = rt6_score_route(nh, fib6_flags, oif, strict);
 744         if (m == RT6_NUD_FAIL_DO_RR) {
 745                 match_do_rr = true;
 746                 m = 0; /* lowest valid score */
 747         } else if (m == RT6_NUD_FAIL_HARD) {
 748                 goto out;
 749         }
 750
 751         if (strict & RT6_LOOKUP_F_REACHABLE)
 752                 rt6_probe(nh);
 753
 754         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 755         if (m > *mpri) {
 756                 *do_rr = match_do_rr;
 757                 *mpri = m;
 758                 rc = true;
 759         }
 760 out:
 761         return rc;
 762 }
 763
 764 struct fib6_nh_frl_arg {
 765         u32             flags;
 766         int             oif;
 767         int             strict;
 768         int             *mpri;
 769         bool            *do_rr;
 770         struct fib6_nh  *nh;
 771 };
 772
 773 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 774 {
 775         struct fib6_nh_frl_arg *arg = _arg;
 776
 777         arg->nh = nh;
 778         return find_match(nh, arg->flags, arg->oif, arg->strict,
 779                           arg->mpri, arg->do_rr);
 780 }
 781
 782 static void __find_rr_leaf(struct fib6_info *f6i_start,
 783                            struct fib6_info *nomatch, u32 metric,
 784                            struct fib6_result *res, struct fib6_info **cont,
 785                            int oif, int strict, bool *do_rr, int *mpri)
 786 {
 787         struct fib6_info *f6i;
 788
 789         for (f6i = f6i_start;
 790              f6i && f6i != nomatch;
 791              f6i = rcu_dereference(f6i->fib6_next)) {
 792                 bool matched = false;
 793                 struct fib6_nh *nh;
 794
 795                 if (cont && f6i->fib6_metric != metric) {
 796                         *cont = f6i;
 797                         return;
 798                 }
 799
 800                 if (fib6_check_expired(f6i))
 801                         continue;
 802
 803                 if (unlikely(f6i->nh)) {
 804                         struct fib6_nh_frl_arg arg = {
 805                                 .flags  = f6i->fib6_flags,
 806                                 .oif    = oif,
 807                                 .strict = strict,
 808                                 .mpri   = mpri,
 809                                 .do_rr  = do_rr
 810                         };
 811
 812                         if (nexthop_is_blackhole(f6i->nh)) {
 813                                 res->fib6_flags = RTF_REJECT;
 814                                 res->fib6_type = RTN_BLACKHOLE;
 815                                 res->f6i = f6i;
 816                                 res->nh = nexthop_fib6_nh(f6i->nh);
 817                                 return;
 818                         }
 819                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 820                                                      &arg)) {
 821                                 matched = true;
 822                                 nh = arg.nh;
 823                         }
 824                 } else {
 825                         nh = f6i->fib6_nh;
 826                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 827                                        mpri, do_rr))
 828                                 matched = true;
 829                 }
 830                 if (matched) {
 831                         res->f6i = f6i;
 832                         res->nh = nh;
 833                         res->fib6_flags = f6i->fib6_flags;
 834                         res->fib6_type = f6i->fib6_type;
 835                 }
 836         }
 837 }
 838
 839 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 840                          struct fib6_info *rr_head, int oif, int strict,
 841                          bool *do_rr, struct fib6_result *res)
 842 {
 843         u32 metric = rr_head->fib6_metric;
 844         struct fib6_info *cont = NULL;
 845         int mpri = -1;
 846
 847         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 848                        oif, strict, do_rr, &mpri);
 849
 850         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 851                        oif, strict, do_rr, &mpri);
 852
 853         if (res->f6i || !cont)
 854                 return;
 855
 856         __find_rr_leaf(cont, NULL, metric, res, NULL,
 857                        oif, strict, do_rr, &mpri);
 858 }
 859
 860 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 861                        struct fib6_result *res, int strict)
 862 {
 863         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 864         struct fib6_info *rt0;
 865         bool do_rr = false;
 866         int key_plen;
 867
 868         /* make sure this function or its helpers sets f6i */
 869         res->f6i = NULL;
 870
 871         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 872                 goto out;
 873
 874         rt0 = rcu_dereference(fn->rr_ptr);
 875         if (!rt0)
 876                 rt0 = leaf;
 877
 878         /* Double check to make sure fn is not an intermediate node
 879          * and fn->leaf does not points to its child's leaf
 880          * (This might happen if all routes under fn are deleted from
 881          * the tree and fib6_repair_tree() is called on the node.)
 882          */
 883         key_plen = rt0->fib6_dst.plen;
 884 #ifdef CONFIG_IPV6_SUBTREES
 885         if (rt0->fib6_src.plen)
 886                 key_plen = rt0->fib6_src.plen;
 887 #endif
 888         if (fn->fn_bit != key_plen)
 889                 goto out;
 890
 891         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 892         if (do_rr) {
 893                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 894
 895                 /* no entries matched; do round-robin */
 896                 if (!next || next->fib6_metric != rt0->fib6_metric)
 897                         next = leaf;
 898
 899                 if (next != rt0) {
 900                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 901                         /* make sure next is not being deleted from the tree */
 902                         if (next->fib6_node)
 903                                 rcu_assign_pointer(fn->rr_ptr, next);
 904                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 905                 }
 906         }
 907
 908 out:
 909         if (!res->f6i) {
 910                 res->f6i = net->ipv6.fib6_null_entry;
 911                 res->nh = res->f6i->fib6_nh;
 912                 res->fib6_flags = res->f6i->fib6_flags;
 913                 res->fib6_type = res->f6i->fib6_type;
 914         }
 915 }
 916
 917 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 918 {
 919         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 920                res->nh->fib_nh_gw_family;
 921 }
 922
 923 #ifdef CONFIG_IPV6_ROUTE_INFO
 924 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 925                   const struct in6_addr *gwaddr)
 926 {
 927         struct net *net = dev_net(dev);
 928         struct route_info *rinfo = (struct route_info *) opt;
 929         struct in6_addr prefix_buf, *prefix;
 930         unsigned int pref;
 931         unsigned long lifetime;
 932         struct fib6_info *rt;
 933
 934         if (len < sizeof(struct route_info)) {
 935                 return -EINVAL;
 936         }
 937
 938         /* Sanity check for prefix_len and length */
 939         if (rinfo->length > 3) {
 940                 return -EINVAL;
 941         } else if (rinfo->prefix_len > 128) {
 942                 return -EINVAL;
 943         } else if (rinfo->prefix_len > 64) {
 944                 if (rinfo->length < 2) {
 945                         return -EINVAL;
 946                 }
 947         } else if (rinfo->prefix_len > 0) {
 948                 if (rinfo->length < 1) {
 949                         return -EINVAL;
 950                 }
 951         }
 952
 953         pref = rinfo->route_pref;
 954         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 955                 return -EINVAL;
 956
 957         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 958
 959         if (rinfo->length == 3)
 960                 prefix = (struct in6_addr *)rinfo->prefix;
 961         else {
 962                 /* this function is safe */
 963                 ipv6_addr_prefix(&prefix_buf,
 964                                  (struct in6_addr *)rinfo->prefix,
 965                                  rinfo->prefix_len);
 966                 prefix = &prefix_buf;
 967         }
 968
 969         if (rinfo->prefix_len == 0)
 970                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 971         else
 972                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 973                                         gwaddr, dev);
 974
 975         if (rt && !lifetime) {
 976                 ip6_del_rt(net, rt, false);
 977                 rt = NULL;
 978         }
 979
 980         if (!rt && lifetime)
 981                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 982                                         dev, pref);
 983         else if (rt)
 984                 rt->fib6_flags = RTF_ROUTEINFO |
 985                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 986
 987         if (rt) {
 988                 if (!addrconf_finite_timeout(lifetime))
 989                         fib6_clean_expires(rt);
 990                 else
 991                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 992
 993                 fib6_info_release(rt);
 994         }
 995         return 0;
 996 }
 997 #endif
 998
 999 /*
1000  *      Misc support functions
1001  */
1002
1003 /* called with rcu_lock held */
1004 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1005 {
1006         struct net_device *dev = res->nh->fib_nh_dev;
1007
1008         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1009                 /* for copies of local routes, dst->dev needs to be the
1010                  * device if it is a master device, the master device if
1011                  * device is enslaved, and the loopback as the default
1012                  */
1013                 if (netif_is_l3_slave(dev) &&
1014                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1015                         dev = l3mdev_master_dev_rcu(dev);
1016                 else if (!netif_is_l3_master(dev))
1017                         dev = dev_net(dev)->loopback_dev;
1018                 /* last case is netif_is_l3_master(dev) is true in which
1019                  * case we want dev returned to be dev
1020                  */
1021         }
1022
1023         return dev;
1024 }
1025
1026 static const int fib6_prop[RTN_MAX + 1] = {
1027         [RTN_UNSPEC]    = 0,
1028         [RTN_UNICAST]   = 0,
1029         [RTN_LOCAL]     = 0,
1030         [RTN_BROADCAST] = 0,
1031         [RTN_ANYCAST]   = 0,
1032         [RTN_MULTICAST] = 0,
1033         [RTN_BLACKHOLE] = -EINVAL,
1034         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1035         [RTN_PROHIBIT]  = -EACCES,
1036         [RTN_THROW]     = -EAGAIN,
1037         [RTN_NAT]       = -EINVAL,
1038         [RTN_XRESOLVE]  = -EINVAL,
1039 };
1040
1041 static int ip6_rt_type_to_error(u8 fib6_type)
1042 {
1043         return fib6_prop[fib6_type];
1044 }
1045
1046 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1047 {
1048         unsigned short flags = 0;
1049
1050         if (rt->dst_nocount)
1051                 flags |= DST_NOCOUNT;
1052         if (rt->dst_nopolicy)
1053                 flags |= DST_NOPOLICY;
1054
1055         return flags;
1056 }
1057
1058 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1059 {
1060         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1061
1062         switch (fib6_type) {
1063         case RTN_BLACKHOLE:
1064                 rt->dst.output = dst_discard_out;
1065                 rt->dst.input = dst_discard;
1066                 break;
1067         case RTN_PROHIBIT:
1068                 rt->dst.output = ip6_pkt_prohibit_out;
1069                 rt->dst.input = ip6_pkt_prohibit;
1070                 break;
1071         case RTN_THROW:
1072         case RTN_UNREACHABLE:
1073         default:
1074                 rt->dst.output = ip6_pkt_discard_out;
1075                 rt->dst.input = ip6_pkt_discard;
1076                 break;
1077         }
1078 }
1079
1080 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1081 {
1082         struct fib6_info *f6i = res->f6i;
1083
1084         if (res->fib6_flags & RTF_REJECT) {
1085                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1086                 return;
1087         }
1088
1089         rt->dst.error = 0;
1090         rt->dst.output = ip6_output;
1091
1092         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1093                 rt->dst.input = ip6_input;
1094         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1095                 rt->dst.input = ip6_mc_input;
1096         } else {
1097                 rt->dst.input = ip6_forward;
1098         }
1099
1100         if (res->nh->fib_nh_lws) {
1101                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1102                 lwtunnel_set_redirect(&rt->dst);
1103         }
1104
1105         rt->dst.lastuse = jiffies;
1106 }
1107
1108 /* Caller must already hold reference to @from */
1109 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1110 {
1111         rt->rt6i_flags &= ~RTF_EXPIRES;
1112         rcu_assign_pointer(rt->from, from);
1113         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1114 }
1115
1116 /* Caller must already hold reference to f6i in result */
1117 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1118 {
1119         const struct fib6_nh *nh = res->nh;
1120         const struct net_device *dev = nh->fib_nh_dev;
1121         struct fib6_info *f6i = res->f6i;
1122
1123         ip6_rt_init_dst(rt, res);
1124
1125         rt->rt6i_dst = f6i->fib6_dst;
1126         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1127         rt->rt6i_flags = res->fib6_flags;
1128         if (nh->fib_nh_gw_family) {
1129                 rt->rt6i_gateway = nh->fib_nh_gw6;
1130                 rt->rt6i_flags |= RTF_GATEWAY;
1131         }
1132         rt6_set_from(rt, f6i);
1133 #ifdef CONFIG_IPV6_SUBTREES
1134         rt->rt6i_src = f6i->fib6_src;
1135 #endif
1136 }
1137
1138 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1139                                         struct in6_addr *saddr)
1140 {
1141         struct fib6_node *pn, *sn;
1142         while (1) {
1143                 if (fn->fn_flags & RTN_TL_ROOT)
1144                         return NULL;
1145                 pn = rcu_dereference(fn->parent);
1146                 sn = FIB6_SUBTREE(pn);
1147                 if (sn && sn != fn)
1148                         fn = fib6_node_lookup(sn, NULL, saddr);
1149                 else
1150                         fn = pn;
1151                 if (fn->fn_flags & RTN_RTINFO)
1152                         return fn;
1153         }
1154 }
1155
1156 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1157 {
1158         struct rt6_info *rt = *prt;
1159
1160         if (dst_hold_safe(&rt->dst))
1161                 return true;
1162         if (net) {
1163                 rt = net->ipv6.ip6_null_entry;
1164                 dst_hold(&rt->dst);
1165         } else {
1166                 rt = NULL;
1167         }
1168         *prt = rt;
1169         return false;
1170 }
1171
1172 /* called with rcu_lock held */
1173 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1174 {
1175         struct net_device *dev = res->nh->fib_nh_dev;
1176         struct fib6_info *f6i = res->f6i;
1177         unsigned short flags;
1178         struct rt6_info *nrt;
1179
1180         if (!fib6_info_hold_safe(f6i))
1181                 goto fallback;
1182
1183         flags = fib6_info_dst_flags(f6i);
1184         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1185         if (!nrt) {
1186                 fib6_info_release(f6i);
1187                 goto fallback;
1188         }
1189
1190         ip6_rt_copy_init(nrt, res);
1191         return nrt;
1192
1193 fallback:
1194         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1195         dst_hold(&nrt->dst);
1196         return nrt;
1197 }
1198
1199 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1200                                              struct fib6_table *table,
1201                                              struct flowi6 *fl6,
1202                                              const struct sk_buff *skb,
1203                                              int flags)
1204 {
1205         struct fib6_result res = {};
1206         struct fib6_node *fn;
1207         struct rt6_info *rt;
1208
1209         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1210                 flags &= ~RT6_LOOKUP_F_IFACE;
1211
1212         rcu_read_lock();
1213         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1214 restart:
1215         res.f6i = rcu_dereference(fn->leaf);
1216         if (!res.f6i)
1217                 res.f6i = net->ipv6.fib6_null_entry;
1218         else
1219                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1220                                  flags);
1221
1222         if (res.f6i == net->ipv6.fib6_null_entry) {
1223                 fn = fib6_backtrack(fn, &fl6->saddr);
1224                 if (fn)
1225                         goto restart;
1226
1227                 rt = net->ipv6.ip6_null_entry;
1228                 dst_hold(&rt->dst);
1229                 goto out;
1230         } else if (res.fib6_flags & RTF_REJECT) {
1231                 goto do_create;
1232         }
1233
1234         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1235                          fl6->flowi6_oif != 0, skb, flags);
1236
1237         /* Search through exception table */
1238         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1239         if (rt) {
1240                 if (ip6_hold_safe(net, &rt))
1241                         dst_use_noref(&rt->dst, jiffies);
1242         } else {
1243 do_create:
1244                 rt = ip6_create_rt_rcu(&res);
1245         }
1246
1247 out:
1248         trace_fib6_table_lookup(net, &res, table, fl6);
1249
1250         rcu_read_unlock();
1251
1252         return rt;
1253 }
1254
1255 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1256                                    const struct sk_buff *skb, int flags)
1257 {
1258         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1259 }
1260 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1261
1262 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1263                             const struct in6_addr *saddr, int oif,
1264                             const struct sk_buff *skb, int strict)
1265 {
1266         struct flowi6 fl6 = {
1267                 .flowi6_oif = oif,
1268                 .daddr = *daddr,
1269         };
1270         struct dst_entry *dst;
1271         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1272
1273         if (saddr) {
1274                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1275                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1276         }
1277
1278         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1279         if (dst->error == 0)
1280                 return (struct rt6_info *) dst;
1281
1282         dst_release(dst);
1283
1284         return NULL;
1285 }
1286 EXPORT_SYMBOL(rt6_lookup);
1287
1288 /* ip6_ins_rt is called with FREE table->tb6_lock.
1289  * It takes new route entry, the addition fails by any reason the
1290  * route is released.
1291  * Caller must hold dst before calling it.
1292  */
1293
1294 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1295                         struct netlink_ext_ack *extack)
1296 {
1297         int err;
1298         struct fib6_table *table;
1299
1300         table = rt->fib6_table;
1301         spin_lock_bh(&table->tb6_lock);
1302         err = fib6_add(&table->tb6_root, rt, info, extack);
1303         spin_unlock_bh(&table->tb6_lock);
1304
1305         return err;
1306 }
1307
1308 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1309 {
1310         struct nl_info info = { .nl_net = net, };
1311
1312         return __ip6_ins_rt(rt, &info, NULL);
1313 }
1314
1315 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1316                                            const struct in6_addr *daddr,
1317                                            const struct in6_addr *saddr)
1318 {
1319         struct fib6_info *f6i = res->f6i;
1320         struct net_device *dev;
1321         struct rt6_info *rt;
1322
1323         /*
1324          *      Clone the route.
1325          */
1326
1327         if (!fib6_info_hold_safe(f6i))
1328                 return NULL;
1329
1330         dev = ip6_rt_get_dev_rcu(res);
1331         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1332         if (!rt) {
1333                 fib6_info_release(f6i);
1334                 return NULL;
1335         }
1336
1337         ip6_rt_copy_init(rt, res);
1338         rt->rt6i_flags |= RTF_CACHE;
1339         rt->rt6i_dst.addr = *daddr;
1340         rt->rt6i_dst.plen = 128;
1341
1342         if (!rt6_is_gw_or_nonexthop(res)) {
1343                 if (f6i->fib6_dst.plen != 128 &&
1344                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1345                         rt->rt6i_flags |= RTF_ANYCAST;
1346 #ifdef CONFIG_IPV6_SUBTREES
1347                 if (rt->rt6i_src.plen && saddr) {
1348                         rt->rt6i_src.addr = *saddr;
1349                         rt->rt6i_src.plen = 128;
1350                 }
1351 #endif
1352         }
1353
1354         return rt;
1355 }
1356
1357 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1358 {
1359         struct fib6_info *f6i = res->f6i;
1360         unsigned short flags = fib6_info_dst_flags(f6i);
1361         struct net_device *dev;
1362         struct rt6_info *pcpu_rt;
1363
1364         if (!fib6_info_hold_safe(f6i))
1365                 return NULL;
1366
1367         rcu_read_lock();
1368         dev = ip6_rt_get_dev_rcu(res);
1369         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1370         rcu_read_unlock();
1371         if (!pcpu_rt) {
1372                 fib6_info_release(f6i);
1373                 return NULL;
1374         }
1375         ip6_rt_copy_init(pcpu_rt, res);
1376         pcpu_rt->rt6i_flags |= RTF_PCPU;
1377
1378         if (f6i->nh)
1379                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1380
1381         return pcpu_rt;
1382 }
1383
1384 static bool rt6_is_valid(const struct rt6_info *rt6)
1385 {
1386         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1387 }
1388
1389 /* It should be called with rcu_read_lock() acquired */
1390 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1391 {
1392         struct rt6_info *pcpu_rt;
1393
1394         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1395
1396         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1397                 struct rt6_info *prev, **p;
1398
1399                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1400                 prev = xchg(p, NULL);
1401                 if (prev) {
1402                         dst_dev_put(&prev->dst);
1403                         dst_release(&prev->dst);
1404                 }
1405
1406                 pcpu_rt = NULL;
1407         }
1408
1409         return pcpu_rt;
1410 }
1411
1412 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1413                                             const struct fib6_result *res)
1414 {
1415         struct rt6_info *pcpu_rt, *prev, **p;
1416
1417         pcpu_rt = ip6_rt_pcpu_alloc(res);
1418         if (!pcpu_rt)
1419                 return NULL;
1420
1421         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1422         prev = cmpxchg(p, NULL, pcpu_rt);
1423         BUG_ON(prev);
1424
1425         if (res->f6i->fib6_destroying) {
1426                 struct fib6_info *from;
1427
1428                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1429                 fib6_info_release(from);
1430         }
1431
1432         return pcpu_rt;
1433 }
1434
1435 /* exception hash table implementation
1436  */
1437 static DEFINE_SPINLOCK(rt6_exception_lock);
1438
1439 /* Remove rt6_ex from hash table and free the memory
1440  * Caller must hold rt6_exception_lock
1441  */
1442 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1443                                  struct rt6_exception *rt6_ex)
1444 {
1445         struct fib6_info *from;
1446         struct net *net;
1447
1448         if (!bucket || !rt6_ex)
1449                 return;
1450
1451         net = dev_net(rt6_ex->rt6i->dst.dev);
1452         net->ipv6.rt6_stats->fib_rt_cache--;
1453
1454         /* purge completely the exception to allow releasing the held resources:
1455          * some [sk] cache may keep the dst around for unlimited time
1456          */
1457         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1458         fib6_info_release(from);
1459         dst_dev_put(&rt6_ex->rt6i->dst);
1460
1461         hlist_del_rcu(&rt6_ex->hlist);
1462         dst_release(&rt6_ex->rt6i->dst);
1463         kfree_rcu(rt6_ex, rcu);
1464         WARN_ON_ONCE(!bucket->depth);
1465         bucket->depth--;
1466 }
1467
1468 /* Remove oldest rt6_ex in bucket and free the memory
1469  * Caller must hold rt6_exception_lock
1470  */
1471 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1472 {
1473         struct rt6_exception *rt6_ex, *oldest = NULL;
1474
1475         if (!bucket)
1476                 return;
1477
1478         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1479                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1480                         oldest = rt6_ex;
1481         }
1482         rt6_remove_exception(bucket, oldest);
1483 }
1484
1485 static u32 rt6_exception_hash(const struct in6_addr *dst,
1486                               const struct in6_addr *src)
1487 {
1488         static siphash_key_t rt6_exception_key __read_mostly;
1489         struct {
1490                 struct in6_addr dst;
1491                 struct in6_addr src;
1492         } __aligned(SIPHASH_ALIGNMENT) combined = {
1493                 .dst = *dst,
1494         };
1495         u64 val;
1496
1497         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1498
1499 #ifdef CONFIG_IPV6_SUBTREES
1500         if (src)
1501                 combined.src = *src;
1502 #endif
1503         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1504
1505         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1506 }
1507
1508 /* Helper function to find the cached rt in the hash table
1509  * and update bucket pointer to point to the bucket for this
1510  * (daddr, saddr) pair
1511  * Caller must hold rt6_exception_lock
1512  */
1513 static struct rt6_exception *
1514 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1515                               const struct in6_addr *daddr,
1516                               const struct in6_addr *saddr)
1517 {
1518         struct rt6_exception *rt6_ex;
1519         u32 hval;
1520
1521         if (!(*bucket) || !daddr)
1522                 return NULL;
1523
1524         hval = rt6_exception_hash(daddr, saddr);
1525         *bucket += hval;
1526
1527         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1528                 struct rt6_info *rt6 = rt6_ex->rt6i;
1529                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1530
1531 #ifdef CONFIG_IPV6_SUBTREES
1532                 if (matched && saddr)
1533                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1534 #endif
1535                 if (matched)
1536                         return rt6_ex;
1537         }
1538         return NULL;
1539 }
1540
1541 /* Helper function to find the cached rt in the hash table
1542  * and update bucket pointer to point to the bucket for this
1543  * (daddr, saddr) pair
1544  * Caller must hold rcu_read_lock()
1545  */
1546 static struct rt6_exception *
1547 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1548                          const struct in6_addr *daddr,
1549                          const struct in6_addr *saddr)
1550 {
1551         struct rt6_exception *rt6_ex;
1552         u32 hval;
1553
1554         WARN_ON_ONCE(!rcu_read_lock_held());
1555
1556         if (!(*bucket) || !daddr)
1557                 return NULL;
1558
1559         hval = rt6_exception_hash(daddr, saddr);
1560         *bucket += hval;
1561
1562         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1563                 struct rt6_info *rt6 = rt6_ex->rt6i;
1564                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1565
1566 #ifdef CONFIG_IPV6_SUBTREES
1567                 if (matched && saddr)
1568                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1569 #endif
1570                 if (matched)
1571                         return rt6_ex;
1572         }
1573         return NULL;
1574 }
1575
1576 static unsigned int fib6_mtu(const struct fib6_result *res)
1577 {
1578         const struct fib6_nh *nh = res->nh;
1579         unsigned int mtu;
1580
1581         if (res->f6i->fib6_pmtu) {
1582                 mtu = res->f6i->fib6_pmtu;
1583         } else {
1584                 struct net_device *dev = nh->fib_nh_dev;
1585                 struct inet6_dev *idev;
1586
1587                 rcu_read_lock();
1588                 idev = __in6_dev_get(dev);
1589                 mtu = idev->cnf.mtu6;
1590                 rcu_read_unlock();
1591         }
1592
1593         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1594
1595         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1596 }
1597
1598 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1599
1600 /* used when the flushed bit is not relevant, only access to the bucket
1601  * (ie., all bucket users except rt6_insert_exception);
1602  *
1603  * called under rcu lock; sometimes called with rt6_exception_lock held
1604  */
1605 static
1606 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1607                                                        spinlock_t *lock)
1608 {
1609         struct rt6_exception_bucket *bucket;
1610
1611         if (lock)
1612                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1613                                                    lockdep_is_held(lock));
1614         else
1615                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1616
1617         /* remove bucket flushed bit if set */
1618         if (bucket) {
1619                 unsigned long p = (unsigned long)bucket;
1620
1621                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1622                 bucket = (struct rt6_exception_bucket *)p;
1623         }
1624
1625         return bucket;
1626 }
1627
1628 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1629 {
1630         unsigned long p = (unsigned long)bucket;
1631
1632         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1633 }
1634
1635 /* called with rt6_exception_lock held */
1636 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1637                                               spinlock_t *lock)
1638 {
1639         struct rt6_exception_bucket *bucket;
1640         unsigned long p;
1641
1642         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1643                                            lockdep_is_held(lock));
1644
1645         p = (unsigned long)bucket;
1646         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1647         bucket = (struct rt6_exception_bucket *)p;
1648         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1649 }
1650
1651 static int rt6_insert_exception(struct rt6_info *nrt,
1652                                 const struct fib6_result *res)
1653 {
1654         struct net *net = dev_net(nrt->dst.dev);
1655         struct rt6_exception_bucket *bucket;
1656         struct fib6_info *f6i = res->f6i;
1657         struct in6_addr *src_key = NULL;
1658         struct rt6_exception *rt6_ex;
1659         struct fib6_nh *nh = res->nh;
1660         int err = 0;
1661
1662         spin_lock_bh(&rt6_exception_lock);
1663
1664         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1665                                           lockdep_is_held(&rt6_exception_lock));
1666         if (!bucket) {
1667                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1668                                  GFP_ATOMIC);
1669                 if (!bucket) {
1670                         err = -ENOMEM;
1671                         goto out;
1672                 }
1673                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1674         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1675                 err = -EINVAL;
1676                 goto out;
1677         }
1678
1679 #ifdef CONFIG_IPV6_SUBTREES
1680         /* fib6_src.plen != 0 indicates f6i is in subtree
1681          * and exception table is indexed by a hash of
1682          * both fib6_dst and fib6_src.
1683          * Otherwise, the exception table is indexed by
1684          * a hash of only fib6_dst.
1685          */
1686         if (f6i->fib6_src.plen)
1687                 src_key = &nrt->rt6i_src.addr;
1688 #endif
1689         /* rt6_mtu_change() might lower mtu on f6i.
1690          * Only insert this exception route if its mtu
1691          * is less than f6i's mtu value.
1692          */
1693         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1694                 err = -EINVAL;
1695                 goto out;
1696         }
1697
1698         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1699                                                src_key);
1700         if (rt6_ex)
1701                 rt6_remove_exception(bucket, rt6_ex);
1702
1703         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1704         if (!rt6_ex) {
1705                 err = -ENOMEM;
1706                 goto out;
1707         }
1708         rt6_ex->rt6i = nrt;
1709         rt6_ex->stamp = jiffies;
1710         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1711         bucket->depth++;
1712         net->ipv6.rt6_stats->fib_rt_cache++;
1713
1714         if (bucket->depth > FIB6_MAX_DEPTH)
1715                 rt6_exception_remove_oldest(bucket);
1716
1717 out:
1718         spin_unlock_bh(&rt6_exception_lock);
1719
1720         /* Update fn->fn_sernum to invalidate all cached dst */
1721         if (!err) {
1722                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1723                 fib6_update_sernum(net, f6i);
1724                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1725                 fib6_force_start_gc(net);
1726         }
1727
1728         return err;
1729 }
1730
1731 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1732 {
1733         struct rt6_exception_bucket *bucket;
1734         struct rt6_exception *rt6_ex;
1735         struct hlist_node *tmp;
1736         int i;
1737
1738         spin_lock_bh(&rt6_exception_lock);
1739
1740         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1741         if (!bucket)
1742                 goto out;
1743
1744         /* Prevent rt6_insert_exception() to recreate the bucket list */
1745         if (!from)
1746                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1747
1748         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1749                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1750                         if (!from ||
1751                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1752                                 rt6_remove_exception(bucket, rt6_ex);
1753                 }
1754                 WARN_ON_ONCE(!from && bucket->depth);
1755                 bucket++;
1756         }
1757 out:
1758         spin_unlock_bh(&rt6_exception_lock);
1759 }
1760
1761 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1762 {
1763         struct fib6_info *f6i = arg;
1764
1765         fib6_nh_flush_exceptions(nh, f6i);
1766
1767         return 0;
1768 }
1769
1770 void rt6_flush_exceptions(struct fib6_info *f6i)
1771 {
1772         if (f6i->nh)
1773                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1774                                          f6i);
1775         else
1776                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1777 }
1778
1779 /* Find cached rt in the hash table inside passed in rt
1780  * Caller has to hold rcu_read_lock()
1781  */
1782 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1783                                            const struct in6_addr *daddr,
1784                                            const struct in6_addr *saddr)
1785 {
1786         const struct in6_addr *src_key = NULL;
1787         struct rt6_exception_bucket *bucket;
1788         struct rt6_exception *rt6_ex;
1789         struct rt6_info *ret = NULL;
1790
1791 #ifdef CONFIG_IPV6_SUBTREES
1792         /* fib6i_src.plen != 0 indicates f6i is in subtree
1793          * and exception table is indexed by a hash of
1794          * both fib6_dst and fib6_src.
1795          * However, the src addr used to create the hash
1796          * might not be exactly the passed in saddr which
1797          * is a /128 addr from the flow.
1798          * So we need to use f6i->fib6_src to redo lookup
1799          * if the passed in saddr does not find anything.
1800          * (See the logic in ip6_rt_cache_alloc() on how
1801          * rt->rt6i_src is updated.)
1802          */
1803         if (res->f6i->fib6_src.plen)
1804                 src_key = saddr;
1805 find_ex:
1806 #endif
1807         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1808         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1809
1810         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1811                 ret = rt6_ex->rt6i;
1812
1813 #ifdef CONFIG_IPV6_SUBTREES
1814         /* Use fib6_src as src_key and redo lookup */
1815         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1816                 src_key = &res->f6i->fib6_src.addr;
1817                 goto find_ex;
1818         }
1819 #endif
1820
1821         return ret;
1822 }
1823
1824 /* Remove the passed in cached rt from the hash table that contains it */
1825 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1826                                     const struct rt6_info *rt)
1827 {
1828         const struct in6_addr *src_key = NULL;
1829         struct rt6_exception_bucket *bucket;
1830         struct rt6_exception *rt6_ex;
1831         int err;
1832
1833         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1834                 return -ENOENT;
1835
1836         spin_lock_bh(&rt6_exception_lock);
1837         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1838
1839 #ifdef CONFIG_IPV6_SUBTREES
1840         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1841          * and exception table is indexed by a hash of
1842          * both rt6i_dst and rt6i_src.
1843          * Otherwise, the exception table is indexed by
1844          * a hash of only rt6i_dst.
1845          */
1846         if (plen)
1847                 src_key = &rt->rt6i_src.addr;
1848 #endif
1849         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1850                                                &rt->rt6i_dst.addr,
1851                                                src_key);
1852         if (rt6_ex) {
1853                 rt6_remove_exception(bucket, rt6_ex);
1854                 err = 0;
1855         } else {
1856                 err = -ENOENT;
1857         }
1858
1859         spin_unlock_bh(&rt6_exception_lock);
1860         return err;
1861 }
1862
1863 struct fib6_nh_excptn_arg {
1864         struct rt6_info *rt;
1865         int             plen;
1866 };
1867
1868 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1869 {
1870         struct fib6_nh_excptn_arg *arg = _arg;
1871         int err;
1872
1873         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1874         if (err == 0)
1875                 return 1;
1876
1877         return 0;
1878 }
1879
1880 static int rt6_remove_exception_rt(struct rt6_info *rt)
1881 {
1882         struct fib6_info *from;
1883
1884         from = rcu_dereference(rt->from);
1885         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1886                 return -EINVAL;
1887
1888         if (from->nh) {
1889                 struct fib6_nh_excptn_arg arg = {
1890                         .rt = rt,
1891                         .plen = from->fib6_src.plen
1892                 };
1893                 int rc;
1894
1895                 /* rc = 1 means an entry was found */
1896                 rc = nexthop_for_each_fib6_nh(from->nh,
1897                                               rt6_nh_remove_exception_rt,
1898                                               &arg);
1899                 return rc ? 0 : -ENOENT;
1900         }
1901
1902         return fib6_nh_remove_exception(from->fib6_nh,
1903                                         from->fib6_src.plen, rt);
1904 }
1905
1906 /* Find rt6_ex which contains the passed in rt cache and
1907  * refresh its stamp
1908  */
1909 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1910                                      const struct rt6_info *rt)
1911 {
1912         const struct in6_addr *src_key = NULL;
1913         struct rt6_exception_bucket *bucket;
1914         struct rt6_exception *rt6_ex;
1915
1916         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1917 #ifdef CONFIG_IPV6_SUBTREES
1918         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1919          * and exception table is indexed by a hash of
1920          * both rt6i_dst and rt6i_src.
1921          * Otherwise, the exception table is indexed by
1922          * a hash of only rt6i_dst.
1923          */
1924         if (plen)
1925                 src_key = &rt->rt6i_src.addr;
1926 #endif
1927         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1928         if (rt6_ex)
1929                 rt6_ex->stamp = jiffies;
1930 }
1931
1932 struct fib6_nh_match_arg {
1933         const struct net_device *dev;
1934         const struct in6_addr   *gw;
1935         struct fib6_nh          *match;
1936 };
1937
1938 /* determine if fib6_nh has given device and gateway */
1939 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1940 {
1941         struct fib6_nh_match_arg *arg = _arg;
1942
1943         if (arg->dev != nh->fib_nh_dev ||
1944             (arg->gw && !nh->fib_nh_gw_family) ||
1945             (!arg->gw && nh->fib_nh_gw_family) ||
1946             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1947                 return 0;
1948
1949         arg->match = nh;
1950
1951         /* found a match, break the loop */
1952         return 1;
1953 }
1954
1955 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1956 {
1957         struct fib6_info *from;
1958         struct fib6_nh *fib6_nh;
1959
1960         rcu_read_lock();
1961
1962         from = rcu_dereference(rt->from);
1963         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1964                 goto unlock;
1965
1966         if (from->nh) {
1967                 struct fib6_nh_match_arg arg = {
1968                         .dev = rt->dst.dev,
1969                         .gw = &rt->rt6i_gateway,
1970                 };
1971
1972                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1973
1974                 if (!arg.match)
1975                         goto unlock;
1976                 fib6_nh = arg.match;
1977         } else {
1978                 fib6_nh = from->fib6_nh;
1979         }
1980         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1981 unlock:
1982         rcu_read_unlock();
1983 }
1984
1985 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1986                                          struct rt6_info *rt, int mtu)
1987 {
1988         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1989          * lowest MTU in the path: always allow updating the route PMTU to
1990          * reflect PMTU decreases.
1991          *
1992          * If the new MTU is higher, and the route PMTU is equal to the local
1993          * MTU, this means the old MTU is the lowest in the path, so allow
1994          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1995          * handle this.
1996          */
1997
1998         if (dst_mtu(&rt->dst) >= mtu)
1999                 return true;
2000
2001         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2002                 return true;
2003
2004         return false;
2005 }
2006
2007 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2008                                        const struct fib6_nh *nh, int mtu)
2009 {
2010         struct rt6_exception_bucket *bucket;
2011         struct rt6_exception *rt6_ex;
2012         int i;
2013
2014         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2015         if (!bucket)
2016                 return;
2017
2018         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2019                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2020                         struct rt6_info *entry = rt6_ex->rt6i;
2021
2022                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2023                          * route), the metrics of its rt->from have already
2024                          * been updated.
2025                          */
2026                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2027                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2028                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2029                 }
2030                 bucket++;
2031         }
2032 }
2033
2034 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2035
2036 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2037                                             const struct in6_addr *gateway)
2038 {
2039         struct rt6_exception_bucket *bucket;
2040         struct rt6_exception *rt6_ex;
2041         struct hlist_node *tmp;
2042         int i;
2043
2044         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2045                 return;
2046
2047         spin_lock_bh(&rt6_exception_lock);
2048         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2049         if (bucket) {
2050                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2051                         hlist_for_each_entry_safe(rt6_ex, tmp,
2052                                                   &bucket->chain, hlist) {
2053                                 struct rt6_info *entry = rt6_ex->rt6i;
2054
2055                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2056                                     RTF_CACHE_GATEWAY &&
2057                                     ipv6_addr_equal(gateway,
2058                                                     &entry->rt6i_gateway)) {
2059                                         rt6_remove_exception(bucket, rt6_ex);
2060                                 }
2061                         }
2062                         bucket++;
2063                 }
2064         }
2065
2066         spin_unlock_bh(&rt6_exception_lock);
2067 }
2068
2069 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2070                                       struct rt6_exception *rt6_ex,
2071                                       struct fib6_gc_args *gc_args,
2072                                       unsigned long now)
2073 {
2074         struct rt6_info *rt = rt6_ex->rt6i;
2075
2076         /* we are pruning and obsoleting aged-out and non gateway exceptions
2077          * even if others have still references to them, so that on next
2078          * dst_check() such references can be dropped.
2079          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2080          * expired, independently from their aging, as per RFC 8201 section 4
2081          */
2082         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2083                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2084                         RT6_TRACE("aging clone %p\n", rt);
2085                         rt6_remove_exception(bucket, rt6_ex);
2086                         return;
2087                 }
2088         } else if (time_after(jiffies, rt->dst.expires)) {
2089                 RT6_TRACE("purging expired route %p\n", rt);
2090                 rt6_remove_exception(bucket, rt6_ex);
2091                 return;
2092         }
2093
2094         if (rt->rt6i_flags & RTF_GATEWAY) {
2095                 struct neighbour *neigh;
2096
2097                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2098
2099                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2100                         RT6_TRACE("purging route %p via non-router but gateway\n",
2101                                   rt);
2102                         rt6_remove_exception(bucket, rt6_ex);
2103                         return;
2104                 }
2105         }
2106
2107         gc_args->more++;
2108 }
2109
2110 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2111                                    struct fib6_gc_args *gc_args,
2112                                    unsigned long now)
2113 {
2114         struct rt6_exception_bucket *bucket;
2115         struct rt6_exception *rt6_ex;
2116         struct hlist_node *tmp;
2117         int i;
2118
2119         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2120                 return;
2121
2122         rcu_read_lock_bh();
2123         spin_lock(&rt6_exception_lock);
2124         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2125         if (bucket) {
2126                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2127                         hlist_for_each_entry_safe(rt6_ex, tmp,
2128                                                   &bucket->chain, hlist) {
2129                                 rt6_age_examine_exception(bucket, rt6_ex,
2130                                                           gc_args, now);
2131                         }
2132                         bucket++;
2133                 }
2134         }
2135         spin_unlock(&rt6_exception_lock);
2136         rcu_read_unlock_bh();
2137 }
2138
2139 struct fib6_nh_age_excptn_arg {
2140         struct fib6_gc_args     *gc_args;
2141         unsigned long           now;
2142 };
2143
2144 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2145 {
2146         struct fib6_nh_age_excptn_arg *arg = _arg;
2147
2148         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2149         return 0;
2150 }
2151
2152 void rt6_age_exceptions(struct fib6_info *f6i,
2153                         struct fib6_gc_args *gc_args,
2154                         unsigned long now)
2155 {
2156         if (f6i->nh) {
2157                 struct fib6_nh_age_excptn_arg arg = {
2158                         .gc_args = gc_args,
2159                         .now = now
2160                 };
2161
2162                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2163                                          &arg);
2164         } else {
2165                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2166         }
2167 }
2168
2169 /* must be called with rcu lock held */
2170 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2171                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2172 {
2173         struct fib6_node *fn, *saved_fn;
2174
2175         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2176         saved_fn = fn;
2177
2178         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2179                 oif = 0;
2180
2181 redo_rt6_select:
2182         rt6_select(net, fn, oif, res, strict);
2183         if (res->f6i == net->ipv6.fib6_null_entry) {
2184                 fn = fib6_backtrack(fn, &fl6->saddr);
2185                 if (fn)
2186                         goto redo_rt6_select;
2187                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2188                         /* also consider unreachable route */
2189                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2190                         fn = saved_fn;
2191                         goto redo_rt6_select;
2192                 }
2193         }
2194
2195         trace_fib6_table_lookup(net, res, table, fl6);
2196
2197         return 0;
2198 }
2199
2200 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2201                                int oif, struct flowi6 *fl6,
2202                                const struct sk_buff *skb, int flags)
2203 {
2204         struct fib6_result res = {};
2205         struct rt6_info *rt = NULL;
2206         int strict = 0;
2207
2208         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2209                      !rcu_read_lock_held());
2210
2211         strict |= flags & RT6_LOOKUP_F_IFACE;
2212         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2213         if (net->ipv6.devconf_all->forwarding == 0)
2214                 strict |= RT6_LOOKUP_F_REACHABLE;
2215
2216         rcu_read_lock();
2217
2218         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2219         if (res.f6i == net->ipv6.fib6_null_entry)
2220                 goto out;
2221
2222         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2223
2224         /*Search through exception table */
2225         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2226         if (rt) {
2227                 goto out;
2228         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2229                             !res.nh->fib_nh_gw_family)) {
2230                 /* Create a RTF_CACHE clone which will not be
2231                  * owned by the fib6 tree.  It is for the special case where
2232                  * the daddr in the skb during the neighbor look-up is different
2233                  * from the fl6->daddr used to look-up route here.
2234                  */
2235                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2236
2237                 if (rt) {
2238                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2239                          * As rt6_uncached_list_add() does not consume refcnt,
2240                          * this refcnt is always returned to the caller even
2241                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2242                          */
2243                         rt6_uncached_list_add(rt);
2244                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2245                         rcu_read_unlock();
2246
2247                         return rt;
2248                 }
2249         } else {
2250                 /* Get a percpu copy */
2251                 local_bh_disable();
2252                 rt = rt6_get_pcpu_route(&res);
2253
2254                 if (!rt)
2255                         rt = rt6_make_pcpu_route(net, &res);
2256
2257                 local_bh_enable();
2258         }
2259 out:
2260         if (!rt)
2261                 rt = net->ipv6.ip6_null_entry;
2262         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2263                 ip6_hold_safe(net, &rt);
2264         rcu_read_unlock();
2265
2266         return rt;
2267 }
2268 EXPORT_SYMBOL_GPL(ip6_pol_route);
2269
2270 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2271                                             struct fib6_table *table,
2272                                             struct flowi6 *fl6,
2273                                             const struct sk_buff *skb,
2274                                             int flags)
2275 {
2276         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2277 }
2278
2279 struct dst_entry *ip6_route_input_lookup(struct net *net,
2280                                          struct net_device *dev,
2281                                          struct flowi6 *fl6,
2282                                          const struct sk_buff *skb,
2283                                          int flags)
2284 {
2285         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2286                 flags |= RT6_LOOKUP_F_IFACE;
2287
2288         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2289 }
2290 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2291
2292 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2293                                   struct flow_keys *keys,
2294                                   struct flow_keys *flkeys)
2295 {
2296         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2297         const struct ipv6hdr *key_iph = outer_iph;
2298         struct flow_keys *_flkeys = flkeys;
2299         const struct ipv6hdr *inner_iph;
2300         const struct icmp6hdr *icmph;
2301         struct ipv6hdr _inner_iph;
2302         struct icmp6hdr _icmph;
2303
2304         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2305                 goto out;
2306
2307         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2308                                    sizeof(_icmph), &_icmph);
2309         if (!icmph)
2310                 goto out;
2311
2312         if (!icmpv6_is_err(icmph->icmp6_type))
2313                 goto out;
2314
2315         inner_iph = skb_header_pointer(skb,
2316                                        skb_transport_offset(skb) + sizeof(*icmph),
2317                                        sizeof(_inner_iph), &_inner_iph);
2318         if (!inner_iph)
2319                 goto out;
2320
2321         key_iph = inner_iph;
2322         _flkeys = NULL;
2323 out:
2324         if (_flkeys) {
2325                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2326                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2327                 keys->tags.flow_label = _flkeys->tags.flow_label;
2328                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2329         } else {
2330                 keys->addrs.v6addrs.src = key_iph->saddr;
2331                 keys->addrs.v6addrs.dst = key_iph->daddr;
2332                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2333                 keys->basic.ip_proto = key_iph->nexthdr;
2334         }
2335 }
2336
2337 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2338                                            const struct sk_buff *skb,
2339                                            bool *p_has_inner)
2340 {
2341         u32 hash_fields = ip6_multipath_hash_fields(net);
2342         struct flow_keys keys, hash_keys;
2343
2344         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2345                 return 0;
2346
2347         memset(&hash_keys, 0, sizeof(hash_keys));
2348         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2349
2350         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2351         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2352                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2353         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2354                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2355         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2356                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2357         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2358                 hash_keys.tags.flow_label = keys.tags.flow_label;
2359         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2360                 hash_keys.ports.src = keys.ports.src;
2361         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2362                 hash_keys.ports.dst = keys.ports.dst;
2363
2364         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2365         return flow_hash_from_keys(&hash_keys);
2366 }
2367
2368 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2369                                            const struct sk_buff *skb,
2370                                            bool has_inner)
2371 {
2372         u32 hash_fields = ip6_multipath_hash_fields(net);
2373         struct flow_keys keys, hash_keys;
2374
2375         /* We assume the packet carries an encapsulation, but if none was
2376          * encountered during dissection of the outer flow, then there is no
2377          * point in calling the flow dissector again.
2378          */
2379         if (!has_inner)
2380                 return 0;
2381
2382         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2383                 return 0;
2384
2385         memset(&hash_keys, 0, sizeof(hash_keys));
2386         skb_flow_dissect_flow_keys(skb, &keys, 0);
2387
2388         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2389                 return 0;
2390
2391         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2392                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2393                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2394                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2395                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2396                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2397         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2398                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2399                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2400                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2401                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2402                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2403                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2404                         hash_keys.tags.flow_label = keys.tags.flow_label;
2405         }
2406
2407         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2408                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2409         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2410                 hash_keys.ports.src = keys.ports.src;
2411         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2412                 hash_keys.ports.dst = keys.ports.dst;
2413
2414         return flow_hash_from_keys(&hash_keys);
2415 }
2416
2417 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2418                                          const struct sk_buff *skb)
2419 {
2420         u32 mhash, mhash_inner;
2421         bool has_inner = true;
2422
2423         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2424         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2425
2426         return jhash_2words(mhash, mhash_inner, 0);
2427 }
2428
2429 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2430                                          const struct flowi6 *fl6)
2431 {
2432         u32 hash_fields = ip6_multipath_hash_fields(net);
2433         struct flow_keys hash_keys;
2434
2435         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2436                 return 0;
2437
2438         memset(&hash_keys, 0, sizeof(hash_keys));
2439         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2440         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2441                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2442         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2443                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2444         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2445                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2446         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2447                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2448         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2449                 hash_keys.ports.src = fl6->fl6_sport;
2450         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2451                 hash_keys.ports.dst = fl6->fl6_dport;
2452
2453         return flow_hash_from_keys(&hash_keys);
2454 }
2455
2456 /* if skb is set it will be used and fl6 can be NULL */
2457 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2458                        const struct sk_buff *skb, struct flow_keys *flkeys)
2459 {
2460         struct flow_keys hash_keys;
2461         u32 mhash = 0;
2462
2463         switch (ip6_multipath_hash_policy(net)) {
2464         case 0:
2465                 memset(&hash_keys, 0, sizeof(hash_keys));
2466                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2467                 if (skb) {
2468                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2469                 } else {
2470                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2471                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2472                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2473                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2474                 }
2475                 mhash = flow_hash_from_keys(&hash_keys);
2476                 break;
2477         case 1:
2478                 if (skb) {
2479                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2480                         struct flow_keys keys;
2481
2482                         /* short-circuit if we already have L4 hash present */
2483                         if (skb->l4_hash)
2484                                 return skb_get_hash_raw(skb) >> 1;
2485
2486                         memset(&hash_keys, 0, sizeof(hash_keys));
2487
2488                         if (!flkeys) {
2489                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2490                                 flkeys = &keys;
2491                         }
2492                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2493                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2494                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2495                         hash_keys.ports.src = flkeys->ports.src;
2496                         hash_keys.ports.dst = flkeys->ports.dst;
2497                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2498                 } else {
2499                         memset(&hash_keys, 0, sizeof(hash_keys));
2500                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2501                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2502                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2503                         hash_keys.ports.src = fl6->fl6_sport;
2504                         hash_keys.ports.dst = fl6->fl6_dport;
2505                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2506                 }
2507                 mhash = flow_hash_from_keys(&hash_keys);
2508                 break;
2509         case 2:
2510                 memset(&hash_keys, 0, sizeof(hash_keys));
2511                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2512                 if (skb) {
2513                         struct flow_keys keys;
2514
2515                         if (!flkeys) {
2516                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2517                                 flkeys = &keys;
2518                         }
2519
2520                         /* Inner can be v4 or v6 */
2521                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2522                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2523                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2524                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2525                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2526                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2527                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2528                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2529                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2530                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2531                         } else {
2532                                 /* Same as case 0 */
2533                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2534                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2535                         }
2536                 } else {
2537                         /* Same as case 0 */
2538                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2539                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2540                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2541                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2542                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2543                 }
2544                 mhash = flow_hash_from_keys(&hash_keys);
2545                 break;
2546         case 3:
2547                 if (skb)
2548                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2549                 else
2550                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2551                 break;
2552         }
2553
2554         return mhash >> 1;
2555 }
2556
2557 /* Called with rcu held */
2558 void ip6_route_input(struct sk_buff *skb)
2559 {
2560         const struct ipv6hdr *iph = ipv6_hdr(skb);
2561         struct net *net = dev_net(skb->dev);
2562         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2563         struct ip_tunnel_info *tun_info;
2564         struct flowi6 fl6 = {
2565                 .flowi6_iif = skb->dev->ifindex,
2566                 .daddr = iph->daddr,
2567                 .saddr = iph->saddr,
2568                 .flowlabel = ip6_flowinfo(iph),
2569                 .flowi6_mark = skb->mark,
2570                 .flowi6_proto = iph->nexthdr,
2571         };
2572         struct flow_keys *flkeys = NULL, _flkeys;
2573
2574         tun_info = skb_tunnel_info(skb);
2575         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2576                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2577
2578         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2579                 flkeys = &_flkeys;
2580
2581         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2582                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2583         skb_dst_drop(skb);
2584         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2585                                                       &fl6, skb, flags));
2586 }
2587
2588 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2589                                              struct fib6_table *table,
2590                                              struct flowi6 *fl6,
2591                                              const struct sk_buff *skb,
2592                                              int flags)
2593 {
2594         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2595 }
2596
2597 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2598                                                const struct sock *sk,
2599                                                struct flowi6 *fl6, int flags)
2600 {
2601         bool any_src;
2602
2603         if (ipv6_addr_type(&fl6->daddr) &
2604             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2605                 struct dst_entry *dst;
2606
2607                 /* This function does not take refcnt on the dst */
2608                 dst = l3mdev_link_scope_lookup(net, fl6);
2609                 if (dst)
2610                         return dst;
2611         }
2612
2613         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2614
2615         flags |= RT6_LOOKUP_F_DST_NOREF;
2616         any_src = ipv6_addr_any(&fl6->saddr);
2617         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2618             (fl6->flowi6_oif && any_src))
2619                 flags |= RT6_LOOKUP_F_IFACE;
2620
2621         if (!any_src)
2622                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2623         else if (sk)
2624                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2625
2626         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2627 }
2628 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2629
2630 struct dst_entry *ip6_route_output_flags(struct net *net,
2631                                          const struct sock *sk,
2632                                          struct flowi6 *fl6,
2633                                          int flags)
2634 {
2635         struct dst_entry *dst;
2636         struct rt6_info *rt6;
2637
2638         rcu_read_lock();
2639         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2640         rt6 = (struct rt6_info *)dst;
2641         /* For dst cached in uncached_list, refcnt is already taken. */
2642         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2643                 dst = &net->ipv6.ip6_null_entry->dst;
2644                 dst_hold(dst);
2645         }
2646         rcu_read_unlock();
2647
2648         return dst;
2649 }
2650 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2651
2652 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2653 {
2654         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2655         struct net_device *loopback_dev = net->loopback_dev;
2656         struct dst_entry *new = NULL;
2657
2658         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2659                        DST_OBSOLETE_DEAD, 0);
2660         if (rt) {
2661                 rt6_info_init(rt);
2662                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2663
2664                 new = &rt->dst;
2665                 new->__use = 1;
2666                 new->input = dst_discard;
2667                 new->output = dst_discard_out;
2668
2669                 dst_copy_metrics(new, &ort->dst);
2670
2671                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2672                 rt->rt6i_gateway = ort->rt6i_gateway;
2673                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2674
2675                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2676 #ifdef CONFIG_IPV6_SUBTREES
2677                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2678 #endif
2679         }
2680
2681         dst_release(dst_orig);
2682         return new ? new : ERR_PTR(-ENOMEM);
2683 }
2684
2685 /*
2686  *      Destination cache support functions
2687  */
2688
2689 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2690 {
2691         u32 rt_cookie = 0;
2692
2693         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2694                 return false;
2695
2696         if (fib6_check_expired(f6i))
2697                 return false;
2698
2699         return true;
2700 }
2701
2702 static struct dst_entry *rt6_check(struct rt6_info *rt,
2703                                    struct fib6_info *from,
2704                                    u32 cookie)
2705 {
2706         u32 rt_cookie = 0;
2707
2708         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2709             rt_cookie != cookie)
2710                 return NULL;
2711
2712         if (rt6_check_expired(rt))
2713                 return NULL;
2714
2715         return &rt->dst;
2716 }
2717
2718 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2719                                             struct fib6_info *from,
2720                                             u32 cookie)
2721 {
2722         if (!__rt6_check_expired(rt) &&
2723             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2724             fib6_check(from, cookie))
2725                 return &rt->dst;
2726         else
2727                 return NULL;
2728 }
2729
2730 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2731                                                         u32 cookie)
2732 {
2733         struct dst_entry *dst_ret;
2734         struct fib6_info *from;
2735         struct rt6_info *rt;
2736
2737         rt = container_of(dst, struct rt6_info, dst);
2738
2739         if (rt->sernum)
2740                 return rt6_is_valid(rt) ? dst : NULL;
2741
2742         rcu_read_lock();
2743
2744         /* All IPV6 dsts are created with ->obsolete set to the value
2745          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2746          * into this function always.
2747          */
2748
2749         from = rcu_dereference(rt->from);
2750
2751         if (from && (rt->rt6i_flags & RTF_PCPU ||
2752             unlikely(!list_empty(&rt->rt6i_uncached))))
2753                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2754         else
2755                 dst_ret = rt6_check(rt, from, cookie);
2756
2757         rcu_read_unlock();
2758
2759         return dst_ret;
2760 }
2761 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2762
2763 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2764 {
2765         struct rt6_info *rt = (struct rt6_info *) dst;
2766
2767         if (rt) {
2768                 if (rt->rt6i_flags & RTF_CACHE) {
2769                         rcu_read_lock();
2770                         if (rt6_check_expired(rt)) {
2771                                 rt6_remove_exception_rt(rt);
2772                                 dst = NULL;
2773                         }
2774                         rcu_read_unlock();
2775                 } else {
2776                         dst_release(dst);
2777                         dst = NULL;
2778                 }
2779         }
2780         return dst;
2781 }
2782
2783 static void ip6_link_failure(struct sk_buff *skb)
2784 {
2785         struct rt6_info *rt;
2786
2787         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2788
2789         rt = (struct rt6_info *) skb_dst(skb);
2790         if (rt) {
2791                 rcu_read_lock();
2792                 if (rt->rt6i_flags & RTF_CACHE) {
2793                         rt6_remove_exception_rt(rt);
2794                 } else {
2795                         struct fib6_info *from;
2796                         struct fib6_node *fn;
2797
2798                         from = rcu_dereference(rt->from);
2799                         if (from) {
2800                                 fn = rcu_dereference(from->fib6_node);
2801                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2802                                         fn->fn_sernum = -1;
2803                         }
2804                 }
2805                 rcu_read_unlock();
2806         }
2807 }
2808
2809 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2810 {
2811         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2812                 struct fib6_info *from;
2813
2814                 rcu_read_lock();
2815                 from = rcu_dereference(rt0->from);
2816                 if (from)
2817                         rt0->dst.expires = from->expires;
2818                 rcu_read_unlock();
2819         }
2820
2821         dst_set_expires(&rt0->dst, timeout);
2822         rt0->rt6i_flags |= RTF_EXPIRES;
2823 }
2824
2825 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2826 {
2827         struct net *net = dev_net(rt->dst.dev);
2828
2829         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2830         rt->rt6i_flags |= RTF_MODIFIED;
2831         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2832 }
2833
2834 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2835 {
2836         return !(rt->rt6i_flags & RTF_CACHE) &&
2837                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2838 }
2839
2840 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2841                                  const struct ipv6hdr *iph, u32 mtu,
2842                                  bool confirm_neigh)
2843 {
2844         const struct in6_addr *daddr, *saddr;
2845         struct rt6_info *rt6 = (struct rt6_info *)dst;
2846
2847         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2848          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2849          * [see also comment in rt6_mtu_change_route()]
2850          */
2851
2852         if (iph) {
2853                 daddr = &iph->daddr;
2854                 saddr = &iph->saddr;
2855         } else if (sk) {
2856                 daddr = &sk->sk_v6_daddr;
2857                 saddr = &inet6_sk(sk)->saddr;
2858         } else {
2859                 daddr = NULL;
2860                 saddr = NULL;
2861         }
2862
2863         if (confirm_neigh)
2864                 dst_confirm_neigh(dst, daddr);
2865
2866         if (mtu < IPV6_MIN_MTU)
2867                 return;
2868         if (mtu >= dst_mtu(dst))
2869                 return;
2870
2871         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2872                 rt6_do_update_pmtu(rt6, mtu);
2873                 /* update rt6_ex->stamp for cache */
2874                 if (rt6->rt6i_flags & RTF_CACHE)
2875                         rt6_update_exception_stamp_rt(rt6);
2876         } else if (daddr) {
2877                 struct fib6_result res = {};
2878                 struct rt6_info *nrt6;
2879
2880                 rcu_read_lock();
2881                 res.f6i = rcu_dereference(rt6->from);
2882                 if (!res.f6i)
2883                         goto out_unlock;
2884
2885                 res.fib6_flags = res.f6i->fib6_flags;
2886                 res.fib6_type = res.f6i->fib6_type;
2887
2888                 if (res.f6i->nh) {
2889                         struct fib6_nh_match_arg arg = {
2890                                 .dev = dst->dev,
2891                                 .gw = &rt6->rt6i_gateway,
2892                         };
2893
2894                         nexthop_for_each_fib6_nh(res.f6i->nh,
2895                                                  fib6_nh_find_match, &arg);
2896
2897                         /* fib6_info uses a nexthop that does not have fib6_nh
2898                          * using the dst->dev + gw. Should be impossible.
2899                          */
2900                         if (!arg.match)
2901                                 goto out_unlock;
2902
2903                         res.nh = arg.match;
2904                 } else {
2905                         res.nh = res.f6i->fib6_nh;
2906                 }
2907
2908                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2909                 if (nrt6) {
2910                         rt6_do_update_pmtu(nrt6, mtu);
2911                         if (rt6_insert_exception(nrt6, &res))
2912                                 dst_release_immediate(&nrt6->dst);
2913                 }
2914 out_unlock:
2915                 rcu_read_unlock();
2916         }
2917 }
2918
2919 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2920                                struct sk_buff *skb, u32 mtu,
2921                                bool confirm_neigh)
2922 {
2923         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2924                              confirm_neigh);
2925 }
2926
2927 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2928                      int oif, u32 mark, kuid_t uid)
2929 {
2930         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2931         struct dst_entry *dst;
2932         struct flowi6 fl6 = {
2933                 .flowi6_oif = oif,
2934                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2935                 .daddr = iph->daddr,
2936                 .saddr = iph->saddr,
2937                 .flowlabel = ip6_flowinfo(iph),
2938                 .flowi6_uid = uid,
2939         };
2940
2941         dst = ip6_route_output(net, NULL, &fl6);
2942         if (!dst->error)
2943                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2944         dst_release(dst);
2945 }
2946 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2947
2948 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2949 {
2950         int oif = sk->sk_bound_dev_if;
2951         struct dst_entry *dst;
2952
2953         if (!oif && skb->dev)
2954                 oif = l3mdev_master_ifindex(skb->dev);
2955
2956         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2957
2958         dst = __sk_dst_get(sk);
2959         if (!dst || !dst->obsolete ||
2960             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2961                 return;
2962
2963         bh_lock_sock(sk);
2964         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2965                 ip6_datagram_dst_update(sk, false);
2966         bh_unlock_sock(sk);
2967 }
2968 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2969
2970 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2971                            const struct flowi6 *fl6)
2972 {
2973 #ifdef CONFIG_IPV6_SUBTREES
2974         struct ipv6_pinfo *np = inet6_sk(sk);
2975 #endif
2976
2977         ip6_dst_store(sk, dst,
2978                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2979                       &sk->sk_v6_daddr : NULL,
2980 #ifdef CONFIG_IPV6_SUBTREES
2981                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2982                       &np->saddr :
2983 #endif
2984                       NULL);
2985 }
2986
2987 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2988                                   struct flowi6 *fl6,
2989                                   const struct in6_addr *gw,
2990                                   struct rt6_info **ret)
2991 {
2992         const struct fib6_nh *nh = res->nh;
2993
2994         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2995             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2996                 return false;
2997
2998         /* rt_cache's gateway might be different from its 'parent'
2999          * in the case of an ip redirect.
3000          * So we keep searching in the exception table if the gateway
3001          * is different.
3002          */
3003         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3004                 struct rt6_info *rt_cache;
3005
3006                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3007                 if (rt_cache &&
3008                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3009                         *ret = rt_cache;
3010                         return true;
3011                 }
3012                 return false;
3013         }
3014         return true;
3015 }
3016
3017 struct fib6_nh_rd_arg {
3018         struct fib6_result      *res;
3019         struct flowi6           *fl6;
3020         const struct in6_addr   *gw;
3021         struct rt6_info         **ret;
3022 };
3023
3024 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3025 {
3026         struct fib6_nh_rd_arg *arg = _arg;
3027
3028         arg->res->nh = nh;
3029         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3030 }
3031
3032 /* Handle redirects */
3033 struct ip6rd_flowi {
3034         struct flowi6 fl6;
3035         struct in6_addr gateway;
3036 };
3037
3038 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3039                                              struct fib6_table *table,
3040                                              struct flowi6 *fl6,
3041                                              const struct sk_buff *skb,
3042                                              int flags)
3043 {
3044         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3045         struct rt6_info *ret = NULL;
3046         struct fib6_result res = {};
3047         struct fib6_nh_rd_arg arg = {
3048                 .res = &res,
3049                 .fl6 = fl6,
3050                 .gw  = &rdfl->gateway,
3051                 .ret = &ret
3052         };
3053         struct fib6_info *rt;
3054         struct fib6_node *fn;
3055
3056         /* l3mdev_update_flow overrides oif if the device is enslaved; in
3057          * this case we must match on the real ingress device, so reset it
3058          */
3059         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
3060                 fl6->flowi6_oif = skb->dev->ifindex;
3061
3062         /* Get the "current" route for this destination and
3063          * check if the redirect has come from appropriate router.
3064          *
3065          * RFC 4861 specifies that redirects should only be
3066          * accepted if they come from the nexthop to the target.
3067          * Due to the way the routes are chosen, this notion
3068          * is a bit fuzzy and one might need to check all possible
3069          * routes.
3070          */
3071
3072         rcu_read_lock();
3073         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3074 restart:
3075         for_each_fib6_node_rt_rcu(fn) {
3076                 res.f6i = rt;
3077                 if (fib6_check_expired(rt))
3078                         continue;
3079                 if (rt->fib6_flags & RTF_REJECT)
3080                         break;
3081                 if (unlikely(rt->nh)) {
3082                         if (nexthop_is_blackhole(rt->nh))
3083                                 continue;
3084                         /* on match, res->nh is filled in and potentially ret */
3085                         if (nexthop_for_each_fib6_nh(rt->nh,
3086                                                      fib6_nh_redirect_match,
3087                                                      &arg))
3088                                 goto out;
3089                 } else {
3090                         res.nh = rt->fib6_nh;
3091                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3092                                                   &ret))
3093                                 goto out;
3094                 }
3095         }
3096
3097         if (!rt)
3098                 rt = net->ipv6.fib6_null_entry;
3099         else if (rt->fib6_flags & RTF_REJECT) {
3100                 ret = net->ipv6.ip6_null_entry;
3101                 goto out;
3102         }
3103
3104         if (rt == net->ipv6.fib6_null_entry) {
3105                 fn = fib6_backtrack(fn, &fl6->saddr);
3106                 if (fn)
3107                         goto restart;
3108         }
3109
3110         res.f6i = rt;
3111         res.nh = rt->fib6_nh;
3112 out:
3113         if (ret) {
3114                 ip6_hold_safe(net, &ret);
3115         } else {
3116                 res.fib6_flags = res.f6i->fib6_flags;
3117                 res.fib6_type = res.f6i->fib6_type;
3118                 ret = ip6_create_rt_rcu(&res);
3119         }
3120
3121         rcu_read_unlock();
3122
3123         trace_fib6_table_lookup(net, &res, table, fl6);
3124         return ret;
3125 };
3126
3127 static struct dst_entry *ip6_route_redirect(struct net *net,
3128                                             const struct flowi6 *fl6,
3129                                             const struct sk_buff *skb,
3130                                             const struct in6_addr *gateway)
3131 {
3132         int flags = RT6_LOOKUP_F_HAS_SADDR;
3133         struct ip6rd_flowi rdfl;
3134
3135         rdfl.fl6 = *fl6;
3136         rdfl.gateway = *gateway;
3137
3138         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3139                                 flags, __ip6_route_redirect);
3140 }
3141
3142 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3143                   kuid_t uid)
3144 {
3145         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3146         struct dst_entry *dst;
3147         struct flowi6 fl6 = {
3148                 .flowi6_iif = LOOPBACK_IFINDEX,
3149                 .flowi6_oif = oif,
3150                 .flowi6_mark = mark,
3151                 .daddr = iph->daddr,
3152                 .saddr = iph->saddr,
3153                 .flowlabel = ip6_flowinfo(iph),
3154                 .flowi6_uid = uid,
3155         };
3156
3157         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3158         rt6_do_redirect(dst, NULL, skb);
3159         dst_release(dst);
3160 }
3161 EXPORT_SYMBOL_GPL(ip6_redirect);
3162
3163 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3164 {
3165         const struct ipv6hdr *iph = ipv6_hdr(skb);
3166         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3167         struct dst_entry *dst;
3168         struct flowi6 fl6 = {
3169                 .flowi6_iif = LOOPBACK_IFINDEX,
3170                 .flowi6_oif = oif,
3171                 .daddr = msg->dest,
3172                 .saddr = iph->daddr,
3173                 .flowi6_uid = sock_net_uid(net, NULL),
3174         };
3175
3176         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3177         rt6_do_redirect(dst, NULL, skb);
3178         dst_release(dst);
3179 }
3180
3181 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3182 {
3183         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3184                      sk->sk_uid);
3185 }
3186 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3187
3188 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3189 {
3190         struct net_device *dev = dst->dev;
3191         unsigned int mtu = dst_mtu(dst);
3192         struct net *net = dev_net(dev);
3193
3194         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3195
3196         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3197                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3198
3199         /*
3200          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3201          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3202          * IPV6_MAXPLEN is also valid and means: "any MSS,
3203          * rely only on pmtu discovery"
3204          */
3205         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3206                 mtu = IPV6_MAXPLEN;
3207         return mtu;
3208 }
3209
3210 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3211 {
3212         struct inet6_dev *idev;
3213         unsigned int mtu;
3214
3215         mtu = dst_metric_raw(dst, RTAX_MTU);
3216         if (mtu)
3217                 goto out;
3218
3219         mtu = IPV6_MIN_MTU;
3220
3221         rcu_read_lock();
3222         idev = __in6_dev_get(dst->dev);
3223         if (idev)
3224                 mtu = idev->cnf.mtu6;
3225         rcu_read_unlock();
3226
3227 out:
3228         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3229
3230         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3231 }
3232 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3233
3234 /* MTU selection:
3235  * 1. mtu on route is locked - use it
3236  * 2. mtu from nexthop exception
3237  * 3. mtu from egress device
3238  *
3239  * based on ip6_dst_mtu_forward and exception logic of
3240  * rt6_find_cached_rt; called with rcu_read_lock
3241  */
3242 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3243                       const struct in6_addr *daddr,
3244                       const struct in6_addr *saddr)
3245 {
3246         const struct fib6_nh *nh = res->nh;
3247         struct fib6_info *f6i = res->f6i;
3248         struct inet6_dev *idev;
3249         struct rt6_info *rt;
3250         u32 mtu = 0;
3251
3252         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3253                 mtu = f6i->fib6_pmtu;
3254                 if (mtu)
3255                         goto out;
3256         }
3257
3258         rt = rt6_find_cached_rt(res, daddr, saddr);
3259         if (unlikely(rt)) {
3260                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3261         } else {
3262                 struct net_device *dev = nh->fib_nh_dev;
3263
3264                 mtu = IPV6_MIN_MTU;
3265                 idev = __in6_dev_get(dev);
3266                 if (idev && idev->cnf.mtu6 > mtu)
3267                         mtu = idev->cnf.mtu6;
3268         }
3269
3270         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3271 out:
3272         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3273 }
3274
3275 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3276                                   struct flowi6 *fl6)
3277 {
3278         struct dst_entry *dst;
3279         struct rt6_info *rt;
3280         struct inet6_dev *idev = in6_dev_get(dev);
3281         struct net *net = dev_net(dev);
3282
3283         if (unlikely(!idev))
3284                 return ERR_PTR(-ENODEV);
3285
3286         rt = ip6_dst_alloc(net, dev, 0);
3287         if (unlikely(!rt)) {
3288                 in6_dev_put(idev);
3289                 dst = ERR_PTR(-ENOMEM);
3290                 goto out;
3291         }
3292
3293         rt->dst.input = ip6_input;
3294         rt->dst.output  = ip6_output;
3295         rt->rt6i_gateway  = fl6->daddr;
3296         rt->rt6i_dst.addr = fl6->daddr;
3297         rt->rt6i_dst.plen = 128;
3298         rt->rt6i_idev     = idev;
3299         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3300
3301         /* Add this dst into uncached_list so that rt6_disable_ip() can
3302          * do proper release of the net_device
3303          */
3304         rt6_uncached_list_add(rt);
3305         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3306
3307         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3308
3309 out:
3310         return dst;
3311 }
3312
3313 static int ip6_dst_gc(struct dst_ops *ops)
3314 {
3315         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3316         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3317         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3318         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3319         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3320         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3321         int entries;
3322
3323         entries = dst_entries_get_fast(ops);
3324         if (entries > rt_max_size)
3325                 entries = dst_entries_get_slow(ops);
3326
3327         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3328             entries <= rt_max_size)
3329                 goto out;
3330
3331         net->ipv6.ip6_rt_gc_expire++;
3332         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3333         entries = dst_entries_get_slow(ops);
3334         if (entries < ops->gc_thresh)
3335                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3336 out:
3337         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3338         return entries > rt_max_size;
3339 }
3340
3341 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3342                                const struct in6_addr *gw_addr, u32 tbid,
3343                                int flags, struct fib6_result *res)
3344 {
3345         struct flowi6 fl6 = {
3346                 .flowi6_oif = cfg->fc_ifindex,
3347                 .daddr = *gw_addr,
3348                 .saddr = cfg->fc_prefsrc,
3349         };
3350         struct fib6_table *table;
3351         int err;
3352
3353         table = fib6_get_table(net, tbid);
3354         if (!table)
3355                 return -EINVAL;
3356
3357         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3358                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3359
3360         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3361
3362         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3363         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3364                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3365                                  cfg->fc_ifindex != 0, NULL, flags);
3366
3367         return err;
3368 }
3369
3370 static int ip6_route_check_nh_onlink(struct net *net,
3371                                      struct fib6_config *cfg,
3372                                      const struct net_device *dev,
3373                                      struct netlink_ext_ack *extack)
3374 {
3375         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3376         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3377         struct fib6_result res = {};
3378         int err;
3379
3380         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3381         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3382             /* ignore match if it is the default route */
3383             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3384             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3385                 NL_SET_ERR_MSG(extack,
3386                                "Nexthop has invalid gateway or device mismatch");
3387                 err = -EINVAL;
3388         }
3389
3390         return err;
3391 }
3392
3393 static int ip6_route_check_nh(struct net *net,
3394                               struct fib6_config *cfg,
3395                               struct net_device **_dev,
3396                               struct inet6_dev **idev)
3397 {
3398         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3399         struct net_device *dev = _dev ? *_dev : NULL;
3400         int flags = RT6_LOOKUP_F_IFACE;
3401         struct fib6_result res = {};
3402         int err = -EHOSTUNREACH;
3403
3404         if (cfg->fc_table) {
3405                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3406                                           cfg->fc_table, flags, &res);
3407                 /* gw_addr can not require a gateway or resolve to a reject
3408                  * route. If a device is given, it must match the result.
3409                  */
3410                 if (err || res.fib6_flags & RTF_REJECT ||
3411                     res.nh->fib_nh_gw_family ||
3412                     (dev && dev != res.nh->fib_nh_dev))
3413                         err = -EHOSTUNREACH;
3414         }
3415
3416         if (err < 0) {
3417                 struct flowi6 fl6 = {
3418                         .flowi6_oif = cfg->fc_ifindex,
3419                         .daddr = *gw_addr,
3420                 };
3421
3422                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3423                 if (err || res.fib6_flags & RTF_REJECT ||
3424                     res.nh->fib_nh_gw_family)
3425                         err = -EHOSTUNREACH;
3426
3427                 if (err)
3428                         return err;
3429
3430                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3431                                  cfg->fc_ifindex != 0, NULL, flags);
3432         }
3433
3434         err = 0;
3435         if (dev) {
3436                 if (dev != res.nh->fib_nh_dev)
3437                         err = -EHOSTUNREACH;
3438         } else {
3439                 *_dev = dev = res.nh->fib_nh_dev;
3440                 dev_hold(dev);
3441                 *idev = in6_dev_get(dev);
3442         }
3443
3444         return err;
3445 }
3446
3447 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3448                            struct net_device **_dev, struct inet6_dev **idev,
3449                            struct netlink_ext_ack *extack)
3450 {
3451         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3452         int gwa_type = ipv6_addr_type(gw_addr);
3453         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3454         const struct net_device *dev = *_dev;
3455         bool need_addr_check = !dev;
3456         int err = -EINVAL;
3457
3458         /* if gw_addr is local we will fail to detect this in case
3459          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3460          * will return already-added prefix route via interface that
3461          * prefix route was assigned to, which might be non-loopback.
3462          */
3463         if (dev &&
3464             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3465                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3466                 goto out;
3467         }
3468
3469         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3470                 /* IPv6 strictly inhibits using not link-local
3471                  * addresses as nexthop address.
3472                  * Otherwise, router will not able to send redirects.
3473                  * It is very good, but in some (rare!) circumstances
3474                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3475                  * some exceptions. --ANK
3476                  * We allow IPv4-mapped nexthops to support RFC4798-type
3477                  * addressing
3478                  */
3479                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3480                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3481                         goto out;
3482                 }
3483
3484                 rcu_read_lock();
3485
3486                 if (cfg->fc_flags & RTNH_F_ONLINK)
3487                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3488                 else
3489                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3490
3491                 rcu_read_unlock();
3492
3493                 if (err)
3494                         goto out;
3495         }
3496
3497         /* reload in case device was changed */
3498         dev = *_dev;
3499
3500         err = -EINVAL;
3501         if (!dev) {
3502                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3503                 goto out;
3504         } else if (dev->flags & IFF_LOOPBACK) {
3505                 NL_SET_ERR_MSG(extack,
3506                                "Egress device can not be loopback device for this route");
3507                 goto out;
3508         }
3509
3510         /* if we did not check gw_addr above, do so now that the
3511          * egress device has been resolved.
3512          */
3513         if (need_addr_check &&
3514             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3515                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3516                 goto out;
3517         }
3518
3519         err = 0;
3520 out:
3521         return err;
3522 }
3523
3524 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3525 {
3526         if ((flags & RTF_REJECT) ||
3527             (dev && (dev->flags & IFF_LOOPBACK) &&
3528              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3529              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3530                 return true;
3531
3532         return false;
3533 }
3534
3535 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3536                  struct fib6_config *cfg, gfp_t gfp_flags,
3537                  struct netlink_ext_ack *extack)
3538 {
3539         struct net_device *dev = NULL;
3540         struct inet6_dev *idev = NULL;
3541         int addr_type;
3542         int err;
3543
3544         fib6_nh->fib_nh_family = AF_INET6;
3545 #ifdef CONFIG_IPV6_ROUTER_PREF
3546         fib6_nh->last_probe = jiffies;
3547 #endif
3548         if (cfg->fc_is_fdb) {
3549                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3550                 fib6_nh->fib_nh_gw_family = AF_INET6;
3551                 return 0;
3552         }
3553
3554         err = -ENODEV;
3555         if (cfg->fc_ifindex) {
3556                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3557                 if (!dev)
3558                         goto out;
3559                 idev = in6_dev_get(dev);
3560                 if (!idev)
3561                         goto out;
3562         }
3563
3564         if (cfg->fc_flags & RTNH_F_ONLINK) {
3565                 if (!dev) {
3566                         NL_SET_ERR_MSG(extack,
3567                                        "Nexthop device required for onlink");
3568                         goto out;
3569                 }
3570
3571                 if (!(dev->flags & IFF_UP)) {
3572                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3573                         err = -ENETDOWN;
3574                         goto out;
3575                 }
3576
3577                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3578         }
3579
3580         fib6_nh->fib_nh_weight = 1;
3581
3582         /* We cannot add true routes via loopback here,
3583          * they would result in kernel looping; promote them to reject routes
3584          */
3585         addr_type = ipv6_addr_type(&cfg->fc_dst);
3586         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3587                 /* hold loopback dev/idev if we haven't done so. */
3588                 if (dev != net->loopback_dev) {
3589                         if (dev) {
3590                                 dev_put(dev);
3591                                 in6_dev_put(idev);
3592                         }
3593                         dev = net->loopback_dev;
3594                         dev_hold(dev);
3595                         idev = in6_dev_get(dev);
3596                         if (!idev) {
3597                                 err = -ENODEV;
3598                                 goto out;
3599                         }
3600                 }
3601                 goto pcpu_alloc;
3602         }
3603
3604         if (cfg->fc_flags & RTF_GATEWAY) {
3605                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3606                 if (err)
3607                         goto out;
3608
3609                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3610                 fib6_nh->fib_nh_gw_family = AF_INET6;
3611         }
3612
3613         err = -ENODEV;
3614         if (!dev)
3615                 goto out;
3616
3617         if (idev->cnf.disable_ipv6) {
3618                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3619                 err = -EACCES;
3620                 goto out;
3621         }
3622
3623         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3624                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3625                 err = -ENETDOWN;
3626                 goto out;
3627         }
3628
3629         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3630             !netif_carrier_ok(dev))
3631                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3632
3633         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3634                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3635         if (err)
3636                 goto out;
3637
3638 pcpu_alloc:
3639         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3640         if (!fib6_nh->rt6i_pcpu) {
3641                 err = -ENOMEM;
3642                 goto out;
3643         }
3644
3645         fib6_nh->fib_nh_dev = dev;
3646         fib6_nh->fib_nh_oif = dev->ifindex;
3647         err = 0;
3648 out:
3649         if (idev)
3650                 in6_dev_put(idev);
3651
3652         if (err) {
3653                 lwtstate_put(fib6_nh->fib_nh_lws);
3654                 fib6_nh->fib_nh_lws = NULL;
3655                 if (dev)
3656                         dev_put(dev);
3657         }
3658
3659         return err;
3660 }
3661
3662 void fib6_nh_release(struct fib6_nh *fib6_nh)
3663 {
3664         struct rt6_exception_bucket *bucket;
3665
3666         rcu_read_lock();
3667
3668         fib6_nh_flush_exceptions(fib6_nh, NULL);
3669         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3670         if (bucket) {
3671                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3672                 kfree(bucket);
3673         }
3674
3675         rcu_read_unlock();
3676
3677         if (fib6_nh->rt6i_pcpu) {
3678                 int cpu;
3679
3680                 for_each_possible_cpu(cpu) {
3681                         struct rt6_info **ppcpu_rt;
3682                         struct rt6_info *pcpu_rt;
3683
3684                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3685                         pcpu_rt = *ppcpu_rt;
3686                         if (pcpu_rt) {
3687                                 dst_dev_put(&pcpu_rt->dst);
3688                                 dst_release(&pcpu_rt->dst);
3689                                 *ppcpu_rt = NULL;
3690                         }
3691                 }
3692
3693                 free_percpu(fib6_nh->rt6i_pcpu);
3694         }
3695
3696         fib_nh_common_release(&fib6_nh->nh_common);
3697 }
3698
3699 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3700                                               gfp_t gfp_flags,
3701                                               struct netlink_ext_ack *extack)
3702 {
3703         struct net *net = cfg->fc_nlinfo.nl_net;
3704         struct fib6_info *rt = NULL;
3705         struct nexthop *nh = NULL;
3706         struct fib6_table *table;
3707         struct fib6_nh *fib6_nh;
3708         int err = -EINVAL;
3709         int addr_type;
3710
3711         /* RTF_PCPU is an internal flag; can not be set by userspace */
3712         if (cfg->fc_flags & RTF_PCPU) {
3713                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3714                 goto out;
3715         }
3716
3717         /* RTF_CACHE is an internal flag; can not be set by userspace */
3718         if (cfg->fc_flags & RTF_CACHE) {
3719                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3720                 goto out;
3721         }
3722
3723         if (cfg->fc_type > RTN_MAX) {
3724                 NL_SET_ERR_MSG(extack, "Invalid route type");
3725                 goto out;
3726         }
3727
3728         if (cfg->fc_dst_len > 128) {
3729                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3730                 goto out;
3731         }
3732         if (cfg->fc_src_len > 128) {
3733                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3734                 goto out;
3735         }
3736 #ifndef CONFIG_IPV6_SUBTREES
3737         if (cfg->fc_src_len) {
3738                 NL_SET_ERR_MSG(extack,
3739                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3740                 goto out;
3741         }
3742 #endif
3743         if (cfg->fc_nh_id) {
3744                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3745                 if (!nh) {
3746                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3747                         goto out;
3748                 }
3749                 err = fib6_check_nexthop(nh, cfg, extack);
3750                 if (err)
3751                         goto out;
3752         }
3753
3754         err = -ENOBUFS;
3755         if (cfg->fc_nlinfo.nlh &&
3756             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3757                 table = fib6_get_table(net, cfg->fc_table);
3758                 if (!table) {
3759                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3760                         table = fib6_new_table(net, cfg->fc_table);
3761                 }
3762         } else {
3763                 table = fib6_new_table(net, cfg->fc_table);
3764         }
3765
3766         if (!table)
3767                 goto out;
3768
3769         err = -ENOMEM;
3770         rt = fib6_info_alloc(gfp_flags, !nh);
3771         if (!rt)
3772                 goto out;
3773
3774         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3775                                                extack);
3776         if (IS_ERR(rt->fib6_metrics)) {
3777                 err = PTR_ERR(rt->fib6_metrics);
3778                 /* Do not leave garbage there. */
3779                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3780                 goto out_free;
3781         }
3782
3783         if (cfg->fc_flags & RTF_ADDRCONF)
3784                 rt->dst_nocount = true;
3785
3786         if (cfg->fc_flags & RTF_EXPIRES)
3787                 fib6_set_expires(rt, jiffies +
3788                                 clock_t_to_jiffies(cfg->fc_expires));
3789         else
3790                 fib6_clean_expires(rt);
3791
3792         if (cfg->fc_protocol == RTPROT_UNSPEC)
3793                 cfg->fc_protocol = RTPROT_BOOT;
3794         rt->fib6_protocol = cfg->fc_protocol;
3795
3796         rt->fib6_table = table;
3797         rt->fib6_metric = cfg->fc_metric;
3798         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3799         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3800
3801         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3802         rt->fib6_dst.plen = cfg->fc_dst_len;
3803
3804 #ifdef CONFIG_IPV6_SUBTREES
3805         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3806         rt->fib6_src.plen = cfg->fc_src_len;
3807 #endif
3808         if (nh) {
3809                 if (rt->fib6_src.plen) {
3810                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3811                         goto out_free;
3812                 }
3813                 if (!nexthop_get(nh)) {
3814                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3815                         goto out_free;
3816                 }
3817                 rt->nh = nh;
3818                 fib6_nh = nexthop_fib6_nh(rt->nh);
3819         } else {
3820                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3821                 if (err)
3822                         goto out;
3823
3824                 fib6_nh = rt->fib6_nh;
3825
3826                 /* We cannot add true routes via loopback here, they would
3827                  * result in kernel looping; promote them to reject routes
3828                  */
3829                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3830                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3831                                    addr_type))
3832                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3833         }
3834
3835         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3836                 struct net_device *dev = fib6_nh->fib_nh_dev;
3837
3838                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3839                         NL_SET_ERR_MSG(extack, "Invalid source address");
3840                         err = -EINVAL;
3841                         goto out;
3842                 }
3843                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3844                 rt->fib6_prefsrc.plen = 128;
3845         } else
3846                 rt->fib6_prefsrc.plen = 0;
3847
3848         return rt;
3849 out:
3850         fib6_info_release(rt);
3851         return ERR_PTR(err);
3852 out_free:
3853         ip_fib_metrics_put(rt->fib6_metrics);
3854         kfree(rt);
3855         return ERR_PTR(err);
3856 }
3857
3858 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3859                   struct netlink_ext_ack *extack)
3860 {
3861         struct fib6_info *rt;
3862         int err;
3863
3864         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3865         if (IS_ERR(rt))
3866                 return PTR_ERR(rt);
3867
3868         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3869         fib6_info_release(rt);
3870
3871         return err;
3872 }
3873
3874 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3875 {
3876         struct net *net = info->nl_net;
3877         struct fib6_table *table;
3878         int err;
3879
3880         if (rt == net->ipv6.fib6_null_entry) {
3881                 err = -ENOENT;
3882                 goto out;
3883         }
3884
3885         table = rt->fib6_table;
3886         spin_lock_bh(&table->tb6_lock);
3887         err = fib6_del(rt, info);
3888         spin_unlock_bh(&table->tb6_lock);
3889
3890 out:
3891         fib6_info_release(rt);
3892         return err;
3893 }
3894
3895 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3896 {
3897         struct nl_info info = {
3898                 .nl_net = net,
3899                 .skip_notify = skip_notify
3900         };
3901
3902         return __ip6_del_rt(rt, &info);
3903 }
3904
3905 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3906 {
3907         struct nl_info *info = &cfg->fc_nlinfo;
3908         struct net *net = info->nl_net;
3909         struct sk_buff *skb = NULL;
3910         struct fib6_table *table;
3911         int err = -ENOENT;
3912
3913         if (rt == net->ipv6.fib6_null_entry)
3914                 goto out_put;
3915         table = rt->fib6_table;
3916         spin_lock_bh(&table->tb6_lock);
3917
3918         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3919                 struct fib6_info *sibling, *next_sibling;
3920                 struct fib6_node *fn;
3921
3922                 /* prefer to send a single notification with all hops */
3923                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3924                 if (skb) {
3925                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3926
3927                         if (rt6_fill_node(net, skb, rt, NULL,
3928                                           NULL, NULL, 0, RTM_DELROUTE,
3929                                           info->portid, seq, 0) < 0) {
3930                                 kfree_skb(skb);
3931                                 skb = NULL;
3932                         } else
3933                                 info->skip_notify = 1;
3934                 }
3935
3936                 /* 'rt' points to the first sibling route. If it is not the
3937                  * leaf, then we do not need to send a notification. Otherwise,
3938                  * we need to check if the last sibling has a next route or not
3939                  * and emit a replace or delete notification, respectively.
3940                  */
3941                 info->skip_notify_kernel = 1;
3942                 fn = rcu_dereference_protected(rt->fib6_node,
3943                                             lockdep_is_held(&table->tb6_lock));
3944                 if (rcu_access_pointer(fn->leaf) == rt) {
3945                         struct fib6_info *last_sibling, *replace_rt;
3946
3947                         last_sibling = list_last_entry(&rt->fib6_siblings,
3948                                                        struct fib6_info,
3949                                                        fib6_siblings);
3950                         replace_rt = rcu_dereference_protected(
3951                                             last_sibling->fib6_next,
3952                                             lockdep_is_held(&table->tb6_lock));
3953                         if (replace_rt)
3954                                 call_fib6_entry_notifiers_replace(net,
3955                                                                   replace_rt);
3956                         else
3957                                 call_fib6_multipath_entry_notifiers(net,
3958                                                        FIB_EVENT_ENTRY_DEL,
3959                                                        rt, rt->fib6_nsiblings,
3960                                                        NULL);
3961                 }
3962                 list_for_each_entry_safe(sibling, next_sibling,
3963                                          &rt->fib6_siblings,
3964                                          fib6_siblings) {
3965                         err = fib6_del(sibling, info);
3966                         if (err)
3967                                 goto out_unlock;
3968                 }
3969         }
3970
3971         err = fib6_del(rt, info);
3972 out_unlock:
3973         spin_unlock_bh(&table->tb6_lock);
3974 out_put:
3975         fib6_info_release(rt);
3976
3977         if (skb) {
3978                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3979                             info->nlh, gfp_any());
3980         }
3981         return err;
3982 }
3983
3984 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3985 {
3986         int rc = -ESRCH;
3987
3988         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3989                 goto out;
3990
3991         if (cfg->fc_flags & RTF_GATEWAY &&
3992             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3993                 goto out;
3994
3995         rc = rt6_remove_exception_rt(rt);
3996 out:
3997         return rc;
3998 }
3999
4000 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
4001                              struct fib6_nh *nh)
4002 {
4003         struct fib6_result res = {
4004                 .f6i = rt,
4005                 .nh = nh,
4006         };
4007         struct rt6_info *rt_cache;
4008
4009         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
4010         if (rt_cache)
4011                 return __ip6_del_cached_rt(rt_cache, cfg);
4012
4013         return 0;
4014 }
4015
4016 struct fib6_nh_del_cached_rt_arg {
4017         struct fib6_config *cfg;
4018         struct fib6_info *f6i;
4019 };
4020
4021 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4022 {
4023         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4024         int rc;
4025
4026         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4027         return rc != -ESRCH ? rc : 0;
4028 }
4029
4030 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4031 {
4032         struct fib6_nh_del_cached_rt_arg arg = {
4033                 .cfg = cfg,
4034                 .f6i = f6i
4035         };
4036
4037         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4038 }
4039
4040 static int ip6_route_del(struct fib6_config *cfg,
4041                          struct netlink_ext_ack *extack)
4042 {
4043         struct fib6_table *table;
4044         struct fib6_info *rt;
4045         struct fib6_node *fn;
4046         int err = -ESRCH;
4047
4048         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4049         if (!table) {
4050                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4051                 return err;
4052         }
4053
4054         rcu_read_lock();
4055
4056         fn = fib6_locate(&table->tb6_root,
4057                          &cfg->fc_dst, cfg->fc_dst_len,
4058                          &cfg->fc_src, cfg->fc_src_len,
4059                          !(cfg->fc_flags & RTF_CACHE));
4060
4061         if (fn) {
4062                 for_each_fib6_node_rt_rcu(fn) {
4063                         struct fib6_nh *nh;
4064
4065                         if (rt->nh && cfg->fc_nh_id &&
4066                             rt->nh->id != cfg->fc_nh_id)
4067                                 continue;
4068
4069                         if (cfg->fc_flags & RTF_CACHE) {
4070                                 int rc = 0;
4071
4072                                 if (rt->nh) {
4073                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4074                                 } else if (cfg->fc_nh_id) {
4075                                         continue;
4076                                 } else {
4077                                         nh = rt->fib6_nh;
4078                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4079                                 }
4080                                 if (rc != -ESRCH) {
4081                                         rcu_read_unlock();
4082                                         return rc;
4083                                 }
4084                                 continue;
4085                         }
4086
4087                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4088                                 continue;
4089                         if (cfg->fc_protocol &&
4090                             cfg->fc_protocol != rt->fib6_protocol)
4091                                 continue;
4092
4093                         if (rt->nh) {
4094                                 if (!fib6_info_hold_safe(rt))
4095                                         continue;
4096                                 rcu_read_unlock();
4097
4098                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4099                         }
4100                         if (cfg->fc_nh_id)
4101                                 continue;
4102
4103                         nh = rt->fib6_nh;
4104                         if (cfg->fc_ifindex &&
4105                             (!nh->fib_nh_dev ||
4106                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4107                                 continue;
4108                         if (cfg->fc_flags & RTF_GATEWAY &&
4109                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4110                                 continue;
4111                         if (!fib6_info_hold_safe(rt))
4112                                 continue;
4113                         rcu_read_unlock();
4114
4115                         /* if gateway was specified only delete the one hop */
4116                         if (cfg->fc_flags & RTF_GATEWAY)
4117                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4118
4119                         return __ip6_del_rt_siblings(rt, cfg);
4120                 }
4121         }
4122         rcu_read_unlock();
4123
4124         return err;
4125 }
4126
4127 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4128 {
4129         struct netevent_redirect netevent;
4130         struct rt6_info *rt, *nrt = NULL;
4131         struct fib6_result res = {};
4132         struct ndisc_options ndopts;
4133         struct inet6_dev *in6_dev;
4134         struct neighbour *neigh;
4135         struct rd_msg *msg;
4136         int optlen, on_link;
4137         u8 *lladdr;
4138
4139         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4140         optlen -= sizeof(*msg);
4141
4142         if (optlen < 0) {
4143                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4144                 return;
4145         }
4146
4147         msg = (struct rd_msg *)icmp6_hdr(skb);
4148
4149         if (ipv6_addr_is_multicast(&msg->dest)) {
4150                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4151                 return;
4152         }
4153
4154         on_link = 0;
4155         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4156                 on_link = 1;
4157         } else if (ipv6_addr_type(&msg->target) !=
4158                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4159                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4160                 return;
4161         }
4162
4163         in6_dev = __in6_dev_get(skb->dev);
4164         if (!in6_dev)
4165                 return;
4166         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4167                 return;
4168
4169         /* RFC2461 8.1:
4170          *      The IP source address of the Redirect MUST be the same as the current
4171          *      first-hop router for the specified ICMP Destination Address.
4172          */
4173
4174         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4175                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4176                 return;
4177         }
4178
4179         lladdr = NULL;
4180         if (ndopts.nd_opts_tgt_lladdr) {
4181                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4182                                              skb->dev);
4183                 if (!lladdr) {
4184                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4185                         return;
4186                 }
4187         }
4188
4189         rt = (struct rt6_info *) dst;
4190         if (rt->rt6i_flags & RTF_REJECT) {
4191                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4192                 return;
4193         }
4194
4195         /* Redirect received -> path was valid.
4196          * Look, redirects are sent only in response to data packets,
4197          * so that this nexthop apparently is reachable. --ANK
4198          */
4199         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4200
4201         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4202         if (!neigh)
4203                 return;
4204
4205         /*
4206          *      We have finally decided to accept it.
4207          */
4208
4209         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4210                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4211                      NEIGH_UPDATE_F_OVERRIDE|
4212                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4213                                      NEIGH_UPDATE_F_ISROUTER)),
4214                      NDISC_REDIRECT, &ndopts);
4215
4216         rcu_read_lock();
4217         res.f6i = rcu_dereference(rt->from);
4218         if (!res.f6i)
4219                 goto out;
4220
4221         if (res.f6i->nh) {
4222                 struct fib6_nh_match_arg arg = {
4223                         .dev = dst->dev,
4224                         .gw = &rt->rt6i_gateway,
4225                 };
4226
4227                 nexthop_for_each_fib6_nh(res.f6i->nh,
4228                                          fib6_nh_find_match, &arg);
4229
4230                 /* fib6_info uses a nexthop that does not have fib6_nh
4231                  * using the dst->dev. Should be impossible
4232                  */
4233                 if (!arg.match)
4234                         goto out;
4235                 res.nh = arg.match;
4236         } else {
4237                 res.nh = res.f6i->fib6_nh;
4238         }
4239
4240         res.fib6_flags = res.f6i->fib6_flags;
4241         res.fib6_type = res.f6i->fib6_type;
4242         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4243         if (!nrt)
4244                 goto out;
4245
4246         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4247         if (on_link)
4248                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4249
4250         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4251
4252         /* rt6_insert_exception() will take care of duplicated exceptions */
4253         if (rt6_insert_exception(nrt, &res)) {
4254                 dst_release_immediate(&nrt->dst);
4255                 goto out;
4256         }
4257
4258         netevent.old = &rt->dst;
4259         netevent.new = &nrt->dst;
4260         netevent.daddr = &msg->dest;
4261         netevent.neigh = neigh;
4262         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4263
4264 out:
4265         rcu_read_unlock();
4266         neigh_release(neigh);
4267 }
4268
4269 #ifdef CONFIG_IPV6_ROUTE_INFO
4270 static struct fib6_info *rt6_get_route_info(struct net *net,
4271                                            const struct in6_addr *prefix, int prefixlen,
4272                                            const struct in6_addr *gwaddr,
4273                                            struct net_device *dev)
4274 {
4275         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4276         int ifindex = dev->ifindex;
4277         struct fib6_node *fn;
4278         struct fib6_info *rt = NULL;
4279         struct fib6_table *table;
4280
4281         table = fib6_get_table(net, tb_id);
4282         if (!table)
4283                 return NULL;
4284
4285         rcu_read_lock();
4286         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4287         if (!fn)
4288                 goto out;
4289
4290         for_each_fib6_node_rt_rcu(fn) {
4291                 /* these routes do not use nexthops */
4292                 if (rt->nh)
4293                         continue;
4294                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4295                         continue;
4296                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4297                     !rt->fib6_nh->fib_nh_gw_family)
4298                         continue;
4299                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4300                         continue;
4301                 if (!fib6_info_hold_safe(rt))
4302                         continue;
4303                 break;
4304         }
4305 out:
4306         rcu_read_unlock();
4307         return rt;
4308 }
4309
4310 static struct fib6_info *rt6_add_route_info(struct net *net,
4311                                            const struct in6_addr *prefix, int prefixlen,
4312                                            const struct in6_addr *gwaddr,
4313                                            struct net_device *dev,
4314                                            unsigned int pref)
4315 {
4316         struct fib6_config cfg = {
4317                 .fc_metric      = IP6_RT_PRIO_USER,
4318                 .fc_ifindex     = dev->ifindex,
4319                 .fc_dst_len     = prefixlen,
4320                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4321                                   RTF_UP | RTF_PREF(pref),
4322                 .fc_protocol = RTPROT_RA,
4323                 .fc_type = RTN_UNICAST,
4324                 .fc_nlinfo.portid = 0,
4325                 .fc_nlinfo.nlh = NULL,
4326                 .fc_nlinfo.nl_net = net,
4327         };
4328
4329         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4330         cfg.fc_dst = *prefix;
4331         cfg.fc_gateway = *gwaddr;
4332
4333         /* We should treat it as a default route if prefix length is 0. */
4334         if (!prefixlen)
4335                 cfg.fc_flags |= RTF_DEFAULT;
4336
4337         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4338
4339         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4340 }
4341 #endif
4342
4343 struct fib6_info *rt6_get_dflt_router(struct net *net,
4344                                      const struct in6_addr *addr,
4345                                      struct net_device *dev)
4346 {
4347         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4348         struct fib6_info *rt;
4349         struct fib6_table *table;
4350
4351         table = fib6_get_table(net, tb_id);
4352         if (!table)
4353                 return NULL;
4354
4355         rcu_read_lock();
4356         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4357                 struct fib6_nh *nh;
4358
4359                 /* RA routes do not use nexthops */
4360                 if (rt->nh)
4361                         continue;
4362
4363                 nh = rt->fib6_nh;
4364                 if (dev == nh->fib_nh_dev &&
4365                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4366                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4367                         break;
4368         }
4369         if (rt && !fib6_info_hold_safe(rt))
4370                 rt = NULL;
4371         rcu_read_unlock();
4372         return rt;
4373 }
4374
4375 struct fib6_info *rt6_add_dflt_router(struct net *net,
4376                                      const struct in6_addr *gwaddr,
4377                                      struct net_device *dev,
4378                                      unsigned int pref,
4379                                      u32 defrtr_usr_metric)
4380 {
4381         struct fib6_config cfg = {
4382                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4383                 .fc_metric      = defrtr_usr_metric,
4384                 .fc_ifindex     = dev->ifindex,
4385                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4386                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4387                 .fc_protocol = RTPROT_RA,
4388                 .fc_type = RTN_UNICAST,
4389                 .fc_nlinfo.portid = 0,
4390                 .fc_nlinfo.nlh = NULL,
4391                 .fc_nlinfo.nl_net = net,
4392         };
4393
4394         cfg.fc_gateway = *gwaddr;
4395
4396         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4397                 struct fib6_table *table;
4398
4399                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4400                 if (table)
4401                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4402         }
4403
4404         return rt6_get_dflt_router(net, gwaddr, dev);
4405 }
4406
4407 static void __rt6_purge_dflt_routers(struct net *net,
4408                                      struct fib6_table *table)
4409 {
4410         struct fib6_info *rt;
4411
4412 restart:
4413         rcu_read_lock();
4414         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4415                 struct net_device *dev = fib6_info_nh_dev(rt);
4416                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4417
4418                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4419                     (!idev || idev->cnf.accept_ra != 2) &&
4420                     fib6_info_hold_safe(rt)) {
4421                         rcu_read_unlock();
4422                         ip6_del_rt(net, rt, false);
4423                         goto restart;
4424                 }
4425         }
4426         rcu_read_unlock();
4427
4428         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4429 }
4430
4431 void rt6_purge_dflt_routers(struct net *net)
4432 {
4433         struct fib6_table *table;
4434         struct hlist_head *head;
4435         unsigned int h;
4436
4437         rcu_read_lock();
4438
4439         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4440                 head = &net->ipv6.fib_table_hash[h];
4441                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4442                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4443                                 __rt6_purge_dflt_routers(net, table);
4444                 }
4445         }
4446
4447         rcu_read_unlock();
4448 }
4449
4450 static void rtmsg_to_fib6_config(struct net *net,
4451                                  struct in6_rtmsg *rtmsg,
4452                                  struct fib6_config *cfg)
4453 {
4454         *cfg = (struct fib6_config){
4455                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4456                          : RT6_TABLE_MAIN,
4457                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4458                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4459                 .fc_expires = rtmsg->rtmsg_info,
4460                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4461                 .fc_src_len = rtmsg->rtmsg_src_len,
4462                 .fc_flags = rtmsg->rtmsg_flags,
4463                 .fc_type = rtmsg->rtmsg_type,
4464
4465                 .fc_nlinfo.nl_net = net,
4466
4467                 .fc_dst = rtmsg->rtmsg_dst,
4468                 .fc_src = rtmsg->rtmsg_src,
4469                 .fc_gateway = rtmsg->rtmsg_gateway,
4470         };
4471 }
4472
4473 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4474 {
4475         struct fib6_config cfg;
4476         int err;
4477
4478         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4479                 return -EINVAL;
4480         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4481                 return -EPERM;
4482
4483         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4484
4485         rtnl_lock();
4486         switch (cmd) {
4487         case SIOCADDRT:
4488                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4489                 break;
4490         case SIOCDELRT:
4491                 err = ip6_route_del(&cfg, NULL);
4492                 break;
4493         }
4494         rtnl_unlock();
4495         return err;
4496 }
4497
4498 /*
4499  *      Drop the packet on the floor
4500  */
4501
4502 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4503 {
4504         struct dst_entry *dst = skb_dst(skb);
4505         struct net *net = dev_net(dst->dev);
4506         struct inet6_dev *idev;
4507         int type;
4508
4509         if (netif_is_l3_master(skb->dev) &&
4510             dst->dev == net->loopback_dev)
4511                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4512         else
4513                 idev = ip6_dst_idev(dst);
4514
4515         switch (ipstats_mib_noroutes) {
4516         case IPSTATS_MIB_INNOROUTES:
4517                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4518                 if (type == IPV6_ADDR_ANY) {
4519                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4520                         break;
4521                 }
4522                 fallthrough;
4523         case IPSTATS_MIB_OUTNOROUTES:
4524                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4525                 break;
4526         }
4527
4528         /* Start over by dropping the dst for l3mdev case */
4529         if (netif_is_l3_master(skb->dev))
4530                 skb_dst_drop(skb);
4531
4532         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4533         kfree_skb(skb);
4534         return 0;
4535 }
4536
4537 static int ip6_pkt_discard(struct sk_buff *skb)
4538 {
4539         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4540 }
4541
4542 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4543 {
4544         skb->dev = skb_dst(skb)->dev;
4545         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4546 }
4547
4548 static int ip6_pkt_prohibit(struct sk_buff *skb)
4549 {
4550         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4551 }
4552
4553 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4554 {
4555         skb->dev = skb_dst(skb)->dev;
4556         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4557 }
4558
4559 /*
4560  *      Allocate a dst for local (unicast / anycast) address.
4561  */
4562
4563 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4564                                      struct inet6_dev *idev,
4565                                      const struct in6_addr *addr,
4566                                      bool anycast, gfp_t gfp_flags)
4567 {
4568         struct fib6_config cfg = {
4569                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4570                 .fc_ifindex = idev->dev->ifindex,
4571                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4572                 .fc_dst = *addr,
4573                 .fc_dst_len = 128,
4574                 .fc_protocol = RTPROT_KERNEL,
4575                 .fc_nlinfo.nl_net = net,
4576                 .fc_ignore_dev_down = true,
4577         };
4578         struct fib6_info *f6i;
4579
4580         if (anycast) {
4581                 cfg.fc_type = RTN_ANYCAST;
4582                 cfg.fc_flags |= RTF_ANYCAST;
4583         } else {
4584                 cfg.fc_type = RTN_LOCAL;
4585                 cfg.fc_flags |= RTF_LOCAL;
4586         }
4587
4588         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4589         if (!IS_ERR(f6i))
4590                 f6i->dst_nocount = true;
4591         return f6i;
4592 }
4593
4594 /* remove deleted ip from prefsrc entries */
4595 struct arg_dev_net_ip {
4596         struct net_device *dev;
4597         struct net *net;
4598         struct in6_addr *addr;
4599 };
4600
4601 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4602 {
4603         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4604         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4605         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4606
4607         if (!rt->nh &&
4608             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4609             rt != net->ipv6.fib6_null_entry &&
4610             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4611                 spin_lock_bh(&rt6_exception_lock);
4612                 /* remove prefsrc entry */
4613                 rt->fib6_prefsrc.plen = 0;
4614                 spin_unlock_bh(&rt6_exception_lock);
4615         }
4616         return 0;
4617 }
4618
4619 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4620 {
4621         struct net *net = dev_net(ifp->idev->dev);
4622         struct arg_dev_net_ip adni = {
4623                 .dev = ifp->idev->dev,
4624                 .net = net,
4625                 .addr = &ifp->addr,
4626         };
4627         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4628 }
4629
4630 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4631
4632 /* Remove routers and update dst entries when gateway turn into host. */
4633 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4634 {
4635         struct in6_addr *gateway = (struct in6_addr *)arg;
4636         struct fib6_nh *nh;
4637
4638         /* RA routes do not use nexthops */
4639         if (rt->nh)
4640                 return 0;
4641
4642         nh = rt->fib6_nh;
4643         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4644             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4645                 return -1;
4646
4647         /* Further clean up cached routes in exception table.
4648          * This is needed because cached route may have a different
4649          * gateway than its 'parent' in the case of an ip redirect.
4650          */
4651         fib6_nh_exceptions_clean_tohost(nh, gateway);
4652
4653         return 0;
4654 }
4655
4656 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4657 {
4658         fib6_clean_all(net, fib6_clean_tohost, gateway);
4659 }
4660
4661 struct arg_netdev_event {
4662         const struct net_device *dev;
4663         union {
4664                 unsigned char nh_flags;
4665                 unsigned long event;
4666         };
4667 };
4668
4669 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4670 {
4671         struct fib6_info *iter;
4672         struct fib6_node *fn;
4673
4674         fn = rcu_dereference_protected(rt->fib6_node,
4675                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4676         iter = rcu_dereference_protected(fn->leaf,
4677                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4678         while (iter) {
4679                 if (iter->fib6_metric == rt->fib6_metric &&
4680                     rt6_qualify_for_ecmp(iter))
4681                         return iter;
4682                 iter = rcu_dereference_protected(iter->fib6_next,
4683                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4684         }
4685
4686         return NULL;
4687 }
4688
4689 /* only called for fib entries with builtin fib6_nh */
4690 static bool rt6_is_dead(const struct fib6_info *rt)
4691 {
4692         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4693             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4694              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4695                 return true;
4696
4697         return false;
4698 }
4699
4700 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4701 {
4702         struct fib6_info *iter;
4703         int total = 0;
4704
4705         if (!rt6_is_dead(rt))
4706                 total += rt->fib6_nh->fib_nh_weight;
4707
4708         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4709                 if (!rt6_is_dead(iter))
4710                         total += iter->fib6_nh->fib_nh_weight;
4711         }
4712
4713         return total;
4714 }
4715
4716 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4717 {
4718         int upper_bound = -1;
4719
4720         if (!rt6_is_dead(rt)) {
4721                 *weight += rt->fib6_nh->fib_nh_weight;
4722                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4723                                                     total) - 1;
4724         }
4725         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4726 }
4727
4728 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4729 {
4730         struct fib6_info *iter;
4731         int weight = 0;
4732
4733         rt6_upper_bound_set(rt, &weight, total);
4734
4735         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4736                 rt6_upper_bound_set(iter, &weight, total);
4737 }
4738
4739 void rt6_multipath_rebalance(struct fib6_info *rt)
4740 {
4741         struct fib6_info *first;
4742         int total;
4743
4744         /* In case the entire multipath route was marked for flushing,
4745          * then there is no need to rebalance upon the removal of every
4746          * sibling route.
4747          */
4748         if (!rt->fib6_nsiblings || rt->should_flush)
4749                 return;
4750
4751         /* During lookup routes are evaluated in order, so we need to
4752          * make sure upper bounds are assigned from the first sibling
4753          * onwards.
4754          */
4755         first = rt6_multipath_first_sibling(rt);
4756         if (WARN_ON_ONCE(!first))
4757                 return;
4758
4759         total = rt6_multipath_total_weight(first);
4760         rt6_multipath_upper_bound_set(first, total);
4761 }
4762
4763 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4764 {
4765         const struct arg_netdev_event *arg = p_arg;
4766         struct net *net = dev_net(arg->dev);
4767
4768         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4769             rt->fib6_nh->fib_nh_dev == arg->dev) {
4770                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4771                 fib6_update_sernum_upto_root(net, rt);
4772                 rt6_multipath_rebalance(rt);
4773         }
4774
4775         return 0;
4776 }
4777
4778 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4779 {
4780         struct arg_netdev_event arg = {
4781                 .dev = dev,
4782                 {
4783                         .nh_flags = nh_flags,
4784                 },
4785         };
4786
4787         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4788                 arg.nh_flags |= RTNH_F_LINKDOWN;
4789
4790         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4791 }
4792
4793 /* only called for fib entries with inline fib6_nh */
4794 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4795                                    const struct net_device *dev)
4796 {
4797         struct fib6_info *iter;
4798
4799         if (rt->fib6_nh->fib_nh_dev == dev)
4800                 return true;
4801         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4802                 if (iter->fib6_nh->fib_nh_dev == dev)
4803                         return true;
4804
4805         return false;
4806 }
4807
4808 static void rt6_multipath_flush(struct fib6_info *rt)
4809 {
4810         struct fib6_info *iter;
4811
4812         rt->should_flush = 1;
4813         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4814                 iter->should_flush = 1;
4815 }
4816
4817 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4818                                              const struct net_device *down_dev)
4819 {
4820         struct fib6_info *iter;
4821         unsigned int dead = 0;
4822
4823         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4824             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4825                 dead++;
4826         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4827                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4828                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4829                         dead++;
4830
4831         return dead;
4832 }
4833
4834 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4835                                        const struct net_device *dev,
4836                                        unsigned char nh_flags)
4837 {
4838         struct fib6_info *iter;
4839
4840         if (rt->fib6_nh->fib_nh_dev == dev)
4841                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4842         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4843                 if (iter->fib6_nh->fib_nh_dev == dev)
4844                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4845 }
4846
4847 /* called with write lock held for table with rt */
4848 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4849 {
4850         const struct arg_netdev_event *arg = p_arg;
4851         const struct net_device *dev = arg->dev;
4852         struct net *net = dev_net(dev);
4853
4854         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4855                 return 0;
4856
4857         switch (arg->event) {
4858         case NETDEV_UNREGISTER:
4859                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4860         case NETDEV_DOWN:
4861                 if (rt->should_flush)
4862                         return -1;
4863                 if (!rt->fib6_nsiblings)
4864                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4865                 if (rt6_multipath_uses_dev(rt, dev)) {
4866                         unsigned int count;
4867
4868                         count = rt6_multipath_dead_count(rt, dev);
4869                         if (rt->fib6_nsiblings + 1 == count) {
4870                                 rt6_multipath_flush(rt);
4871                                 return -1;
4872                         }
4873                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4874                                                    RTNH_F_LINKDOWN);
4875                         fib6_update_sernum(net, rt);
4876                         rt6_multipath_rebalance(rt);
4877                 }
4878                 return -2;
4879         case NETDEV_CHANGE:
4880                 if (rt->fib6_nh->fib_nh_dev != dev ||
4881                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4882                         break;
4883                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4884                 rt6_multipath_rebalance(rt);
4885                 break;
4886         }
4887
4888         return 0;
4889 }
4890
4891 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4892 {
4893         struct arg_netdev_event arg = {
4894                 .dev = dev,
4895                 {
4896                         .event = event,
4897                 },
4898         };
4899         struct net *net = dev_net(dev);
4900
4901         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4902                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4903         else
4904                 fib6_clean_all(net, fib6_ifdown, &arg);
4905 }
4906
4907 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4908 {
4909         rt6_sync_down_dev(dev, event);
4910         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4911         neigh_ifdown(&nd_tbl, dev);
4912 }
4913
4914 struct rt6_mtu_change_arg {
4915         struct net_device *dev;
4916         unsigned int mtu;
4917         struct fib6_info *f6i;
4918 };
4919
4920 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4921 {
4922         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4923         struct fib6_info *f6i = arg->f6i;
4924
4925         /* For administrative MTU increase, there is no way to discover
4926          * IPv6 PMTU increase, so PMTU increase should be updated here.
4927          * Since RFC 1981 doesn't include administrative MTU increase
4928          * update PMTU increase is a MUST. (i.e. jumbo frame)
4929          */
4930         if (nh->fib_nh_dev == arg->dev) {
4931                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4932                 u32 mtu = f6i->fib6_pmtu;
4933
4934                 if (mtu >= arg->mtu ||
4935                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4936                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4937
4938                 spin_lock_bh(&rt6_exception_lock);
4939                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4940                 spin_unlock_bh(&rt6_exception_lock);
4941         }
4942
4943         return 0;
4944 }
4945
4946 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4947 {
4948         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4949         struct inet6_dev *idev;
4950
4951         /* In IPv6 pmtu discovery is not optional,
4952            so that RTAX_MTU lock cannot disable it.
4953            We still use this lock to block changes
4954            caused by addrconf/ndisc.
4955         */
4956
4957         idev = __in6_dev_get(arg->dev);
4958         if (!idev)
4959                 return 0;
4960
4961         if (fib6_metric_locked(f6i, RTAX_MTU))
4962                 return 0;
4963
4964         arg->f6i = f6i;
4965         if (f6i->nh) {
4966                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4967                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4968                                                 arg);
4969         }
4970
4971         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4972 }
4973
4974 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4975 {
4976         struct rt6_mtu_change_arg arg = {
4977                 .dev = dev,
4978                 .mtu = mtu,
4979         };
4980
4981         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4982 }
4983
4984 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4985         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4986         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4987         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4988         [RTA_OIF]               = { .type = NLA_U32 },
4989         [RTA_IIF]               = { .type = NLA_U32 },
4990         [RTA_PRIORITY]          = { .type = NLA_U32 },
4991         [RTA_METRICS]           = { .type = NLA_NESTED },
4992         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4993         [RTA_PREF]              = { .type = NLA_U8 },
4994         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4995         [RTA_ENCAP]             = { .type = NLA_NESTED },
4996         [RTA_EXPIRES]           = { .type = NLA_U32 },
4997         [RTA_UID]               = { .type = NLA_U32 },
4998         [RTA_MARK]              = { .type = NLA_U32 },
4999         [RTA_TABLE]             = { .type = NLA_U32 },
5000         [RTA_IP_PROTO]          = { .type = NLA_U8 },
5001         [RTA_SPORT]             = { .type = NLA_U16 },
5002         [RTA_DPORT]             = { .type = NLA_U16 },
5003         [RTA_NH_ID]             = { .type = NLA_U32 },
5004 };
5005
5006 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
5007                               struct fib6_config *cfg,
5008                               struct netlink_ext_ack *extack)
5009 {
5010         struct rtmsg *rtm;
5011         struct nlattr *tb[RTA_MAX+1];
5012         unsigned int pref;
5013         int err;
5014
5015         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5016                                      rtm_ipv6_policy, extack);
5017         if (err < 0)
5018                 goto errout;
5019
5020         err = -EINVAL;
5021         rtm = nlmsg_data(nlh);
5022
5023         *cfg = (struct fib6_config){
5024                 .fc_table = rtm->rtm_table,
5025                 .fc_dst_len = rtm->rtm_dst_len,
5026                 .fc_src_len = rtm->rtm_src_len,
5027                 .fc_flags = RTF_UP,
5028                 .fc_protocol = rtm->rtm_protocol,
5029                 .fc_type = rtm->rtm_type,
5030
5031                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5032                 .fc_nlinfo.nlh = nlh,
5033                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5034         };
5035
5036         if (rtm->rtm_type == RTN_UNREACHABLE ||
5037             rtm->rtm_type == RTN_BLACKHOLE ||
5038             rtm->rtm_type == RTN_PROHIBIT ||
5039             rtm->rtm_type == RTN_THROW)
5040                 cfg->fc_flags |= RTF_REJECT;
5041
5042         if (rtm->rtm_type == RTN_LOCAL)
5043                 cfg->fc_flags |= RTF_LOCAL;
5044
5045         if (rtm->rtm_flags & RTM_F_CLONED)
5046                 cfg->fc_flags |= RTF_CACHE;
5047
5048         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5049
5050         if (tb[RTA_NH_ID]) {
5051                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5052                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5053                         NL_SET_ERR_MSG(extack,
5054                                        "Nexthop specification and nexthop id are mutually exclusive");
5055                         goto errout;
5056                 }
5057                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5058         }
5059
5060         if (tb[RTA_GATEWAY]) {
5061                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5062                 cfg->fc_flags |= RTF_GATEWAY;
5063         }
5064         if (tb[RTA_VIA]) {
5065                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5066                 goto errout;
5067         }
5068
5069         if (tb[RTA_DST]) {
5070                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5071
5072                 if (nla_len(tb[RTA_DST]) < plen)
5073                         goto errout;
5074
5075                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5076         }
5077
5078         if (tb[RTA_SRC]) {
5079                 int plen = (rtm->rtm_src_len + 7) >> 3;
5080
5081                 if (nla_len(tb[RTA_SRC]) < plen)
5082                         goto errout;
5083
5084                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5085         }
5086
5087         if (tb[RTA_PREFSRC])
5088                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5089
5090         if (tb[RTA_OIF])
5091                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5092
5093         if (tb[RTA_PRIORITY])
5094                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5095
5096         if (tb[RTA_METRICS]) {
5097                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5098                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5099         }
5100
5101         if (tb[RTA_TABLE])
5102                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5103
5104         if (tb[RTA_MULTIPATH]) {
5105                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5106                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5107
5108                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5109                                                      cfg->fc_mp_len, extack);
5110                 if (err < 0)
5111                         goto errout;
5112         }
5113
5114         if (tb[RTA_PREF]) {
5115                 pref = nla_get_u8(tb[RTA_PREF]);
5116                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5117                     pref != ICMPV6_ROUTER_PREF_HIGH)
5118                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5119                 cfg->fc_flags |= RTF_PREF(pref);
5120         }
5121
5122         if (tb[RTA_ENCAP])
5123                 cfg->fc_encap = tb[RTA_ENCAP];
5124
5125         if (tb[RTA_ENCAP_TYPE]) {
5126                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5127
5128                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5129                 if (err < 0)
5130                         goto errout;
5131         }
5132
5133         if (tb[RTA_EXPIRES]) {
5134                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5135
5136                 if (addrconf_finite_timeout(timeout)) {
5137                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5138                         cfg->fc_flags |= RTF_EXPIRES;
5139                 }
5140         }
5141
5142         err = 0;
5143 errout:
5144         return err;
5145 }
5146
5147 struct rt6_nh {
5148         struct fib6_info *fib6_info;
5149         struct fib6_config r_cfg;
5150         struct list_head next;
5151 };
5152
5153 static int ip6_route_info_append(struct net *net,
5154                                  struct list_head *rt6_nh_list,
5155                                  struct fib6_info *rt,
5156                                  struct fib6_config *r_cfg)
5157 {
5158         struct rt6_nh *nh;
5159         int err = -EEXIST;
5160
5161         list_for_each_entry(nh, rt6_nh_list, next) {
5162                 /* check if fib6_info already exists */
5163                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5164                         return err;
5165         }
5166
5167         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5168         if (!nh)
5169                 return -ENOMEM;
5170         nh->fib6_info = rt;
5171         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5172         list_add_tail(&nh->next, rt6_nh_list);
5173
5174         return 0;
5175 }
5176
5177 static void ip6_route_mpath_notify(struct fib6_info *rt,
5178                                    struct fib6_info *rt_last,
5179                                    struct nl_info *info,
5180                                    __u16 nlflags)
5181 {
5182         /* if this is an APPEND route, then rt points to the first route
5183          * inserted and rt_last points to last route inserted. Userspace
5184          * wants a consistent dump of the route which starts at the first
5185          * nexthop. Since sibling routes are always added at the end of
5186          * the list, find the first sibling of the last route appended
5187          */
5188         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5189                 rt = list_first_entry(&rt_last->fib6_siblings,
5190                                       struct fib6_info,
5191                                       fib6_siblings);
5192         }
5193
5194         if (rt)
5195                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5196 }
5197
5198 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5199 {
5200         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5201         bool should_notify = false;
5202         struct fib6_info *leaf;
5203         struct fib6_node *fn;
5204
5205         rcu_read_lock();
5206         fn = rcu_dereference(rt->fib6_node);
5207         if (!fn)
5208                 goto out;
5209
5210         leaf = rcu_dereference(fn->leaf);
5211         if (!leaf)
5212                 goto out;
5213
5214         if (rt == leaf ||
5215             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5216              rt6_qualify_for_ecmp(leaf)))
5217                 should_notify = true;
5218 out:
5219         rcu_read_unlock();
5220
5221         return should_notify;
5222 }
5223
5224 static int ip6_route_multipath_add(struct fib6_config *cfg,
5225                                    struct netlink_ext_ack *extack)
5226 {
5227         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5228         struct nl_info *info = &cfg->fc_nlinfo;
5229         struct fib6_config r_cfg;
5230         struct rtnexthop *rtnh;
5231         struct fib6_info *rt;
5232         struct rt6_nh *err_nh;
5233         struct rt6_nh *nh, *nh_safe;
5234         __u16 nlflags;
5235         int remaining;
5236         int attrlen;
5237         int err = 1;
5238         int nhn = 0;
5239         int replace = (cfg->fc_nlinfo.nlh &&
5240                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5241         LIST_HEAD(rt6_nh_list);
5242
5243         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5244         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5245                 nlflags |= NLM_F_APPEND;
5246
5247         remaining = cfg->fc_mp_len;
5248         rtnh = (struct rtnexthop *)cfg->fc_mp;
5249
5250         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5251          * fib6_info structs per nexthop
5252          */
5253         while (rtnh_ok(rtnh, remaining)) {
5254                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5255                 if (rtnh->rtnh_ifindex)
5256                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5257
5258                 attrlen = rtnh_attrlen(rtnh);
5259                 if (attrlen > 0) {
5260                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5261
5262                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5263                         if (nla) {
5264                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5265                                 r_cfg.fc_flags |= RTF_GATEWAY;
5266                         }
5267                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5268                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5269                         if (nla)
5270                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5271                 }
5272
5273                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5274                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5275                 if (IS_ERR(rt)) {
5276                         err = PTR_ERR(rt);
5277                         rt = NULL;
5278                         goto cleanup;
5279                 }
5280                 if (!rt6_qualify_for_ecmp(rt)) {
5281                         err = -EINVAL;
5282                         NL_SET_ERR_MSG(extack,
5283                                        "Device only routes can not be added for IPv6 using the multipath API.");
5284                         fib6_info_release(rt);
5285                         goto cleanup;
5286                 }
5287
5288                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5289
5290                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5291                                             rt, &r_cfg);
5292                 if (err) {
5293                         fib6_info_release(rt);
5294                         goto cleanup;
5295                 }
5296
5297                 rtnh = rtnh_next(rtnh, &remaining);
5298         }
5299
5300         if (list_empty(&rt6_nh_list)) {
5301                 NL_SET_ERR_MSG(extack,
5302                                "Invalid nexthop configuration - no valid nexthops");
5303                 return -EINVAL;
5304         }
5305
5306         /* for add and replace send one notification with all nexthops.
5307          * Skip the notification in fib6_add_rt2node and send one with
5308          * the full route when done
5309          */
5310         info->skip_notify = 1;
5311
5312         /* For add and replace, send one notification with all nexthops. For
5313          * append, send one notification with all appended nexthops.
5314          */
5315         info->skip_notify_kernel = 1;
5316
5317         err_nh = NULL;
5318         list_for_each_entry(nh, &rt6_nh_list, next) {
5319                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5320                 fib6_info_release(nh->fib6_info);
5321
5322                 if (!err) {
5323                         /* save reference to last route successfully inserted */
5324                         rt_last = nh->fib6_info;
5325
5326                         /* save reference to first route for notification */
5327                         if (!rt_notif)
5328                                 rt_notif = nh->fib6_info;
5329                 }
5330
5331                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5332                 nh->fib6_info = NULL;
5333                 if (err) {
5334                         if (replace && nhn)
5335                                 NL_SET_ERR_MSG_MOD(extack,
5336                                                    "multipath route replace failed (check consistency of installed routes)");
5337                         err_nh = nh;
5338                         goto add_errout;
5339                 }
5340
5341                 /* Because each route is added like a single route we remove
5342                  * these flags after the first nexthop: if there is a collision,
5343                  * we have already failed to add the first nexthop:
5344                  * fib6_add_rt2node() has rejected it; when replacing, old
5345                  * nexthops have been replaced by first new, the rest should
5346                  * be added to it.
5347                  */
5348                 if (cfg->fc_nlinfo.nlh) {
5349                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5350                                                              NLM_F_REPLACE);
5351                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5352                 }
5353                 nhn++;
5354         }
5355
5356         /* An in-kernel notification should only be sent in case the new
5357          * multipath route is added as the first route in the node, or if
5358          * it was appended to it. We pass 'rt_notif' since it is the first
5359          * sibling and might allow us to skip some checks in the replace case.
5360          */
5361         if (ip6_route_mpath_should_notify(rt_notif)) {
5362                 enum fib_event_type fib_event;
5363
5364                 if (rt_notif->fib6_nsiblings != nhn - 1)
5365                         fib_event = FIB_EVENT_ENTRY_APPEND;
5366                 else
5367                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5368
5369                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5370                                                           fib_event, rt_notif,
5371                                                           nhn - 1, extack);
5372                 if (err) {
5373                         /* Delete all the siblings that were just added */
5374                         err_nh = NULL;
5375                         goto add_errout;
5376                 }
5377         }
5378
5379         /* success ... tell user about new route */
5380         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5381         goto cleanup;
5382
5383 add_errout:
5384         /* send notification for routes that were added so that
5385          * the delete notifications sent by ip6_route_del are
5386          * coherent
5387          */
5388         if (rt_notif)
5389                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5390
5391         /* Delete routes that were already added */
5392         list_for_each_entry(nh, &rt6_nh_list, next) {
5393                 if (err_nh == nh)
5394                         break;
5395                 ip6_route_del(&nh->r_cfg, extack);
5396         }
5397
5398 cleanup:
5399         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5400                 if (nh->fib6_info)
5401                         fib6_info_release(nh->fib6_info);
5402                 list_del(&nh->next);
5403                 kfree(nh);
5404         }
5405
5406         return err;
5407 }
5408
5409 static int ip6_route_multipath_del(struct fib6_config *cfg,
5410                                    struct netlink_ext_ack *extack)
5411 {
5412         struct fib6_config r_cfg;
5413         struct rtnexthop *rtnh;
5414         int last_err = 0;
5415         int remaining;
5416         int attrlen;
5417         int err;
5418
5419         remaining = cfg->fc_mp_len;
5420         rtnh = (struct rtnexthop *)cfg->fc_mp;
5421
5422         /* Parse a Multipath Entry */
5423         while (rtnh_ok(rtnh, remaining)) {
5424                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5425                 if (rtnh->rtnh_ifindex)
5426                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5427
5428                 attrlen = rtnh_attrlen(rtnh);
5429                 if (attrlen > 0) {
5430                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5431
5432                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5433                         if (nla) {
5434                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5435                                 r_cfg.fc_flags |= RTF_GATEWAY;
5436                         }
5437                 }
5438                 err = ip6_route_del(&r_cfg, extack);
5439                 if (err)
5440                         last_err = err;
5441
5442                 rtnh = rtnh_next(rtnh, &remaining);
5443         }
5444
5445         return last_err;
5446 }
5447
5448 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5449                               struct netlink_ext_ack *extack)
5450 {
5451         struct fib6_config cfg;
5452         int err;
5453
5454         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5455         if (err < 0)
5456                 return err;
5457
5458         if (cfg.fc_nh_id &&
5459             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5460                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5461                 return -EINVAL;
5462         }
5463
5464         if (cfg.fc_mp)
5465                 return ip6_route_multipath_del(&cfg, extack);
5466         else {
5467                 cfg.fc_delete_all_nh = 1;
5468                 return ip6_route_del(&cfg, extack);
5469         }
5470 }
5471
5472 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5473                               struct netlink_ext_ack *extack)
5474 {
5475         struct fib6_config cfg;
5476         int err;
5477
5478         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5479         if (err < 0)
5480                 return err;
5481
5482         if (cfg.fc_metric == 0)
5483                 cfg.fc_metric = IP6_RT_PRIO_USER;
5484
5485         if (cfg.fc_mp)
5486                 return ip6_route_multipath_add(&cfg, extack);
5487         else
5488                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5489 }
5490
5491 /* add the overhead of this fib6_nh to nexthop_len */
5492 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5493 {
5494         int *nexthop_len = arg;
5495
5496         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5497                      + NLA_ALIGN(sizeof(struct rtnexthop))
5498                      + nla_total_size(16); /* RTA_GATEWAY */
5499
5500         if (nh->fib_nh_lws) {
5501                 /* RTA_ENCAP_TYPE */
5502                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5503                 /* RTA_ENCAP */
5504                 *nexthop_len += nla_total_size(2);
5505         }
5506
5507         return 0;
5508 }
5509
5510 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5511 {
5512         int nexthop_len;
5513
5514         if (f6i->nh) {
5515                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5516                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5517                                          &nexthop_len);
5518         } else {
5519                 struct fib6_nh *nh = f6i->fib6_nh;
5520
5521                 nexthop_len = 0;
5522                 if (f6i->fib6_nsiblings) {
5523                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5524                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5525                                     + nla_total_size(16) /* RTA_GATEWAY */
5526                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5527
5528                         nexthop_len *= f6i->fib6_nsiblings;
5529                 }
5530                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5531         }
5532
5533         return NLMSG_ALIGN(sizeof(struct rtmsg))
5534                + nla_total_size(16) /* RTA_SRC */
5535                + nla_total_size(16) /* RTA_DST */
5536                + nla_total_size(16) /* RTA_GATEWAY */
5537                + nla_total_size(16) /* RTA_PREFSRC */
5538                + nla_total_size(4) /* RTA_TABLE */
5539                + nla_total_size(4) /* RTA_IIF */
5540                + nla_total_size(4) /* RTA_OIF */
5541                + nla_total_size(4) /* RTA_PRIORITY */
5542                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5543                + nla_total_size(sizeof(struct rta_cacheinfo))
5544                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5545                + nla_total_size(1) /* RTA_PREF */
5546                + nexthop_len;
5547 }
5548
5549 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5550                                  unsigned char *flags)
5551 {
5552         if (nexthop_is_multipath(nh)) {
5553                 struct nlattr *mp;
5554
5555                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5556                 if (!mp)
5557                         goto nla_put_failure;
5558
5559                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5560                         goto nla_put_failure;
5561
5562                 nla_nest_end(skb, mp);
5563         } else {
5564                 struct fib6_nh *fib6_nh;
5565
5566                 fib6_nh = nexthop_fib6_nh(nh);
5567                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5568                                      flags, false) < 0)
5569                         goto nla_put_failure;
5570         }
5571
5572         return 0;
5573
5574 nla_put_failure:
5575         return -EMSGSIZE;
5576 }
5577
5578 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5579                          struct fib6_info *rt, struct dst_entry *dst,
5580                          struct in6_addr *dest, struct in6_addr *src,
5581                          int iif, int type, u32 portid, u32 seq,
5582                          unsigned int flags)
5583 {
5584         struct rt6_info *rt6 = (struct rt6_info *)dst;
5585         struct rt6key *rt6_dst, *rt6_src;
5586         u32 *pmetrics, table, rt6_flags;
5587         unsigned char nh_flags = 0;
5588         struct nlmsghdr *nlh;
5589         struct rtmsg *rtm;
5590         long expires = 0;
5591
5592         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5593         if (!nlh)
5594                 return -EMSGSIZE;
5595
5596         if (rt6) {
5597                 rt6_dst = &rt6->rt6i_dst;
5598                 rt6_src = &rt6->rt6i_src;
5599                 rt6_flags = rt6->rt6i_flags;
5600         } else {
5601                 rt6_dst = &rt->fib6_dst;
5602                 rt6_src = &rt->fib6_src;
5603                 rt6_flags = rt->fib6_flags;
5604         }
5605
5606         rtm = nlmsg_data(nlh);
5607         rtm->rtm_family = AF_INET6;
5608         rtm->rtm_dst_len = rt6_dst->plen;
5609         rtm->rtm_src_len = rt6_src->plen;
5610         rtm->rtm_tos = 0;
5611         if (rt->fib6_table)
5612                 table = rt->fib6_table->tb6_id;
5613         else
5614                 table = RT6_TABLE_UNSPEC;
5615         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5616         if (nla_put_u32(skb, RTA_TABLE, table))
5617                 goto nla_put_failure;
5618
5619         rtm->rtm_type = rt->fib6_type;
5620         rtm->rtm_flags = 0;
5621         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5622         rtm->rtm_protocol = rt->fib6_protocol;
5623
5624         if (rt6_flags & RTF_CACHE)
5625                 rtm->rtm_flags |= RTM_F_CLONED;
5626
5627         if (dest) {
5628                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5629                         goto nla_put_failure;
5630                 rtm->rtm_dst_len = 128;
5631         } else if (rtm->rtm_dst_len)
5632                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5633                         goto nla_put_failure;
5634 #ifdef CONFIG_IPV6_SUBTREES
5635         if (src) {
5636                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5637                         goto nla_put_failure;
5638                 rtm->rtm_src_len = 128;
5639         } else if (rtm->rtm_src_len &&
5640                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5641                 goto nla_put_failure;
5642 #endif
5643         if (iif) {
5644 #ifdef CONFIG_IPV6_MROUTE
5645                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5646                         int err = ip6mr_get_route(net, skb, rtm, portid);
5647
5648                         if (err == 0)
5649                                 return 0;
5650                         if (err < 0)
5651                                 goto nla_put_failure;
5652                 } else
5653 #endif
5654                         if (nla_put_u32(skb, RTA_IIF, iif))
5655                                 goto nla_put_failure;
5656         } else if (dest) {
5657                 struct in6_addr saddr_buf;
5658                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5659                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5660                         goto nla_put_failure;
5661         }
5662
5663         if (rt->fib6_prefsrc.plen) {
5664                 struct in6_addr saddr_buf;
5665                 saddr_buf = rt->fib6_prefsrc.addr;
5666                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5667                         goto nla_put_failure;
5668         }
5669
5670         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5671         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5672                 goto nla_put_failure;
5673
5674         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5675                 goto nla_put_failure;
5676
5677         /* For multipath routes, walk the siblings list and add
5678          * each as a nexthop within RTA_MULTIPATH.
5679          */
5680         if (rt6) {
5681                 if (rt6_flags & RTF_GATEWAY &&
5682                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5683                         goto nla_put_failure;
5684
5685                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5686                         goto nla_put_failure;
5687
5688                 if (dst->lwtstate &&
5689                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5690                         goto nla_put_failure;
5691         } else if (rt->fib6_nsiblings) {
5692                 struct fib6_info *sibling, *next_sibling;
5693                 struct nlattr *mp;
5694
5695                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5696                 if (!mp)
5697                         goto nla_put_failure;
5698
5699                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5700                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5701                         goto nla_put_failure;
5702
5703                 list_for_each_entry_safe(sibling, next_sibling,
5704                                          &rt->fib6_siblings, fib6_siblings) {
5705                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5706                                             sibling->fib6_nh->fib_nh_weight,
5707                                             AF_INET6) < 0)
5708                                 goto nla_put_failure;
5709                 }
5710
5711                 nla_nest_end(skb, mp);
5712         } else if (rt->nh) {
5713                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5714                         goto nla_put_failure;
5715
5716                 if (nexthop_is_blackhole(rt->nh))
5717                         rtm->rtm_type = RTN_BLACKHOLE;
5718
5719                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5720                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5721                         goto nla_put_failure;
5722
5723                 rtm->rtm_flags |= nh_flags;
5724         } else {
5725                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5726                                      &nh_flags, false) < 0)
5727                         goto nla_put_failure;
5728
5729                 rtm->rtm_flags |= nh_flags;
5730         }
5731
5732         if (rt6_flags & RTF_EXPIRES) {
5733                 expires = dst ? dst->expires : rt->expires;
5734                 expires -= jiffies;
5735         }
5736
5737         if (!dst) {
5738                 if (rt->offload)
5739                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5740                 if (rt->trap)
5741                         rtm->rtm_flags |= RTM_F_TRAP;
5742                 if (rt->offload_failed)
5743                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5744         }
5745
5746         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5747                 goto nla_put_failure;
5748
5749         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5750                 goto nla_put_failure;
5751
5752
5753         nlmsg_end(skb, nlh);
5754         return 0;
5755
5756 nla_put_failure:
5757         nlmsg_cancel(skb, nlh);
5758         return -EMSGSIZE;
5759 }
5760
5761 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5762 {
5763         const struct net_device *dev = arg;
5764
5765         if (nh->fib_nh_dev == dev)
5766                 return 1;
5767
5768         return 0;
5769 }
5770
5771 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5772                                const struct net_device *dev)
5773 {
5774         if (f6i->nh) {
5775                 struct net_device *_dev = (struct net_device *)dev;
5776
5777                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5778                                                   fib6_info_nh_uses_dev,
5779                                                   _dev);
5780         }
5781
5782         if (f6i->fib6_nh->fib_nh_dev == dev)
5783                 return true;
5784
5785         if (f6i->fib6_nsiblings) {
5786                 struct fib6_info *sibling, *next_sibling;
5787
5788                 list_for_each_entry_safe(sibling, next_sibling,
5789                                          &f6i->fib6_siblings, fib6_siblings) {
5790                         if (sibling->fib6_nh->fib_nh_dev == dev)
5791                                 return true;
5792                 }
5793         }
5794
5795         return false;
5796 }
5797
5798 struct fib6_nh_exception_dump_walker {
5799         struct rt6_rtnl_dump_arg *dump;
5800         struct fib6_info *rt;
5801         unsigned int flags;
5802         unsigned int skip;
5803         unsigned int count;
5804 };
5805
5806 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5807 {
5808         struct fib6_nh_exception_dump_walker *w = arg;
5809         struct rt6_rtnl_dump_arg *dump = w->dump;
5810         struct rt6_exception_bucket *bucket;
5811         struct rt6_exception *rt6_ex;
5812         int i, err;
5813
5814         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5815         if (!bucket)
5816                 return 0;
5817
5818         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5819                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5820                         if (w->skip) {
5821                                 w->skip--;
5822                                 continue;
5823                         }
5824
5825                         /* Expiration of entries doesn't bump sernum, insertion
5826                          * does. Removal is triggered by insertion, so we can
5827                          * rely on the fact that if entries change between two
5828                          * partial dumps, this node is scanned again completely,
5829                          * see rt6_insert_exception() and fib6_dump_table().
5830                          *
5831                          * Count expired entries we go through as handled
5832                          * entries that we'll skip next time, in case of partial
5833                          * node dump. Otherwise, if entries expire meanwhile,
5834                          * we'll skip the wrong amount.
5835                          */
5836                         if (rt6_check_expired(rt6_ex->rt6i)) {
5837                                 w->count++;
5838                                 continue;
5839                         }
5840
5841                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5842                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5843                                             RTM_NEWROUTE,
5844                                             NETLINK_CB(dump->cb->skb).portid,
5845                                             dump->cb->nlh->nlmsg_seq, w->flags);
5846                         if (err)
5847                                 return err;
5848
5849                         w->count++;
5850                 }
5851                 bucket++;
5852         }
5853
5854         return 0;
5855 }
5856
5857 /* Return -1 if done with node, number of handled routes on partial dump */
5858 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5859 {
5860         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5861         struct fib_dump_filter *filter = &arg->filter;
5862         unsigned int flags = NLM_F_MULTI;
5863         struct net *net = arg->net;
5864         int count = 0;
5865
5866         if (rt == net->ipv6.fib6_null_entry)
5867                 return -1;
5868
5869         if ((filter->flags & RTM_F_PREFIX) &&
5870             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5871                 /* success since this is not a prefix route */
5872                 return -1;
5873         }
5874         if (filter->filter_set &&
5875             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5876              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5877              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5878                 return -1;
5879         }
5880
5881         if (filter->filter_set ||
5882             !filter->dump_routes || !filter->dump_exceptions) {
5883                 flags |= NLM_F_DUMP_FILTERED;
5884         }
5885
5886         if (filter->dump_routes) {
5887                 if (skip) {
5888                         skip--;
5889                 } else {
5890                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5891                                           0, RTM_NEWROUTE,
5892                                           NETLINK_CB(arg->cb->skb).portid,
5893                                           arg->cb->nlh->nlmsg_seq, flags)) {
5894                                 return 0;
5895                         }
5896                         count++;
5897                 }
5898         }
5899
5900         if (filter->dump_exceptions) {
5901                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5902                                                            .rt = rt,
5903                                                            .flags = flags,
5904                                                            .skip = skip,
5905                                                            .count = 0 };
5906                 int err;
5907
5908                 rcu_read_lock();
5909                 if (rt->nh) {
5910                         err = nexthop_for_each_fib6_nh(rt->nh,
5911                                                        rt6_nh_dump_exceptions,
5912                                                        &w);
5913                 } else {
5914                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5915                 }
5916                 rcu_read_unlock();
5917
5918                 if (err)
5919                         return count += w.count;
5920         }
5921
5922         return -1;
5923 }
5924
5925 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5926                                         const struct nlmsghdr *nlh,
5927                                         struct nlattr **tb,
5928                                         struct netlink_ext_ack *extack)
5929 {
5930         struct rtmsg *rtm;
5931         int i, err;
5932
5933         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5934                 NL_SET_ERR_MSG_MOD(extack,
5935                                    "Invalid header for get route request");
5936                 return -EINVAL;
5937         }
5938
5939         if (!netlink_strict_get_check(skb))
5940                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5941                                               rtm_ipv6_policy, extack);
5942
5943         rtm = nlmsg_data(nlh);
5944         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5945             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5946             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5947             rtm->rtm_type) {
5948                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5949                 return -EINVAL;
5950         }
5951         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5952                 NL_SET_ERR_MSG_MOD(extack,
5953                                    "Invalid flags for get route request");
5954                 return -EINVAL;
5955         }
5956
5957         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5958                                             rtm_ipv6_policy, extack);
5959         if (err)
5960                 return err;
5961
5962         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5963             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5964                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5965                 return -EINVAL;
5966         }
5967
5968         for (i = 0; i <= RTA_MAX; i++) {
5969                 if (!tb[i])
5970                         continue;
5971
5972                 switch (i) {
5973                 case RTA_SRC:
5974                 case RTA_DST:
5975                 case RTA_IIF:
5976                 case RTA_OIF:
5977                 case RTA_MARK:
5978                 case RTA_UID:
5979                 case RTA_SPORT:
5980                 case RTA_DPORT:
5981                 case RTA_IP_PROTO:
5982                         break;
5983                 default:
5984                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5985                         return -EINVAL;
5986                 }
5987         }
5988
5989         return 0;
5990 }
5991
5992 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5993                               struct netlink_ext_ack *extack)
5994 {
5995         struct net *net = sock_net(in_skb->sk);
5996         struct nlattr *tb[RTA_MAX+1];
5997         int err, iif = 0, oif = 0;
5998         struct fib6_info *from;
5999         struct dst_entry *dst;
6000         struct rt6_info *rt;
6001         struct sk_buff *skb;
6002         struct rtmsg *rtm;
6003         struct flowi6 fl6 = {};
6004         bool fibmatch;
6005
6006         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6007         if (err < 0)
6008                 goto errout;
6009
6010         err = -EINVAL;
6011         rtm = nlmsg_data(nlh);
6012         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6013         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6014
6015         if (tb[RTA_SRC]) {
6016                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6017                         goto errout;
6018
6019                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6020         }
6021
6022         if (tb[RTA_DST]) {
6023                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6024                         goto errout;
6025
6026                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6027         }
6028
6029         if (tb[RTA_IIF])
6030                 iif = nla_get_u32(tb[RTA_IIF]);
6031
6032         if (tb[RTA_OIF])
6033                 oif = nla_get_u32(tb[RTA_OIF]);
6034
6035         if (tb[RTA_MARK])
6036                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6037
6038         if (tb[RTA_UID])
6039                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6040                                            nla_get_u32(tb[RTA_UID]));
6041         else
6042                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6043
6044         if (tb[RTA_SPORT])
6045                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6046
6047         if (tb[RTA_DPORT])
6048                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6049
6050         if (tb[RTA_IP_PROTO]) {
6051                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6052                                                   &fl6.flowi6_proto, AF_INET6,
6053                                                   extack);
6054                 if (err)
6055                         goto errout;
6056         }
6057
6058         if (iif) {
6059                 struct net_device *dev;
6060                 int flags = 0;
6061
6062                 rcu_read_lock();
6063
6064                 dev = dev_get_by_index_rcu(net, iif);
6065                 if (!dev) {
6066                         rcu_read_unlock();
6067                         err = -ENODEV;
6068                         goto errout;
6069                 }
6070
6071                 fl6.flowi6_iif = iif;
6072
6073                 if (!ipv6_addr_any(&fl6.saddr))
6074                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6075
6076                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6077
6078                 rcu_read_unlock();
6079         } else {
6080                 fl6.flowi6_oif = oif;
6081
6082                 dst = ip6_route_output(net, NULL, &fl6);
6083         }
6084
6085
6086         rt = container_of(dst, struct rt6_info, dst);
6087         if (rt->dst.error) {
6088                 err = rt->dst.error;
6089                 ip6_rt_put(rt);
6090                 goto errout;
6091         }
6092
6093         if (rt == net->ipv6.ip6_null_entry) {
6094                 err = rt->dst.error;
6095                 ip6_rt_put(rt);
6096                 goto errout;
6097         }
6098
6099         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6100         if (!skb) {
6101                 ip6_rt_put(rt);
6102                 err = -ENOBUFS;
6103                 goto errout;
6104         }
6105
6106         skb_dst_set(skb, &rt->dst);
6107
6108         rcu_read_lock();
6109         from = rcu_dereference(rt->from);
6110         if (from) {
6111                 if (fibmatch)
6112                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6113                                             iif, RTM_NEWROUTE,
6114                                             NETLINK_CB(in_skb).portid,
6115                                             nlh->nlmsg_seq, 0);
6116                 else
6117                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6118                                             &fl6.saddr, iif, RTM_NEWROUTE,
6119                                             NETLINK_CB(in_skb).portid,
6120                                             nlh->nlmsg_seq, 0);
6121         } else {
6122                 err = -ENETUNREACH;
6123         }
6124         rcu_read_unlock();
6125
6126         if (err < 0) {
6127                 kfree_skb(skb);
6128                 goto errout;
6129         }
6130
6131         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6132 errout:
6133         return err;
6134 }
6135
6136 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6137                      unsigned int nlm_flags)
6138 {
6139         struct sk_buff *skb;
6140         struct net *net = info->nl_net;
6141         u32 seq;
6142         int err;
6143
6144         err = -ENOBUFS;
6145         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6146
6147         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6148         if (!skb)
6149                 goto errout;
6150
6151         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6152                             event, info->portid, seq, nlm_flags);
6153         if (err < 0) {
6154                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6155                 WARN_ON(err == -EMSGSIZE);
6156                 kfree_skb(skb);
6157                 goto errout;
6158         }
6159         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6160                     info->nlh, gfp_any());
6161         return;
6162 errout:
6163         if (err < 0)
6164                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6165 }
6166
6167 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6168                     struct nl_info *info)
6169 {
6170         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6171         struct sk_buff *skb;
6172         int err = -ENOBUFS;
6173
6174         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6175         if (!skb)
6176                 goto errout;
6177
6178         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6179                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6180         if (err < 0) {
6181                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6182                 WARN_ON(err == -EMSGSIZE);
6183                 kfree_skb(skb);
6184                 goto errout;
6185         }
6186         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6187                     info->nlh, gfp_any());
6188         return;
6189 errout:
6190         if (err < 0)
6191                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6192 }
6193
6194 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6195                             bool offload, bool trap, bool offload_failed)
6196 {
6197         struct sk_buff *skb;
6198         int err;
6199
6200         if (f6i->offload == offload && f6i->trap == trap &&
6201             f6i->offload_failed == offload_failed)
6202                 return;
6203
6204         f6i->offload = offload;
6205         f6i->trap = trap;
6206
6207         /* 2 means send notifications only if offload_failed was changed. */
6208         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6209             f6i->offload_failed == offload_failed)
6210                 return;
6211
6212         f6i->offload_failed = offload_failed;
6213
6214         if (!rcu_access_pointer(f6i->fib6_node))
6215                 /* The route was removed from the tree, do not send
6216                  * notification.
6217                  */
6218                 return;
6219
6220         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6221                 return;
6222
6223         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6224         if (!skb) {
6225                 err = -ENOBUFS;
6226                 goto errout;
6227         }
6228
6229         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6230                             0, 0);
6231         if (err < 0) {
6232                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6233                 WARN_ON(err == -EMSGSIZE);
6234                 kfree_skb(skb);
6235                 goto errout;
6236         }
6237
6238         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6239         return;
6240
6241 errout:
6242         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6243 }
6244 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6245
6246 static int ip6_route_dev_notify(struct notifier_block *this,
6247                                 unsigned long event, void *ptr)
6248 {
6249         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6250         struct net *net = dev_net(dev);
6251
6252         if (!(dev->flags & IFF_LOOPBACK))
6253                 return NOTIFY_OK;
6254
6255         if (event == NETDEV_REGISTER) {
6256                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6257                 net->ipv6.ip6_null_entry->dst.dev = dev;
6258                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6259 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6260                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6261                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6262                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6263                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6264 #endif
6265          } else if (event == NETDEV_UNREGISTER &&
6266                     dev->reg_state != NETREG_UNREGISTERED) {
6267                 /* NETDEV_UNREGISTER could be fired for multiple times by
6268                  * netdev_wait_allrefs(). Make sure we only call this once.
6269                  */
6270                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6271 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6272                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6273                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6274 #endif
6275         }
6276
6277         return NOTIFY_OK;
6278 }
6279
6280 /*
6281  *      /proc
6282  */
6283
6284 #ifdef CONFIG_PROC_FS
6285 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6286 {
6287         struct net *net = (struct net *)seq->private;
6288         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6289                    net->ipv6.rt6_stats->fib_nodes,
6290                    net->ipv6.rt6_stats->fib_route_nodes,
6291                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6292                    net->ipv6.rt6_stats->fib_rt_entries,
6293                    net->ipv6.rt6_stats->fib_rt_cache,
6294                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6295                    net->ipv6.rt6_stats->fib_discarded_routes);
6296
6297         return 0;
6298 }
6299 #endif  /* CONFIG_PROC_FS */
6300
6301 #ifdef CONFIG_SYSCTL
6302
6303 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6304                               void *buffer, size_t *lenp, loff_t *ppos)
6305 {
6306         struct net *net;
6307         int delay;
6308         int ret;
6309         if (!write)
6310                 return -EINVAL;
6311
6312         net = (struct net *)ctl->extra1;
6313         delay = net->ipv6.sysctl.flush_delay;
6314         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6315         if (ret)
6316                 return ret;
6317
6318         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6319         return 0;
6320 }
6321
6322 static struct ctl_table ipv6_route_table_template[] = {
6323         {
6324                 .procname       =       "flush",
6325                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6326                 .maxlen         =       sizeof(int),
6327                 .mode           =       0200,
6328                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6329         },
6330         {
6331                 .procname       =       "gc_thresh",
6332                 .data           =       &ip6_dst_ops_template.gc_thresh,
6333                 .maxlen         =       sizeof(int),
6334                 .mode           =       0644,
6335                 .proc_handler   =       proc_dointvec,
6336         },
6337         {
6338                 .procname       =       "max_size",
6339                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6340                 .maxlen         =       sizeof(int),
6341                 .mode           =       0644,
6342                 .proc_handler   =       proc_dointvec,
6343         },
6344         {
6345                 .procname       =       "gc_min_interval",
6346                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6347                 .maxlen         =       sizeof(int),
6348                 .mode           =       0644,
6349                 .proc_handler   =       proc_dointvec_jiffies,
6350         },
6351         {
6352                 .procname       =       "gc_timeout",
6353                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6354                 .maxlen         =       sizeof(int),
6355                 .mode           =       0644,
6356                 .proc_handler   =       proc_dointvec_jiffies,
6357         },
6358         {
6359                 .procname       =       "gc_interval",
6360                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6361                 .maxlen         =       sizeof(int),
6362                 .mode           =       0644,
6363                 .proc_handler   =       proc_dointvec_jiffies,
6364         },
6365         {
6366                 .procname       =       "gc_elasticity",
6367                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6368                 .maxlen         =       sizeof(int),
6369                 .mode           =       0644,
6370                 .proc_handler   =       proc_dointvec,
6371         },
6372         {
6373                 .procname       =       "mtu_expires",
6374                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6375                 .maxlen         =       sizeof(int),
6376                 .mode           =       0644,
6377                 .proc_handler   =       proc_dointvec_jiffies,
6378         },
6379         {
6380                 .procname       =       "min_adv_mss",
6381                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6382                 .maxlen         =       sizeof(int),
6383                 .mode           =       0644,
6384                 .proc_handler   =       proc_dointvec,
6385         },
6386         {
6387                 .procname       =       "gc_min_interval_ms",
6388                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6389                 .maxlen         =       sizeof(int),
6390                 .mode           =       0644,
6391                 .proc_handler   =       proc_dointvec_ms_jiffies,
6392         },
6393         {
6394                 .procname       =       "skip_notify_on_dev_down",
6395                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6396                 .maxlen         =       sizeof(int),
6397                 .mode           =       0644,
6398                 .proc_handler   =       proc_dointvec_minmax,
6399                 .extra1         =       SYSCTL_ZERO,
6400                 .extra2         =       SYSCTL_ONE,
6401         },
6402         { }
6403 };
6404
6405 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6406 {
6407         struct ctl_table *table;
6408
6409         table = kmemdup(ipv6_route_table_template,
6410                         sizeof(ipv6_route_table_template),
6411                         GFP_KERNEL);
6412
6413         if (table) {
6414                 table[0].data = &net->ipv6.sysctl.flush_delay;
6415                 table[0].extra1 = net;
6416                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6417                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6418                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6419                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6420                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6421                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6422                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6423                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6424                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6425                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6426
6427                 /* Don't export sysctls to unprivileged users */
6428                 if (net->user_ns != &init_user_ns)
6429                         table[0].procname = NULL;
6430         }
6431
6432         return table;
6433 }
6434 #endif
6435
6436 static int __net_init ip6_route_net_init(struct net *net)
6437 {
6438         int ret = -ENOMEM;
6439
6440         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6441                sizeof(net->ipv6.ip6_dst_ops));
6442
6443         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6444                 goto out_ip6_dst_ops;
6445
6446         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6447         if (!net->ipv6.fib6_null_entry)
6448                 goto out_ip6_dst_entries;
6449         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6450                sizeof(*net->ipv6.fib6_null_entry));
6451
6452         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6453                                            sizeof(*net->ipv6.ip6_null_entry),
6454                                            GFP_KERNEL);
6455         if (!net->ipv6.ip6_null_entry)
6456                 goto out_fib6_null_entry;
6457         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6458         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6459                          ip6_template_metrics, true);
6460         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6461
6462 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6463         net->ipv6.fib6_has_custom_rules = false;
6464         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6465                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6466                                                GFP_KERNEL);
6467         if (!net->ipv6.ip6_prohibit_entry)
6468                 goto out_ip6_null_entry;
6469         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6470         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6471                          ip6_template_metrics, true);
6472         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6473
6474         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6475                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6476                                                GFP_KERNEL);
6477         if (!net->ipv6.ip6_blk_hole_entry)
6478                 goto out_ip6_prohibit_entry;
6479         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6480         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6481                          ip6_template_metrics, true);
6482         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6483 #ifdef CONFIG_IPV6_SUBTREES
6484         net->ipv6.fib6_routes_require_src = 0;
6485 #endif
6486 #endif
6487
6488         net->ipv6.sysctl.flush_delay = 0;
6489         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6490         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6491         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6492         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6493         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6494         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6495         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6496         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6497
6498         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6499
6500         ret = 0;
6501 out:
6502         return ret;
6503
6504 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6505 out_ip6_prohibit_entry:
6506         kfree(net->ipv6.ip6_prohibit_entry);
6507 out_ip6_null_entry:
6508         kfree(net->ipv6.ip6_null_entry);
6509 #endif
6510 out_fib6_null_entry:
6511         kfree(net->ipv6.fib6_null_entry);
6512 out_ip6_dst_entries:
6513         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6514 out_ip6_dst_ops:
6515         goto out;
6516 }
6517
6518 static void __net_exit ip6_route_net_exit(struct net *net)
6519 {
6520         kfree(net->ipv6.fib6_null_entry);
6521         kfree(net->ipv6.ip6_null_entry);
6522 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6523         kfree(net->ipv6.ip6_prohibit_entry);
6524         kfree(net->ipv6.ip6_blk_hole_entry);
6525 #endif
6526         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6527 }
6528
6529 static int __net_init ip6_route_net_init_late(struct net *net)
6530 {
6531 #ifdef CONFIG_PROC_FS
6532         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6533                         sizeof(struct ipv6_route_iter));
6534         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6535                         rt6_stats_seq_show, NULL);
6536 #endif
6537         return 0;
6538 }
6539
6540 static void __net_exit ip6_route_net_exit_late(struct net *net)
6541 {
6542 #ifdef CONFIG_PROC_FS
6543         remove_proc_entry("ipv6_route", net->proc_net);
6544         remove_proc_entry("rt6_stats", net->proc_net);
6545 #endif
6546 }
6547
6548 static struct pernet_operations ip6_route_net_ops = {
6549         .init = ip6_route_net_init,
6550         .exit = ip6_route_net_exit,
6551 };
6552
6553 static int __net_init ipv6_inetpeer_init(struct net *net)
6554 {
6555         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6556
6557         if (!bp)
6558                 return -ENOMEM;
6559         inet_peer_base_init(bp);
6560         net->ipv6.peers = bp;
6561         return 0;
6562 }
6563
6564 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6565 {
6566         struct inet_peer_base *bp = net->ipv6.peers;
6567
6568         net->ipv6.peers = NULL;
6569         inetpeer_invalidate_tree(bp);
6570         kfree(bp);
6571 }
6572
6573 static struct pernet_operations ipv6_inetpeer_ops = {
6574         .init   =       ipv6_inetpeer_init,
6575         .exit   =       ipv6_inetpeer_exit,
6576 };
6577
6578 static struct pernet_operations ip6_route_net_late_ops = {
6579         .init = ip6_route_net_init_late,
6580         .exit = ip6_route_net_exit_late,
6581 };
6582
6583 static struct notifier_block ip6_route_dev_notifier = {
6584         .notifier_call = ip6_route_dev_notify,
6585         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6586 };
6587
6588 void __init ip6_route_init_special_entries(void)
6589 {
6590         /* Registering of the loopback is done before this portion of code,
6591          * the loopback reference in rt6_info will not be taken, do it
6592          * manually for init_net */
6593         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6594         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6595         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6596   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6597         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6598         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6599         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6600         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6601   #endif
6602 }
6603
6604 #if IS_BUILTIN(CONFIG_IPV6)
6605 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6606 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6607
6608 BTF_ID_LIST(btf_fib6_info_id)
6609 BTF_ID(struct, fib6_info)
6610
6611 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6612         .seq_ops                = &ipv6_route_seq_ops,
6613         .init_seq_private       = bpf_iter_init_seq_net,
6614         .fini_seq_private       = bpf_iter_fini_seq_net,
6615         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6616 };
6617
6618 static struct bpf_iter_reg ipv6_route_reg_info = {
6619         .target                 = "ipv6_route",
6620         .ctx_arg_info_size      = 1,
6621         .ctx_arg_info           = {
6622                 { offsetof(struct bpf_iter__ipv6_route, rt),
6623                   PTR_TO_BTF_ID_OR_NULL },
6624         },
6625         .seq_info               = &ipv6_route_seq_info,
6626 };
6627
6628 static int __init bpf_iter_register(void)
6629 {
6630         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6631         return bpf_iter_reg_target(&ipv6_route_reg_info);
6632 }
6633
6634 static void bpf_iter_unregister(void)
6635 {
6636         bpf_iter_unreg_target(&ipv6_route_reg_info);
6637 }
6638 #endif
6639 #endif
6640
6641 int __init ip6_route_init(void)
6642 {
6643         int ret;
6644         int cpu;
6645
6646         ret = -ENOMEM;
6647         ip6_dst_ops_template.kmem_cachep =
6648                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6649                                   SLAB_HWCACHE_ALIGN, NULL);
6650         if (!ip6_dst_ops_template.kmem_cachep)
6651                 goto out;
6652
6653         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6654         if (ret)
6655                 goto out_kmem_cache;
6656
6657         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6658         if (ret)
6659                 goto out_dst_entries;
6660
6661         ret = register_pernet_subsys(&ip6_route_net_ops);
6662         if (ret)
6663                 goto out_register_inetpeer;
6664
6665         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6666
6667         ret = fib6_init();
6668         if (ret)
6669                 goto out_register_subsys;
6670
6671         ret = xfrm6_init();
6672         if (ret)
6673                 goto out_fib6_init;
6674
6675         ret = fib6_rules_init();
6676         if (ret)
6677                 goto xfrm6_init;
6678
6679         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6680         if (ret)
6681                 goto fib6_rules_init;
6682
6683         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6684                                    inet6_rtm_newroute, NULL, 0);
6685         if (ret < 0)
6686                 goto out_register_late_subsys;
6687
6688         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6689                                    inet6_rtm_delroute, NULL, 0);
6690         if (ret < 0)
6691                 goto out_register_late_subsys;
6692
6693         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6694                                    inet6_rtm_getroute, NULL,
6695                                    RTNL_FLAG_DOIT_UNLOCKED);
6696         if (ret < 0)
6697                 goto out_register_late_subsys;
6698
6699         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6700         if (ret)
6701                 goto out_register_late_subsys;
6702
6703 #if IS_BUILTIN(CONFIG_IPV6)
6704 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6705         ret = bpf_iter_register();
6706         if (ret)
6707                 goto out_register_late_subsys;
6708 #endif
6709 #endif
6710
6711         for_each_possible_cpu(cpu) {
6712                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6713
6714                 INIT_LIST_HEAD(&ul->head);
6715                 spin_lock_init(&ul->lock);
6716         }
6717
6718 out:
6719         return ret;
6720
6721 out_register_late_subsys:
6722         rtnl_unregister_all(PF_INET6);
6723         unregister_pernet_subsys(&ip6_route_net_late_ops);
6724 fib6_rules_init:
6725         fib6_rules_cleanup();
6726 xfrm6_init:
6727         xfrm6_fini();
6728 out_fib6_init:
6729         fib6_gc_cleanup();
6730 out_register_subsys:
6731         unregister_pernet_subsys(&ip6_route_net_ops);
6732 out_register_inetpeer:
6733         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6734 out_dst_entries:
6735         dst_entries_destroy(&ip6_dst_blackhole_ops);
6736 out_kmem_cache:
6737         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6738         goto out;
6739 }
6740
6741 void ip6_route_cleanup(void)
6742 {
6743 #if IS_BUILTIN(CONFIG_IPV6)
6744 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6745         bpf_iter_unregister();
6746 #endif
6747 #endif
6748         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6749         unregister_pernet_subsys(&ip6_route_net_late_ops);
6750         fib6_rules_cleanup();
6751         xfrm6_fini();
6752         fib6_gc_cleanup();
6753         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6754         unregister_pernet_subsys(&ip6_route_net_ops);
6755         dst_entries_destroy(&ip6_dst_blackhole_ops);
6756         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6757 }