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         return ip6_dst_mtu_maybe_forward(dst, false);
3213 }
3214 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3215
3216 /* MTU selection:
3217  * 1. mtu on route is locked - use it
3218  * 2. mtu from nexthop exception
3219  * 3. mtu from egress device
3220  *
3221  * based on ip6_dst_mtu_forward and exception logic of
3222  * rt6_find_cached_rt; called with rcu_read_lock
3223  */
3224 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3225                       const struct in6_addr *daddr,
3226                       const struct in6_addr *saddr)
3227 {
3228         const struct fib6_nh *nh = res->nh;
3229         struct fib6_info *f6i = res->f6i;
3230         struct inet6_dev *idev;
3231         struct rt6_info *rt;
3232         u32 mtu = 0;
3233
3234         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3235                 mtu = f6i->fib6_pmtu;
3236                 if (mtu)
3237                         goto out;
3238         }
3239
3240         rt = rt6_find_cached_rt(res, daddr, saddr);
3241         if (unlikely(rt)) {
3242                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3243         } else {
3244                 struct net_device *dev = nh->fib_nh_dev;
3245
3246                 mtu = IPV6_MIN_MTU;
3247                 idev = __in6_dev_get(dev);
3248                 if (idev && idev->cnf.mtu6 > mtu)
3249                         mtu = idev->cnf.mtu6;
3250         }
3251
3252         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3253 out:
3254         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3255 }
3256
3257 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3258                                   struct flowi6 *fl6)
3259 {
3260         struct dst_entry *dst;
3261         struct rt6_info *rt;
3262         struct inet6_dev *idev = in6_dev_get(dev);
3263         struct net *net = dev_net(dev);
3264
3265         if (unlikely(!idev))
3266                 return ERR_PTR(-ENODEV);
3267
3268         rt = ip6_dst_alloc(net, dev, 0);
3269         if (unlikely(!rt)) {
3270                 in6_dev_put(idev);
3271                 dst = ERR_PTR(-ENOMEM);
3272                 goto out;
3273         }
3274
3275         rt->dst.input = ip6_input;
3276         rt->dst.output  = ip6_output;
3277         rt->rt6i_gateway  = fl6->daddr;
3278         rt->rt6i_dst.addr = fl6->daddr;
3279         rt->rt6i_dst.plen = 128;
3280         rt->rt6i_idev     = idev;
3281         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3282
3283         /* Add this dst into uncached_list so that rt6_disable_ip() can
3284          * do proper release of the net_device
3285          */
3286         rt6_uncached_list_add(rt);
3287         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3288
3289         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3290
3291 out:
3292         return dst;
3293 }
3294
3295 static int ip6_dst_gc(struct dst_ops *ops)
3296 {
3297         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3298         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3299         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3300         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3301         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3302         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3303         int entries;
3304
3305         entries = dst_entries_get_fast(ops);
3306         if (entries > rt_max_size)
3307                 entries = dst_entries_get_slow(ops);
3308
3309         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3310             entries <= rt_max_size)
3311                 goto out;
3312
3313         net->ipv6.ip6_rt_gc_expire++;
3314         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3315         entries = dst_entries_get_slow(ops);
3316         if (entries < ops->gc_thresh)
3317                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3318 out:
3319         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3320         return entries > rt_max_size;
3321 }
3322
3323 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3324                                const struct in6_addr *gw_addr, u32 tbid,
3325                                int flags, struct fib6_result *res)
3326 {
3327         struct flowi6 fl6 = {
3328                 .flowi6_oif = cfg->fc_ifindex,
3329                 .daddr = *gw_addr,
3330                 .saddr = cfg->fc_prefsrc,
3331         };
3332         struct fib6_table *table;
3333         int err;
3334
3335         table = fib6_get_table(net, tbid);
3336         if (!table)
3337                 return -EINVAL;
3338
3339         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3340                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3341
3342         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3343
3344         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3345         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3346                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3347                                  cfg->fc_ifindex != 0, NULL, flags);
3348
3349         return err;
3350 }
3351
3352 static int ip6_route_check_nh_onlink(struct net *net,
3353                                      struct fib6_config *cfg,
3354                                      const struct net_device *dev,
3355                                      struct netlink_ext_ack *extack)
3356 {
3357         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3358         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3359         struct fib6_result res = {};
3360         int err;
3361
3362         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3363         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3364             /* ignore match if it is the default route */
3365             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3366             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3367                 NL_SET_ERR_MSG(extack,
3368                                "Nexthop has invalid gateway or device mismatch");
3369                 err = -EINVAL;
3370         }
3371
3372         return err;
3373 }
3374
3375 static int ip6_route_check_nh(struct net *net,
3376                               struct fib6_config *cfg,
3377                               struct net_device **_dev,
3378                               struct inet6_dev **idev)
3379 {
3380         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3381         struct net_device *dev = _dev ? *_dev : NULL;
3382         int flags = RT6_LOOKUP_F_IFACE;
3383         struct fib6_result res = {};
3384         int err = -EHOSTUNREACH;
3385
3386         if (cfg->fc_table) {
3387                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3388                                           cfg->fc_table, flags, &res);
3389                 /* gw_addr can not require a gateway or resolve to a reject
3390                  * route. If a device is given, it must match the result.
3391                  */
3392                 if (err || res.fib6_flags & RTF_REJECT ||
3393                     res.nh->fib_nh_gw_family ||
3394                     (dev && dev != res.nh->fib_nh_dev))
3395                         err = -EHOSTUNREACH;
3396         }
3397
3398         if (err < 0) {
3399                 struct flowi6 fl6 = {
3400                         .flowi6_oif = cfg->fc_ifindex,
3401                         .daddr = *gw_addr,
3402                 };
3403
3404                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3405                 if (err || res.fib6_flags & RTF_REJECT ||
3406                     res.nh->fib_nh_gw_family)
3407                         err = -EHOSTUNREACH;
3408
3409                 if (err)
3410                         return err;
3411
3412                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3413                                  cfg->fc_ifindex != 0, NULL, flags);
3414         }
3415
3416         err = 0;
3417         if (dev) {
3418                 if (dev != res.nh->fib_nh_dev)
3419                         err = -EHOSTUNREACH;
3420         } else {
3421                 *_dev = dev = res.nh->fib_nh_dev;
3422                 dev_hold(dev);
3423                 *idev = in6_dev_get(dev);
3424         }
3425
3426         return err;
3427 }
3428
3429 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3430                            struct net_device **_dev, struct inet6_dev **idev,
3431                            struct netlink_ext_ack *extack)
3432 {
3433         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3434         int gwa_type = ipv6_addr_type(gw_addr);
3435         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3436         const struct net_device *dev = *_dev;
3437         bool need_addr_check = !dev;
3438         int err = -EINVAL;
3439
3440         /* if gw_addr is local we will fail to detect this in case
3441          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3442          * will return already-added prefix route via interface that
3443          * prefix route was assigned to, which might be non-loopback.
3444          */
3445         if (dev &&
3446             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3447                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3448                 goto out;
3449         }
3450
3451         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3452                 /* IPv6 strictly inhibits using not link-local
3453                  * addresses as nexthop address.
3454                  * Otherwise, router will not able to send redirects.
3455                  * It is very good, but in some (rare!) circumstances
3456                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3457                  * some exceptions. --ANK
3458                  * We allow IPv4-mapped nexthops to support RFC4798-type
3459                  * addressing
3460                  */
3461                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3462                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3463                         goto out;
3464                 }
3465
3466                 rcu_read_lock();
3467
3468                 if (cfg->fc_flags & RTNH_F_ONLINK)
3469                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3470                 else
3471                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3472
3473                 rcu_read_unlock();
3474
3475                 if (err)
3476                         goto out;
3477         }
3478
3479         /* reload in case device was changed */
3480         dev = *_dev;
3481
3482         err = -EINVAL;
3483         if (!dev) {
3484                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3485                 goto out;
3486         } else if (dev->flags & IFF_LOOPBACK) {
3487                 NL_SET_ERR_MSG(extack,
3488                                "Egress device can not be loopback device for this route");
3489                 goto out;
3490         }
3491
3492         /* if we did not check gw_addr above, do so now that the
3493          * egress device has been resolved.
3494          */
3495         if (need_addr_check &&
3496             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3497                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3498                 goto out;
3499         }
3500
3501         err = 0;
3502 out:
3503         return err;
3504 }
3505
3506 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3507 {
3508         if ((flags & RTF_REJECT) ||
3509             (dev && (dev->flags & IFF_LOOPBACK) &&
3510              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3511              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3512                 return true;
3513
3514         return false;
3515 }
3516
3517 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3518                  struct fib6_config *cfg, gfp_t gfp_flags,
3519                  struct netlink_ext_ack *extack)
3520 {
3521         struct net_device *dev = NULL;
3522         struct inet6_dev *idev = NULL;
3523         int addr_type;
3524         int err;
3525
3526         fib6_nh->fib_nh_family = AF_INET6;
3527 #ifdef CONFIG_IPV6_ROUTER_PREF
3528         fib6_nh->last_probe = jiffies;
3529 #endif
3530         if (cfg->fc_is_fdb) {
3531                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3532                 fib6_nh->fib_nh_gw_family = AF_INET6;
3533                 return 0;
3534         }
3535
3536         err = -ENODEV;
3537         if (cfg->fc_ifindex) {
3538                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3539                 if (!dev)
3540                         goto out;
3541                 idev = in6_dev_get(dev);
3542                 if (!idev)
3543                         goto out;
3544         }
3545
3546         if (cfg->fc_flags & RTNH_F_ONLINK) {
3547                 if (!dev) {
3548                         NL_SET_ERR_MSG(extack,
3549                                        "Nexthop device required for onlink");
3550                         goto out;
3551                 }
3552
3553                 if (!(dev->flags & IFF_UP)) {
3554                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3555                         err = -ENETDOWN;
3556                         goto out;
3557                 }
3558
3559                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3560         }
3561
3562         fib6_nh->fib_nh_weight = 1;
3563
3564         /* We cannot add true routes via loopback here,
3565          * they would result in kernel looping; promote them to reject routes
3566          */
3567         addr_type = ipv6_addr_type(&cfg->fc_dst);
3568         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3569                 /* hold loopback dev/idev if we haven't done so. */
3570                 if (dev != net->loopback_dev) {
3571                         if (dev) {
3572                                 dev_put(dev);
3573                                 in6_dev_put(idev);
3574                         }
3575                         dev = net->loopback_dev;
3576                         dev_hold(dev);
3577                         idev = in6_dev_get(dev);
3578                         if (!idev) {
3579                                 err = -ENODEV;
3580                                 goto out;
3581                         }
3582                 }
3583                 goto pcpu_alloc;
3584         }
3585
3586         if (cfg->fc_flags & RTF_GATEWAY) {
3587                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3588                 if (err)
3589                         goto out;
3590
3591                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3592                 fib6_nh->fib_nh_gw_family = AF_INET6;
3593         }
3594
3595         err = -ENODEV;
3596         if (!dev)
3597                 goto out;
3598
3599         if (idev->cnf.disable_ipv6) {
3600                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3601                 err = -EACCES;
3602                 goto out;
3603         }
3604
3605         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3606                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3607                 err = -ENETDOWN;
3608                 goto out;
3609         }
3610
3611         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3612             !netif_carrier_ok(dev))
3613                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3614
3615         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3616                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3617         if (err)
3618                 goto out;
3619
3620 pcpu_alloc:
3621         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3622         if (!fib6_nh->rt6i_pcpu) {
3623                 err = -ENOMEM;
3624                 goto out;
3625         }
3626
3627         fib6_nh->fib_nh_dev = dev;
3628         fib6_nh->fib_nh_oif = dev->ifindex;
3629         err = 0;
3630 out:
3631         if (idev)
3632                 in6_dev_put(idev);
3633
3634         if (err) {
3635                 lwtstate_put(fib6_nh->fib_nh_lws);
3636                 fib6_nh->fib_nh_lws = NULL;
3637                 dev_put(dev);
3638         }
3639
3640         return err;
3641 }
3642
3643 void fib6_nh_release(struct fib6_nh *fib6_nh)
3644 {
3645         struct rt6_exception_bucket *bucket;
3646
3647         rcu_read_lock();
3648
3649         fib6_nh_flush_exceptions(fib6_nh, NULL);
3650         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3651         if (bucket) {
3652                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3653                 kfree(bucket);
3654         }
3655
3656         rcu_read_unlock();
3657
3658         if (fib6_nh->rt6i_pcpu) {
3659                 int cpu;
3660
3661                 for_each_possible_cpu(cpu) {
3662                         struct rt6_info **ppcpu_rt;
3663                         struct rt6_info *pcpu_rt;
3664
3665                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3666                         pcpu_rt = *ppcpu_rt;
3667                         if (pcpu_rt) {
3668                                 dst_dev_put(&pcpu_rt->dst);
3669                                 dst_release(&pcpu_rt->dst);
3670                                 *ppcpu_rt = NULL;
3671                         }
3672                 }
3673
3674                 free_percpu(fib6_nh->rt6i_pcpu);
3675         }
3676
3677         fib_nh_common_release(&fib6_nh->nh_common);
3678 }
3679
3680 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3681                                               gfp_t gfp_flags,
3682                                               struct netlink_ext_ack *extack)
3683 {
3684         struct net *net = cfg->fc_nlinfo.nl_net;
3685         struct fib6_info *rt = NULL;
3686         struct nexthop *nh = NULL;
3687         struct fib6_table *table;
3688         struct fib6_nh *fib6_nh;
3689         int err = -EINVAL;
3690         int addr_type;
3691
3692         /* RTF_PCPU is an internal flag; can not be set by userspace */
3693         if (cfg->fc_flags & RTF_PCPU) {
3694                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3695                 goto out;
3696         }
3697
3698         /* RTF_CACHE is an internal flag; can not be set by userspace */
3699         if (cfg->fc_flags & RTF_CACHE) {
3700                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3701                 goto out;
3702         }
3703
3704         if (cfg->fc_type > RTN_MAX) {
3705                 NL_SET_ERR_MSG(extack, "Invalid route type");
3706                 goto out;
3707         }
3708
3709         if (cfg->fc_dst_len > 128) {
3710                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3711                 goto out;
3712         }
3713         if (cfg->fc_src_len > 128) {
3714                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3715                 goto out;
3716         }
3717 #ifndef CONFIG_IPV6_SUBTREES
3718         if (cfg->fc_src_len) {
3719                 NL_SET_ERR_MSG(extack,
3720                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3721                 goto out;
3722         }
3723 #endif
3724         if (cfg->fc_nh_id) {
3725                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3726                 if (!nh) {
3727                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3728                         goto out;
3729                 }
3730                 err = fib6_check_nexthop(nh, cfg, extack);
3731                 if (err)
3732                         goto out;
3733         }
3734
3735         err = -ENOBUFS;
3736         if (cfg->fc_nlinfo.nlh &&
3737             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3738                 table = fib6_get_table(net, cfg->fc_table);
3739                 if (!table) {
3740                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3741                         table = fib6_new_table(net, cfg->fc_table);
3742                 }
3743         } else {
3744                 table = fib6_new_table(net, cfg->fc_table);
3745         }
3746
3747         if (!table)
3748                 goto out;
3749
3750         err = -ENOMEM;
3751         rt = fib6_info_alloc(gfp_flags, !nh);
3752         if (!rt)
3753                 goto out;
3754
3755         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3756                                                extack);
3757         if (IS_ERR(rt->fib6_metrics)) {
3758                 err = PTR_ERR(rt->fib6_metrics);
3759                 /* Do not leave garbage there. */
3760                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3761                 goto out_free;
3762         }
3763
3764         if (cfg->fc_flags & RTF_ADDRCONF)
3765                 rt->dst_nocount = true;
3766
3767         if (cfg->fc_flags & RTF_EXPIRES)
3768                 fib6_set_expires(rt, jiffies +
3769                                 clock_t_to_jiffies(cfg->fc_expires));
3770         else
3771                 fib6_clean_expires(rt);
3772
3773         if (cfg->fc_protocol == RTPROT_UNSPEC)
3774                 cfg->fc_protocol = RTPROT_BOOT;
3775         rt->fib6_protocol = cfg->fc_protocol;
3776
3777         rt->fib6_table = table;
3778         rt->fib6_metric = cfg->fc_metric;
3779         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3780         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3781
3782         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3783         rt->fib6_dst.plen = cfg->fc_dst_len;
3784
3785 #ifdef CONFIG_IPV6_SUBTREES
3786         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3787         rt->fib6_src.plen = cfg->fc_src_len;
3788 #endif
3789         if (nh) {
3790                 if (rt->fib6_src.plen) {
3791                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3792                         goto out_free;
3793                 }
3794                 if (!nexthop_get(nh)) {
3795                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3796                         goto out_free;
3797                 }
3798                 rt->nh = nh;
3799                 fib6_nh = nexthop_fib6_nh(rt->nh);
3800         } else {
3801                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3802                 if (err)
3803                         goto out;
3804
3805                 fib6_nh = rt->fib6_nh;
3806
3807                 /* We cannot add true routes via loopback here, they would
3808                  * result in kernel looping; promote them to reject routes
3809                  */
3810                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3811                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3812                                    addr_type))
3813                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3814         }
3815
3816         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3817                 struct net_device *dev = fib6_nh->fib_nh_dev;
3818
3819                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3820                         NL_SET_ERR_MSG(extack, "Invalid source address");
3821                         err = -EINVAL;
3822                         goto out;
3823                 }
3824                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3825                 rt->fib6_prefsrc.plen = 128;
3826         } else
3827                 rt->fib6_prefsrc.plen = 0;
3828
3829         return rt;
3830 out:
3831         fib6_info_release(rt);
3832         return ERR_PTR(err);
3833 out_free:
3834         ip_fib_metrics_put(rt->fib6_metrics);
3835         kfree(rt);
3836         return ERR_PTR(err);
3837 }
3838
3839 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3840                   struct netlink_ext_ack *extack)
3841 {
3842         struct fib6_info *rt;
3843         int err;
3844
3845         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3846         if (IS_ERR(rt))
3847                 return PTR_ERR(rt);
3848
3849         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3850         fib6_info_release(rt);
3851
3852         return err;
3853 }
3854
3855 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3856 {
3857         struct net *net = info->nl_net;
3858         struct fib6_table *table;
3859         int err;
3860
3861         if (rt == net->ipv6.fib6_null_entry) {
3862                 err = -ENOENT;
3863                 goto out;
3864         }
3865
3866         table = rt->fib6_table;
3867         spin_lock_bh(&table->tb6_lock);
3868         err = fib6_del(rt, info);
3869         spin_unlock_bh(&table->tb6_lock);
3870
3871 out:
3872         fib6_info_release(rt);
3873         return err;
3874 }
3875
3876 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3877 {
3878         struct nl_info info = {
3879                 .nl_net = net,
3880                 .skip_notify = skip_notify
3881         };
3882
3883         return __ip6_del_rt(rt, &info);
3884 }
3885
3886 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3887 {
3888         struct nl_info *info = &cfg->fc_nlinfo;
3889         struct net *net = info->nl_net;
3890         struct sk_buff *skb = NULL;
3891         struct fib6_table *table;
3892         int err = -ENOENT;
3893
3894         if (rt == net->ipv6.fib6_null_entry)
3895                 goto out_put;
3896         table = rt->fib6_table;
3897         spin_lock_bh(&table->tb6_lock);
3898
3899         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3900                 struct fib6_info *sibling, *next_sibling;
3901                 struct fib6_node *fn;
3902
3903                 /* prefer to send a single notification with all hops */
3904                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3905                 if (skb) {
3906                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3907
3908                         if (rt6_fill_node(net, skb, rt, NULL,
3909                                           NULL, NULL, 0, RTM_DELROUTE,
3910                                           info->portid, seq, 0) < 0) {
3911                                 kfree_skb(skb);
3912                                 skb = NULL;
3913                         } else
3914                                 info->skip_notify = 1;
3915                 }
3916
3917                 /* 'rt' points to the first sibling route. If it is not the
3918                  * leaf, then we do not need to send a notification. Otherwise,
3919                  * we need to check if the last sibling has a next route or not
3920                  * and emit a replace or delete notification, respectively.
3921                  */
3922                 info->skip_notify_kernel = 1;
3923                 fn = rcu_dereference_protected(rt->fib6_node,
3924                                             lockdep_is_held(&table->tb6_lock));
3925                 if (rcu_access_pointer(fn->leaf) == rt) {
3926                         struct fib6_info *last_sibling, *replace_rt;
3927
3928                         last_sibling = list_last_entry(&rt->fib6_siblings,
3929                                                        struct fib6_info,
3930                                                        fib6_siblings);
3931                         replace_rt = rcu_dereference_protected(
3932                                             last_sibling->fib6_next,
3933                                             lockdep_is_held(&table->tb6_lock));
3934                         if (replace_rt)
3935                                 call_fib6_entry_notifiers_replace(net,
3936                                                                   replace_rt);
3937                         else
3938                                 call_fib6_multipath_entry_notifiers(net,
3939                                                        FIB_EVENT_ENTRY_DEL,
3940                                                        rt, rt->fib6_nsiblings,
3941                                                        NULL);
3942                 }
3943                 list_for_each_entry_safe(sibling, next_sibling,
3944                                          &rt->fib6_siblings,
3945                                          fib6_siblings) {
3946                         err = fib6_del(sibling, info);
3947                         if (err)
3948                                 goto out_unlock;
3949                 }
3950         }
3951
3952         err = fib6_del(rt, info);
3953 out_unlock:
3954         spin_unlock_bh(&table->tb6_lock);
3955 out_put:
3956         fib6_info_release(rt);
3957
3958         if (skb) {
3959                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3960                             info->nlh, gfp_any());
3961         }
3962         return err;
3963 }
3964
3965 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3966 {
3967         int rc = -ESRCH;
3968
3969         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3970                 goto out;
3971
3972         if (cfg->fc_flags & RTF_GATEWAY &&
3973             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3974                 goto out;
3975
3976         rc = rt6_remove_exception_rt(rt);
3977 out:
3978         return rc;
3979 }
3980
3981 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3982                              struct fib6_nh *nh)
3983 {
3984         struct fib6_result res = {
3985                 .f6i = rt,
3986                 .nh = nh,
3987         };
3988         struct rt6_info *rt_cache;
3989
3990         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3991         if (rt_cache)
3992                 return __ip6_del_cached_rt(rt_cache, cfg);
3993
3994         return 0;
3995 }
3996
3997 struct fib6_nh_del_cached_rt_arg {
3998         struct fib6_config *cfg;
3999         struct fib6_info *f6i;
4000 };
4001
4002 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4003 {
4004         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4005         int rc;
4006
4007         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4008         return rc != -ESRCH ? rc : 0;
4009 }
4010
4011 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4012 {
4013         struct fib6_nh_del_cached_rt_arg arg = {
4014                 .cfg = cfg,
4015                 .f6i = f6i
4016         };
4017
4018         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4019 }
4020
4021 static int ip6_route_del(struct fib6_config *cfg,
4022                          struct netlink_ext_ack *extack)
4023 {
4024         struct fib6_table *table;
4025         struct fib6_info *rt;
4026         struct fib6_node *fn;
4027         int err = -ESRCH;
4028
4029         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4030         if (!table) {
4031                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4032                 return err;
4033         }
4034
4035         rcu_read_lock();
4036
4037         fn = fib6_locate(&table->tb6_root,
4038                          &cfg->fc_dst, cfg->fc_dst_len,
4039                          &cfg->fc_src, cfg->fc_src_len,
4040                          !(cfg->fc_flags & RTF_CACHE));
4041
4042         if (fn) {
4043                 for_each_fib6_node_rt_rcu(fn) {
4044                         struct fib6_nh *nh;
4045
4046                         if (rt->nh && cfg->fc_nh_id &&
4047                             rt->nh->id != cfg->fc_nh_id)
4048                                 continue;
4049
4050                         if (cfg->fc_flags & RTF_CACHE) {
4051                                 int rc = 0;
4052
4053                                 if (rt->nh) {
4054                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4055                                 } else if (cfg->fc_nh_id) {
4056                                         continue;
4057                                 } else {
4058                                         nh = rt->fib6_nh;
4059                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4060                                 }
4061                                 if (rc != -ESRCH) {
4062                                         rcu_read_unlock();
4063                                         return rc;
4064                                 }
4065                                 continue;
4066                         }
4067
4068                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4069                                 continue;
4070                         if (cfg->fc_protocol &&
4071                             cfg->fc_protocol != rt->fib6_protocol)
4072                                 continue;
4073
4074                         if (rt->nh) {
4075                                 if (!fib6_info_hold_safe(rt))
4076                                         continue;
4077                                 rcu_read_unlock();
4078
4079                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4080                         }
4081                         if (cfg->fc_nh_id)
4082                                 continue;
4083
4084                         nh = rt->fib6_nh;
4085                         if (cfg->fc_ifindex &&
4086                             (!nh->fib_nh_dev ||
4087                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4088                                 continue;
4089                         if (cfg->fc_flags & RTF_GATEWAY &&
4090                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4091                                 continue;
4092                         if (!fib6_info_hold_safe(rt))
4093                                 continue;
4094                         rcu_read_unlock();
4095
4096                         /* if gateway was specified only delete the one hop */
4097                         if (cfg->fc_flags & RTF_GATEWAY)
4098                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4099
4100                         return __ip6_del_rt_siblings(rt, cfg);
4101                 }
4102         }
4103         rcu_read_unlock();
4104
4105         return err;
4106 }
4107
4108 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4109 {
4110         struct netevent_redirect netevent;
4111         struct rt6_info *rt, *nrt = NULL;
4112         struct fib6_result res = {};
4113         struct ndisc_options ndopts;
4114         struct inet6_dev *in6_dev;
4115         struct neighbour *neigh;
4116         struct rd_msg *msg;
4117         int optlen, on_link;
4118         u8 *lladdr;
4119
4120         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4121         optlen -= sizeof(*msg);
4122
4123         if (optlen < 0) {
4124                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4125                 return;
4126         }
4127
4128         msg = (struct rd_msg *)icmp6_hdr(skb);
4129
4130         if (ipv6_addr_is_multicast(&msg->dest)) {
4131                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4132                 return;
4133         }
4134
4135         on_link = 0;
4136         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4137                 on_link = 1;
4138         } else if (ipv6_addr_type(&msg->target) !=
4139                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4140                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4141                 return;
4142         }
4143
4144         in6_dev = __in6_dev_get(skb->dev);
4145         if (!in6_dev)
4146                 return;
4147         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4148                 return;
4149
4150         /* RFC2461 8.1:
4151          *      The IP source address of the Redirect MUST be the same as the current
4152          *      first-hop router for the specified ICMP Destination Address.
4153          */
4154
4155         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4156                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4157                 return;
4158         }
4159
4160         lladdr = NULL;
4161         if (ndopts.nd_opts_tgt_lladdr) {
4162                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4163                                              skb->dev);
4164                 if (!lladdr) {
4165                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4166                         return;
4167                 }
4168         }
4169
4170         rt = (struct rt6_info *) dst;
4171         if (rt->rt6i_flags & RTF_REJECT) {
4172                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4173                 return;
4174         }
4175
4176         /* Redirect received -> path was valid.
4177          * Look, redirects are sent only in response to data packets,
4178          * so that this nexthop apparently is reachable. --ANK
4179          */
4180         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4181
4182         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4183         if (!neigh)
4184                 return;
4185
4186         /*
4187          *      We have finally decided to accept it.
4188          */
4189
4190         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4191                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4192                      NEIGH_UPDATE_F_OVERRIDE|
4193                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4194                                      NEIGH_UPDATE_F_ISROUTER)),
4195                      NDISC_REDIRECT, &ndopts);
4196
4197         rcu_read_lock();
4198         res.f6i = rcu_dereference(rt->from);
4199         if (!res.f6i)
4200                 goto out;
4201
4202         if (res.f6i->nh) {
4203                 struct fib6_nh_match_arg arg = {
4204                         .dev = dst->dev,
4205                         .gw = &rt->rt6i_gateway,
4206                 };
4207
4208                 nexthop_for_each_fib6_nh(res.f6i->nh,
4209                                          fib6_nh_find_match, &arg);
4210
4211                 /* fib6_info uses a nexthop that does not have fib6_nh
4212                  * using the dst->dev. Should be impossible
4213                  */
4214                 if (!arg.match)
4215                         goto out;
4216                 res.nh = arg.match;
4217         } else {
4218                 res.nh = res.f6i->fib6_nh;
4219         }
4220
4221         res.fib6_flags = res.f6i->fib6_flags;
4222         res.fib6_type = res.f6i->fib6_type;
4223         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4224         if (!nrt)
4225                 goto out;
4226
4227         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4228         if (on_link)
4229                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4230
4231         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4232
4233         /* rt6_insert_exception() will take care of duplicated exceptions */
4234         if (rt6_insert_exception(nrt, &res)) {
4235                 dst_release_immediate(&nrt->dst);
4236                 goto out;
4237         }
4238
4239         netevent.old = &rt->dst;
4240         netevent.new = &nrt->dst;
4241         netevent.daddr = &msg->dest;
4242         netevent.neigh = neigh;
4243         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4244
4245 out:
4246         rcu_read_unlock();
4247         neigh_release(neigh);
4248 }
4249
4250 #ifdef CONFIG_IPV6_ROUTE_INFO
4251 static struct fib6_info *rt6_get_route_info(struct net *net,
4252                                            const struct in6_addr *prefix, int prefixlen,
4253                                            const struct in6_addr *gwaddr,
4254                                            struct net_device *dev)
4255 {
4256         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4257         int ifindex = dev->ifindex;
4258         struct fib6_node *fn;
4259         struct fib6_info *rt = NULL;
4260         struct fib6_table *table;
4261
4262         table = fib6_get_table(net, tb_id);
4263         if (!table)
4264                 return NULL;
4265
4266         rcu_read_lock();
4267         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4268         if (!fn)
4269                 goto out;
4270
4271         for_each_fib6_node_rt_rcu(fn) {
4272                 /* these routes do not use nexthops */
4273                 if (rt->nh)
4274                         continue;
4275                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4276                         continue;
4277                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4278                     !rt->fib6_nh->fib_nh_gw_family)
4279                         continue;
4280                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4281                         continue;
4282                 if (!fib6_info_hold_safe(rt))
4283                         continue;
4284                 break;
4285         }
4286 out:
4287         rcu_read_unlock();
4288         return rt;
4289 }
4290
4291 static struct fib6_info *rt6_add_route_info(struct net *net,
4292                                            const struct in6_addr *prefix, int prefixlen,
4293                                            const struct in6_addr *gwaddr,
4294                                            struct net_device *dev,
4295                                            unsigned int pref)
4296 {
4297         struct fib6_config cfg = {
4298                 .fc_metric      = IP6_RT_PRIO_USER,
4299                 .fc_ifindex     = dev->ifindex,
4300                 .fc_dst_len     = prefixlen,
4301                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4302                                   RTF_UP | RTF_PREF(pref),
4303                 .fc_protocol = RTPROT_RA,
4304                 .fc_type = RTN_UNICAST,
4305                 .fc_nlinfo.portid = 0,
4306                 .fc_nlinfo.nlh = NULL,
4307                 .fc_nlinfo.nl_net = net,
4308         };
4309
4310         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4311         cfg.fc_dst = *prefix;
4312         cfg.fc_gateway = *gwaddr;
4313
4314         /* We should treat it as a default route if prefix length is 0. */
4315         if (!prefixlen)
4316                 cfg.fc_flags |= RTF_DEFAULT;
4317
4318         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4319
4320         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4321 }
4322 #endif
4323
4324 struct fib6_info *rt6_get_dflt_router(struct net *net,
4325                                      const struct in6_addr *addr,
4326                                      struct net_device *dev)
4327 {
4328         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4329         struct fib6_info *rt;
4330         struct fib6_table *table;
4331
4332         table = fib6_get_table(net, tb_id);
4333         if (!table)
4334                 return NULL;
4335
4336         rcu_read_lock();
4337         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4338                 struct fib6_nh *nh;
4339
4340                 /* RA routes do not use nexthops */
4341                 if (rt->nh)
4342                         continue;
4343
4344                 nh = rt->fib6_nh;
4345                 if (dev == nh->fib_nh_dev &&
4346                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4347                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4348                         break;
4349         }
4350         if (rt && !fib6_info_hold_safe(rt))
4351                 rt = NULL;
4352         rcu_read_unlock();
4353         return rt;
4354 }
4355
4356 struct fib6_info *rt6_add_dflt_router(struct net *net,
4357                                      const struct in6_addr *gwaddr,
4358                                      struct net_device *dev,
4359                                      unsigned int pref,
4360                                      u32 defrtr_usr_metric)
4361 {
4362         struct fib6_config cfg = {
4363                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4364                 .fc_metric      = defrtr_usr_metric,
4365                 .fc_ifindex     = dev->ifindex,
4366                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4367                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4368                 .fc_protocol = RTPROT_RA,
4369                 .fc_type = RTN_UNICAST,
4370                 .fc_nlinfo.portid = 0,
4371                 .fc_nlinfo.nlh = NULL,
4372                 .fc_nlinfo.nl_net = net,
4373         };
4374
4375         cfg.fc_gateway = *gwaddr;
4376
4377         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4378                 struct fib6_table *table;
4379
4380                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4381                 if (table)
4382                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4383         }
4384
4385         return rt6_get_dflt_router(net, gwaddr, dev);
4386 }
4387
4388 static void __rt6_purge_dflt_routers(struct net *net,
4389                                      struct fib6_table *table)
4390 {
4391         struct fib6_info *rt;
4392
4393 restart:
4394         rcu_read_lock();
4395         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4396                 struct net_device *dev = fib6_info_nh_dev(rt);
4397                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4398
4399                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4400                     (!idev || idev->cnf.accept_ra != 2) &&
4401                     fib6_info_hold_safe(rt)) {
4402                         rcu_read_unlock();
4403                         ip6_del_rt(net, rt, false);
4404                         goto restart;
4405                 }
4406         }
4407         rcu_read_unlock();
4408
4409         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4410 }
4411
4412 void rt6_purge_dflt_routers(struct net *net)
4413 {
4414         struct fib6_table *table;
4415         struct hlist_head *head;
4416         unsigned int h;
4417
4418         rcu_read_lock();
4419
4420         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4421                 head = &net->ipv6.fib_table_hash[h];
4422                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4423                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4424                                 __rt6_purge_dflt_routers(net, table);
4425                 }
4426         }
4427
4428         rcu_read_unlock();
4429 }
4430
4431 static void rtmsg_to_fib6_config(struct net *net,
4432                                  struct in6_rtmsg *rtmsg,
4433                                  struct fib6_config *cfg)
4434 {
4435         *cfg = (struct fib6_config){
4436                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4437                          : RT6_TABLE_MAIN,
4438                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4439                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4440                 .fc_expires = rtmsg->rtmsg_info,
4441                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4442                 .fc_src_len = rtmsg->rtmsg_src_len,
4443                 .fc_flags = rtmsg->rtmsg_flags,
4444                 .fc_type = rtmsg->rtmsg_type,
4445
4446                 .fc_nlinfo.nl_net = net,
4447
4448                 .fc_dst = rtmsg->rtmsg_dst,
4449                 .fc_src = rtmsg->rtmsg_src,
4450                 .fc_gateway = rtmsg->rtmsg_gateway,
4451         };
4452 }
4453
4454 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4455 {
4456         struct fib6_config cfg;
4457         int err;
4458
4459         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4460                 return -EINVAL;
4461         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4462                 return -EPERM;
4463
4464         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4465
4466         rtnl_lock();
4467         switch (cmd) {
4468         case SIOCADDRT:
4469                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4470                 break;
4471         case SIOCDELRT:
4472                 err = ip6_route_del(&cfg, NULL);
4473                 break;
4474         }
4475         rtnl_unlock();
4476         return err;
4477 }
4478
4479 /*
4480  *      Drop the packet on the floor
4481  */
4482
4483 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4484 {
4485         struct dst_entry *dst = skb_dst(skb);
4486         struct net *net = dev_net(dst->dev);
4487         struct inet6_dev *idev;
4488         int type;
4489
4490         if (netif_is_l3_master(skb->dev) &&
4491             dst->dev == net->loopback_dev)
4492                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4493         else
4494                 idev = ip6_dst_idev(dst);
4495
4496         switch (ipstats_mib_noroutes) {
4497         case IPSTATS_MIB_INNOROUTES:
4498                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4499                 if (type == IPV6_ADDR_ANY) {
4500                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4501                         break;
4502                 }
4503                 fallthrough;
4504         case IPSTATS_MIB_OUTNOROUTES:
4505                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4506                 break;
4507         }
4508
4509         /* Start over by dropping the dst for l3mdev case */
4510         if (netif_is_l3_master(skb->dev))
4511                 skb_dst_drop(skb);
4512
4513         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4514         kfree_skb(skb);
4515         return 0;
4516 }
4517
4518 static int ip6_pkt_discard(struct sk_buff *skb)
4519 {
4520         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4521 }
4522
4523 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4524 {
4525         skb->dev = skb_dst(skb)->dev;
4526         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4527 }
4528
4529 static int ip6_pkt_prohibit(struct sk_buff *skb)
4530 {
4531         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4532 }
4533
4534 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4535 {
4536         skb->dev = skb_dst(skb)->dev;
4537         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4538 }
4539
4540 /*
4541  *      Allocate a dst for local (unicast / anycast) address.
4542  */
4543
4544 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4545                                      struct inet6_dev *idev,
4546                                      const struct in6_addr *addr,
4547                                      bool anycast, gfp_t gfp_flags)
4548 {
4549         struct fib6_config cfg = {
4550                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4551                 .fc_ifindex = idev->dev->ifindex,
4552                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4553                 .fc_dst = *addr,
4554                 .fc_dst_len = 128,
4555                 .fc_protocol = RTPROT_KERNEL,
4556                 .fc_nlinfo.nl_net = net,
4557                 .fc_ignore_dev_down = true,
4558         };
4559         struct fib6_info *f6i;
4560
4561         if (anycast) {
4562                 cfg.fc_type = RTN_ANYCAST;
4563                 cfg.fc_flags |= RTF_ANYCAST;
4564         } else {
4565                 cfg.fc_type = RTN_LOCAL;
4566                 cfg.fc_flags |= RTF_LOCAL;
4567         }
4568
4569         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4570         if (!IS_ERR(f6i))
4571                 f6i->dst_nocount = true;
4572         return f6i;
4573 }
4574
4575 /* remove deleted ip from prefsrc entries */
4576 struct arg_dev_net_ip {
4577         struct net_device *dev;
4578         struct net *net;
4579         struct in6_addr *addr;
4580 };
4581
4582 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4583 {
4584         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4585         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4586         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4587
4588         if (!rt->nh &&
4589             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4590             rt != net->ipv6.fib6_null_entry &&
4591             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4592                 spin_lock_bh(&rt6_exception_lock);
4593                 /* remove prefsrc entry */
4594                 rt->fib6_prefsrc.plen = 0;
4595                 spin_unlock_bh(&rt6_exception_lock);
4596         }
4597         return 0;
4598 }
4599
4600 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4601 {
4602         struct net *net = dev_net(ifp->idev->dev);
4603         struct arg_dev_net_ip adni = {
4604                 .dev = ifp->idev->dev,
4605                 .net = net,
4606                 .addr = &ifp->addr,
4607         };
4608         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4609 }
4610
4611 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4612
4613 /* Remove routers and update dst entries when gateway turn into host. */
4614 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4615 {
4616         struct in6_addr *gateway = (struct in6_addr *)arg;
4617         struct fib6_nh *nh;
4618
4619         /* RA routes do not use nexthops */
4620         if (rt->nh)
4621                 return 0;
4622
4623         nh = rt->fib6_nh;
4624         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4625             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4626                 return -1;
4627
4628         /* Further clean up cached routes in exception table.
4629          * This is needed because cached route may have a different
4630          * gateway than its 'parent' in the case of an ip redirect.
4631          */
4632         fib6_nh_exceptions_clean_tohost(nh, gateway);
4633
4634         return 0;
4635 }
4636
4637 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4638 {
4639         fib6_clean_all(net, fib6_clean_tohost, gateway);
4640 }
4641
4642 struct arg_netdev_event {
4643         const struct net_device *dev;
4644         union {
4645                 unsigned char nh_flags;
4646                 unsigned long event;
4647         };
4648 };
4649
4650 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4651 {
4652         struct fib6_info *iter;
4653         struct fib6_node *fn;
4654
4655         fn = rcu_dereference_protected(rt->fib6_node,
4656                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4657         iter = rcu_dereference_protected(fn->leaf,
4658                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4659         while (iter) {
4660                 if (iter->fib6_metric == rt->fib6_metric &&
4661                     rt6_qualify_for_ecmp(iter))
4662                         return iter;
4663                 iter = rcu_dereference_protected(iter->fib6_next,
4664                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4665         }
4666
4667         return NULL;
4668 }
4669
4670 /* only called for fib entries with builtin fib6_nh */
4671 static bool rt6_is_dead(const struct fib6_info *rt)
4672 {
4673         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4674             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4675              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4676                 return true;
4677
4678         return false;
4679 }
4680
4681 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4682 {
4683         struct fib6_info *iter;
4684         int total = 0;
4685
4686         if (!rt6_is_dead(rt))
4687                 total += rt->fib6_nh->fib_nh_weight;
4688
4689         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4690                 if (!rt6_is_dead(iter))
4691                         total += iter->fib6_nh->fib_nh_weight;
4692         }
4693
4694         return total;
4695 }
4696
4697 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4698 {
4699         int upper_bound = -1;
4700
4701         if (!rt6_is_dead(rt)) {
4702                 *weight += rt->fib6_nh->fib_nh_weight;
4703                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4704                                                     total) - 1;
4705         }
4706         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4707 }
4708
4709 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4710 {
4711         struct fib6_info *iter;
4712         int weight = 0;
4713
4714         rt6_upper_bound_set(rt, &weight, total);
4715
4716         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4717                 rt6_upper_bound_set(iter, &weight, total);
4718 }
4719
4720 void rt6_multipath_rebalance(struct fib6_info *rt)
4721 {
4722         struct fib6_info *first;
4723         int total;
4724
4725         /* In case the entire multipath route was marked for flushing,
4726          * then there is no need to rebalance upon the removal of every
4727          * sibling route.
4728          */
4729         if (!rt->fib6_nsiblings || rt->should_flush)
4730                 return;
4731
4732         /* During lookup routes are evaluated in order, so we need to
4733          * make sure upper bounds are assigned from the first sibling
4734          * onwards.
4735          */
4736         first = rt6_multipath_first_sibling(rt);
4737         if (WARN_ON_ONCE(!first))
4738                 return;
4739
4740         total = rt6_multipath_total_weight(first);
4741         rt6_multipath_upper_bound_set(first, total);
4742 }
4743
4744 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4745 {
4746         const struct arg_netdev_event *arg = p_arg;
4747         struct net *net = dev_net(arg->dev);
4748
4749         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4750             rt->fib6_nh->fib_nh_dev == arg->dev) {
4751                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4752                 fib6_update_sernum_upto_root(net, rt);
4753                 rt6_multipath_rebalance(rt);
4754         }
4755
4756         return 0;
4757 }
4758
4759 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4760 {
4761         struct arg_netdev_event arg = {
4762                 .dev = dev,
4763                 {
4764                         .nh_flags = nh_flags,
4765                 },
4766         };
4767
4768         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4769                 arg.nh_flags |= RTNH_F_LINKDOWN;
4770
4771         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4772 }
4773
4774 /* only called for fib entries with inline fib6_nh */
4775 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4776                                    const struct net_device *dev)
4777 {
4778         struct fib6_info *iter;
4779
4780         if (rt->fib6_nh->fib_nh_dev == dev)
4781                 return true;
4782         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4783                 if (iter->fib6_nh->fib_nh_dev == dev)
4784                         return true;
4785
4786         return false;
4787 }
4788
4789 static void rt6_multipath_flush(struct fib6_info *rt)
4790 {
4791         struct fib6_info *iter;
4792
4793         rt->should_flush = 1;
4794         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4795                 iter->should_flush = 1;
4796 }
4797
4798 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4799                                              const struct net_device *down_dev)
4800 {
4801         struct fib6_info *iter;
4802         unsigned int dead = 0;
4803
4804         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4805             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4806                 dead++;
4807         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4808                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4809                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4810                         dead++;
4811
4812         return dead;
4813 }
4814
4815 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4816                                        const struct net_device *dev,
4817                                        unsigned char nh_flags)
4818 {
4819         struct fib6_info *iter;
4820
4821         if (rt->fib6_nh->fib_nh_dev == dev)
4822                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4823         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4824                 if (iter->fib6_nh->fib_nh_dev == dev)
4825                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4826 }
4827
4828 /* called with write lock held for table with rt */
4829 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4830 {
4831         const struct arg_netdev_event *arg = p_arg;
4832         const struct net_device *dev = arg->dev;
4833         struct net *net = dev_net(dev);
4834
4835         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4836                 return 0;
4837
4838         switch (arg->event) {
4839         case NETDEV_UNREGISTER:
4840                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4841         case NETDEV_DOWN:
4842                 if (rt->should_flush)
4843                         return -1;
4844                 if (!rt->fib6_nsiblings)
4845                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4846                 if (rt6_multipath_uses_dev(rt, dev)) {
4847                         unsigned int count;
4848
4849                         count = rt6_multipath_dead_count(rt, dev);
4850                         if (rt->fib6_nsiblings + 1 == count) {
4851                                 rt6_multipath_flush(rt);
4852                                 return -1;
4853                         }
4854                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4855                                                    RTNH_F_LINKDOWN);
4856                         fib6_update_sernum(net, rt);
4857                         rt6_multipath_rebalance(rt);
4858                 }
4859                 return -2;
4860         case NETDEV_CHANGE:
4861                 if (rt->fib6_nh->fib_nh_dev != dev ||
4862                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4863                         break;
4864                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4865                 rt6_multipath_rebalance(rt);
4866                 break;
4867         }
4868
4869         return 0;
4870 }
4871
4872 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4873 {
4874         struct arg_netdev_event arg = {
4875                 .dev = dev,
4876                 {
4877                         .event = event,
4878                 },
4879         };
4880         struct net *net = dev_net(dev);
4881
4882         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4883                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4884         else
4885                 fib6_clean_all(net, fib6_ifdown, &arg);
4886 }
4887
4888 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4889 {
4890         rt6_sync_down_dev(dev, event);
4891         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4892         neigh_ifdown(&nd_tbl, dev);
4893 }
4894
4895 struct rt6_mtu_change_arg {
4896         struct net_device *dev;
4897         unsigned int mtu;
4898         struct fib6_info *f6i;
4899 };
4900
4901 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4902 {
4903         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4904         struct fib6_info *f6i = arg->f6i;
4905
4906         /* For administrative MTU increase, there is no way to discover
4907          * IPv6 PMTU increase, so PMTU increase should be updated here.
4908          * Since RFC 1981 doesn't include administrative MTU increase
4909          * update PMTU increase is a MUST. (i.e. jumbo frame)
4910          */
4911         if (nh->fib_nh_dev == arg->dev) {
4912                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4913                 u32 mtu = f6i->fib6_pmtu;
4914
4915                 if (mtu >= arg->mtu ||
4916                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4917                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4918
4919                 spin_lock_bh(&rt6_exception_lock);
4920                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4921                 spin_unlock_bh(&rt6_exception_lock);
4922         }
4923
4924         return 0;
4925 }
4926
4927 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4928 {
4929         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4930         struct inet6_dev *idev;
4931
4932         /* In IPv6 pmtu discovery is not optional,
4933            so that RTAX_MTU lock cannot disable it.
4934            We still use this lock to block changes
4935            caused by addrconf/ndisc.
4936         */
4937
4938         idev = __in6_dev_get(arg->dev);
4939         if (!idev)
4940                 return 0;
4941
4942         if (fib6_metric_locked(f6i, RTAX_MTU))
4943                 return 0;
4944
4945         arg->f6i = f6i;
4946         if (f6i->nh) {
4947                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4948                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4949                                                 arg);
4950         }
4951
4952         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4953 }
4954
4955 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4956 {
4957         struct rt6_mtu_change_arg arg = {
4958                 .dev = dev,
4959                 .mtu = mtu,
4960         };
4961
4962         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4963 }
4964
4965 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4966         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4967         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4968         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4969         [RTA_OIF]               = { .type = NLA_U32 },
4970         [RTA_IIF]               = { .type = NLA_U32 },
4971         [RTA_PRIORITY]          = { .type = NLA_U32 },
4972         [RTA_METRICS]           = { .type = NLA_NESTED },
4973         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4974         [RTA_PREF]              = { .type = NLA_U8 },
4975         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4976         [RTA_ENCAP]             = { .type = NLA_NESTED },
4977         [RTA_EXPIRES]           = { .type = NLA_U32 },
4978         [RTA_UID]               = { .type = NLA_U32 },
4979         [RTA_MARK]              = { .type = NLA_U32 },
4980         [RTA_TABLE]             = { .type = NLA_U32 },
4981         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4982         [RTA_SPORT]             = { .type = NLA_U16 },
4983         [RTA_DPORT]             = { .type = NLA_U16 },
4984         [RTA_NH_ID]             = { .type = NLA_U32 },
4985 };
4986
4987 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4988                               struct fib6_config *cfg,
4989                               struct netlink_ext_ack *extack)
4990 {
4991         struct rtmsg *rtm;
4992         struct nlattr *tb[RTA_MAX+1];
4993         unsigned int pref;
4994         int err;
4995
4996         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4997                                      rtm_ipv6_policy, extack);
4998         if (err < 0)
4999                 goto errout;
5000
5001         err = -EINVAL;
5002         rtm = nlmsg_data(nlh);
5003
5004         *cfg = (struct fib6_config){
5005                 .fc_table = rtm->rtm_table,
5006                 .fc_dst_len = rtm->rtm_dst_len,
5007                 .fc_src_len = rtm->rtm_src_len,
5008                 .fc_flags = RTF_UP,
5009                 .fc_protocol = rtm->rtm_protocol,
5010                 .fc_type = rtm->rtm_type,
5011
5012                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5013                 .fc_nlinfo.nlh = nlh,
5014                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5015         };
5016
5017         if (rtm->rtm_type == RTN_UNREACHABLE ||
5018             rtm->rtm_type == RTN_BLACKHOLE ||
5019             rtm->rtm_type == RTN_PROHIBIT ||
5020             rtm->rtm_type == RTN_THROW)
5021                 cfg->fc_flags |= RTF_REJECT;
5022
5023         if (rtm->rtm_type == RTN_LOCAL)
5024                 cfg->fc_flags |= RTF_LOCAL;
5025
5026         if (rtm->rtm_flags & RTM_F_CLONED)
5027                 cfg->fc_flags |= RTF_CACHE;
5028
5029         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5030
5031         if (tb[RTA_NH_ID]) {
5032                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5033                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5034                         NL_SET_ERR_MSG(extack,
5035                                        "Nexthop specification and nexthop id are mutually exclusive");
5036                         goto errout;
5037                 }
5038                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5039         }
5040
5041         if (tb[RTA_GATEWAY]) {
5042                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5043                 cfg->fc_flags |= RTF_GATEWAY;
5044         }
5045         if (tb[RTA_VIA]) {
5046                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5047                 goto errout;
5048         }
5049
5050         if (tb[RTA_DST]) {
5051                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5052
5053                 if (nla_len(tb[RTA_DST]) < plen)
5054                         goto errout;
5055
5056                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5057         }
5058
5059         if (tb[RTA_SRC]) {
5060                 int plen = (rtm->rtm_src_len + 7) >> 3;
5061
5062                 if (nla_len(tb[RTA_SRC]) < plen)
5063                         goto errout;
5064
5065                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5066         }
5067
5068         if (tb[RTA_PREFSRC])
5069                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5070
5071         if (tb[RTA_OIF])
5072                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5073
5074         if (tb[RTA_PRIORITY])
5075                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5076
5077         if (tb[RTA_METRICS]) {
5078                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5079                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5080         }
5081
5082         if (tb[RTA_TABLE])
5083                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5084
5085         if (tb[RTA_MULTIPATH]) {
5086                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5087                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5088
5089                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5090                                                      cfg->fc_mp_len, extack);
5091                 if (err < 0)
5092                         goto errout;
5093         }
5094
5095         if (tb[RTA_PREF]) {
5096                 pref = nla_get_u8(tb[RTA_PREF]);
5097                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5098                     pref != ICMPV6_ROUTER_PREF_HIGH)
5099                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5100                 cfg->fc_flags |= RTF_PREF(pref);
5101         }
5102
5103         if (tb[RTA_ENCAP])
5104                 cfg->fc_encap = tb[RTA_ENCAP];
5105
5106         if (tb[RTA_ENCAP_TYPE]) {
5107                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5108
5109                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5110                 if (err < 0)
5111                         goto errout;
5112         }
5113
5114         if (tb[RTA_EXPIRES]) {
5115                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5116
5117                 if (addrconf_finite_timeout(timeout)) {
5118                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5119                         cfg->fc_flags |= RTF_EXPIRES;
5120                 }
5121         }
5122
5123         err = 0;
5124 errout:
5125         return err;
5126 }
5127
5128 struct rt6_nh {
5129         struct fib6_info *fib6_info;
5130         struct fib6_config r_cfg;
5131         struct list_head next;
5132 };
5133
5134 static int ip6_route_info_append(struct net *net,
5135                                  struct list_head *rt6_nh_list,
5136                                  struct fib6_info *rt,
5137                                  struct fib6_config *r_cfg)
5138 {
5139         struct rt6_nh *nh;
5140         int err = -EEXIST;
5141
5142         list_for_each_entry(nh, rt6_nh_list, next) {
5143                 /* check if fib6_info already exists */
5144                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5145                         return err;
5146         }
5147
5148         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5149         if (!nh)
5150                 return -ENOMEM;
5151         nh->fib6_info = rt;
5152         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5153         list_add_tail(&nh->next, rt6_nh_list);
5154
5155         return 0;
5156 }
5157
5158 static void ip6_route_mpath_notify(struct fib6_info *rt,
5159                                    struct fib6_info *rt_last,
5160                                    struct nl_info *info,
5161                                    __u16 nlflags)
5162 {
5163         /* if this is an APPEND route, then rt points to the first route
5164          * inserted and rt_last points to last route inserted. Userspace
5165          * wants a consistent dump of the route which starts at the first
5166          * nexthop. Since sibling routes are always added at the end of
5167          * the list, find the first sibling of the last route appended
5168          */
5169         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5170                 rt = list_first_entry(&rt_last->fib6_siblings,
5171                                       struct fib6_info,
5172                                       fib6_siblings);
5173         }
5174
5175         if (rt)
5176                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5177 }
5178
5179 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5180 {
5181         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5182         bool should_notify = false;
5183         struct fib6_info *leaf;
5184         struct fib6_node *fn;
5185
5186         rcu_read_lock();
5187         fn = rcu_dereference(rt->fib6_node);
5188         if (!fn)
5189                 goto out;
5190
5191         leaf = rcu_dereference(fn->leaf);
5192         if (!leaf)
5193                 goto out;
5194
5195         if (rt == leaf ||
5196             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5197              rt6_qualify_for_ecmp(leaf)))
5198                 should_notify = true;
5199 out:
5200         rcu_read_unlock();
5201
5202         return should_notify;
5203 }
5204
5205 static int ip6_route_multipath_add(struct fib6_config *cfg,
5206                                    struct netlink_ext_ack *extack)
5207 {
5208         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5209         struct nl_info *info = &cfg->fc_nlinfo;
5210         struct fib6_config r_cfg;
5211         struct rtnexthop *rtnh;
5212         struct fib6_info *rt;
5213         struct rt6_nh *err_nh;
5214         struct rt6_nh *nh, *nh_safe;
5215         __u16 nlflags;
5216         int remaining;
5217         int attrlen;
5218         int err = 1;
5219         int nhn = 0;
5220         int replace = (cfg->fc_nlinfo.nlh &&
5221                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5222         LIST_HEAD(rt6_nh_list);
5223
5224         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5225         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5226                 nlflags |= NLM_F_APPEND;
5227
5228         remaining = cfg->fc_mp_len;
5229         rtnh = (struct rtnexthop *)cfg->fc_mp;
5230
5231         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5232          * fib6_info structs per nexthop
5233          */
5234         while (rtnh_ok(rtnh, remaining)) {
5235                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5236                 if (rtnh->rtnh_ifindex)
5237                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5238
5239                 attrlen = rtnh_attrlen(rtnh);
5240                 if (attrlen > 0) {
5241                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5242
5243                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5244                         if (nla) {
5245                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5246                                 r_cfg.fc_flags |= RTF_GATEWAY;
5247                         }
5248                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5249                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5250                         if (nla)
5251                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5252                 }
5253
5254                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5255                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5256                 if (IS_ERR(rt)) {
5257                         err = PTR_ERR(rt);
5258                         rt = NULL;
5259                         goto cleanup;
5260                 }
5261                 if (!rt6_qualify_for_ecmp(rt)) {
5262                         err = -EINVAL;
5263                         NL_SET_ERR_MSG(extack,
5264                                        "Device only routes can not be added for IPv6 using the multipath API.");
5265                         fib6_info_release(rt);
5266                         goto cleanup;
5267                 }
5268
5269                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5270
5271                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5272                                             rt, &r_cfg);
5273                 if (err) {
5274                         fib6_info_release(rt);
5275                         goto cleanup;
5276                 }
5277
5278                 rtnh = rtnh_next(rtnh, &remaining);
5279         }
5280
5281         if (list_empty(&rt6_nh_list)) {
5282                 NL_SET_ERR_MSG(extack,
5283                                "Invalid nexthop configuration - no valid nexthops");
5284                 return -EINVAL;
5285         }
5286
5287         /* for add and replace send one notification with all nexthops.
5288          * Skip the notification in fib6_add_rt2node and send one with
5289          * the full route when done
5290          */
5291         info->skip_notify = 1;
5292
5293         /* For add and replace, send one notification with all nexthops. For
5294          * append, send one notification with all appended nexthops.
5295          */
5296         info->skip_notify_kernel = 1;
5297
5298         err_nh = NULL;
5299         list_for_each_entry(nh, &rt6_nh_list, next) {
5300                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5301                 fib6_info_release(nh->fib6_info);
5302
5303                 if (!err) {
5304                         /* save reference to last route successfully inserted */
5305                         rt_last = nh->fib6_info;
5306
5307                         /* save reference to first route for notification */
5308                         if (!rt_notif)
5309                                 rt_notif = nh->fib6_info;
5310                 }
5311
5312                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5313                 nh->fib6_info = NULL;
5314                 if (err) {
5315                         if (replace && nhn)
5316                                 NL_SET_ERR_MSG_MOD(extack,
5317                                                    "multipath route replace failed (check consistency of installed routes)");
5318                         err_nh = nh;
5319                         goto add_errout;
5320                 }
5321
5322                 /* Because each route is added like a single route we remove
5323                  * these flags after the first nexthop: if there is a collision,
5324                  * we have already failed to add the first nexthop:
5325                  * fib6_add_rt2node() has rejected it; when replacing, old
5326                  * nexthops have been replaced by first new, the rest should
5327                  * be added to it.
5328                  */
5329                 if (cfg->fc_nlinfo.nlh) {
5330                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5331                                                              NLM_F_REPLACE);
5332                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5333                 }
5334                 nhn++;
5335         }
5336
5337         /* An in-kernel notification should only be sent in case the new
5338          * multipath route is added as the first route in the node, or if
5339          * it was appended to it. We pass 'rt_notif' since it is the first
5340          * sibling and might allow us to skip some checks in the replace case.
5341          */
5342         if (ip6_route_mpath_should_notify(rt_notif)) {
5343                 enum fib_event_type fib_event;
5344
5345                 if (rt_notif->fib6_nsiblings != nhn - 1)
5346                         fib_event = FIB_EVENT_ENTRY_APPEND;
5347                 else
5348                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5349
5350                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5351                                                           fib_event, rt_notif,
5352                                                           nhn - 1, extack);
5353                 if (err) {
5354                         /* Delete all the siblings that were just added */
5355                         err_nh = NULL;
5356                         goto add_errout;
5357                 }
5358         }
5359
5360         /* success ... tell user about new route */
5361         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5362         goto cleanup;
5363
5364 add_errout:
5365         /* send notification for routes that were added so that
5366          * the delete notifications sent by ip6_route_del are
5367          * coherent
5368          */
5369         if (rt_notif)
5370                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5371
5372         /* Delete routes that were already added */
5373         list_for_each_entry(nh, &rt6_nh_list, next) {
5374                 if (err_nh == nh)
5375                         break;
5376                 ip6_route_del(&nh->r_cfg, extack);
5377         }
5378
5379 cleanup:
5380         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5381                 if (nh->fib6_info)
5382                         fib6_info_release(nh->fib6_info);
5383                 list_del(&nh->next);
5384                 kfree(nh);
5385         }
5386
5387         return err;
5388 }
5389
5390 static int ip6_route_multipath_del(struct fib6_config *cfg,
5391                                    struct netlink_ext_ack *extack)
5392 {
5393         struct fib6_config r_cfg;
5394         struct rtnexthop *rtnh;
5395         int last_err = 0;
5396         int remaining;
5397         int attrlen;
5398         int err;
5399
5400         remaining = cfg->fc_mp_len;
5401         rtnh = (struct rtnexthop *)cfg->fc_mp;
5402
5403         /* Parse a Multipath Entry */
5404         while (rtnh_ok(rtnh, remaining)) {
5405                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5406                 if (rtnh->rtnh_ifindex)
5407                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5408
5409                 attrlen = rtnh_attrlen(rtnh);
5410                 if (attrlen > 0) {
5411                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5412
5413                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5414                         if (nla) {
5415                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5416                                 r_cfg.fc_flags |= RTF_GATEWAY;
5417                         }
5418                 }
5419                 err = ip6_route_del(&r_cfg, extack);
5420                 if (err)
5421                         last_err = err;
5422
5423                 rtnh = rtnh_next(rtnh, &remaining);
5424         }
5425
5426         return last_err;
5427 }
5428
5429 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5430                               struct netlink_ext_ack *extack)
5431 {
5432         struct fib6_config cfg;
5433         int err;
5434
5435         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5436         if (err < 0)
5437                 return err;
5438
5439         if (cfg.fc_nh_id &&
5440             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5441                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5442                 return -EINVAL;
5443         }
5444
5445         if (cfg.fc_mp)
5446                 return ip6_route_multipath_del(&cfg, extack);
5447         else {
5448                 cfg.fc_delete_all_nh = 1;
5449                 return ip6_route_del(&cfg, extack);
5450         }
5451 }
5452
5453 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5454                               struct netlink_ext_ack *extack)
5455 {
5456         struct fib6_config cfg;
5457         int err;
5458
5459         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5460         if (err < 0)
5461                 return err;
5462
5463         if (cfg.fc_metric == 0)
5464                 cfg.fc_metric = IP6_RT_PRIO_USER;
5465
5466         if (cfg.fc_mp)
5467                 return ip6_route_multipath_add(&cfg, extack);
5468         else
5469                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5470 }
5471
5472 /* add the overhead of this fib6_nh to nexthop_len */
5473 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5474 {
5475         int *nexthop_len = arg;
5476
5477         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5478                      + NLA_ALIGN(sizeof(struct rtnexthop))
5479                      + nla_total_size(16); /* RTA_GATEWAY */
5480
5481         if (nh->fib_nh_lws) {
5482                 /* RTA_ENCAP_TYPE */
5483                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5484                 /* RTA_ENCAP */
5485                 *nexthop_len += nla_total_size(2);
5486         }
5487
5488         return 0;
5489 }
5490
5491 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5492 {
5493         int nexthop_len;
5494
5495         if (f6i->nh) {
5496                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5497                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5498                                          &nexthop_len);
5499         } else {
5500                 struct fib6_nh *nh = f6i->fib6_nh;
5501
5502                 nexthop_len = 0;
5503                 if (f6i->fib6_nsiblings) {
5504                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5505                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5506                                     + nla_total_size(16) /* RTA_GATEWAY */
5507                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5508
5509                         nexthop_len *= f6i->fib6_nsiblings;
5510                 }
5511                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5512         }
5513
5514         return NLMSG_ALIGN(sizeof(struct rtmsg))
5515                + nla_total_size(16) /* RTA_SRC */
5516                + nla_total_size(16) /* RTA_DST */
5517                + nla_total_size(16) /* RTA_GATEWAY */
5518                + nla_total_size(16) /* RTA_PREFSRC */
5519                + nla_total_size(4) /* RTA_TABLE */
5520                + nla_total_size(4) /* RTA_IIF */
5521                + nla_total_size(4) /* RTA_OIF */
5522                + nla_total_size(4) /* RTA_PRIORITY */
5523                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5524                + nla_total_size(sizeof(struct rta_cacheinfo))
5525                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5526                + nla_total_size(1) /* RTA_PREF */
5527                + nexthop_len;
5528 }
5529
5530 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5531                                  unsigned char *flags)
5532 {
5533         if (nexthop_is_multipath(nh)) {
5534                 struct nlattr *mp;
5535
5536                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5537                 if (!mp)
5538                         goto nla_put_failure;
5539
5540                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5541                         goto nla_put_failure;
5542
5543                 nla_nest_end(skb, mp);
5544         } else {
5545                 struct fib6_nh *fib6_nh;
5546
5547                 fib6_nh = nexthop_fib6_nh(nh);
5548                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5549                                      flags, false) < 0)
5550                         goto nla_put_failure;
5551         }
5552
5553         return 0;
5554
5555 nla_put_failure:
5556         return -EMSGSIZE;
5557 }
5558
5559 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5560                          struct fib6_info *rt, struct dst_entry *dst,
5561                          struct in6_addr *dest, struct in6_addr *src,
5562                          int iif, int type, u32 portid, u32 seq,
5563                          unsigned int flags)
5564 {
5565         struct rt6_info *rt6 = (struct rt6_info *)dst;
5566         struct rt6key *rt6_dst, *rt6_src;
5567         u32 *pmetrics, table, rt6_flags;
5568         unsigned char nh_flags = 0;
5569         struct nlmsghdr *nlh;
5570         struct rtmsg *rtm;
5571         long expires = 0;
5572
5573         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5574         if (!nlh)
5575                 return -EMSGSIZE;
5576
5577         if (rt6) {
5578                 rt6_dst = &rt6->rt6i_dst;
5579                 rt6_src = &rt6->rt6i_src;
5580                 rt6_flags = rt6->rt6i_flags;
5581         } else {
5582                 rt6_dst = &rt->fib6_dst;
5583                 rt6_src = &rt->fib6_src;
5584                 rt6_flags = rt->fib6_flags;
5585         }
5586
5587         rtm = nlmsg_data(nlh);
5588         rtm->rtm_family = AF_INET6;
5589         rtm->rtm_dst_len = rt6_dst->plen;
5590         rtm->rtm_src_len = rt6_src->plen;
5591         rtm->rtm_tos = 0;
5592         if (rt->fib6_table)
5593                 table = rt->fib6_table->tb6_id;
5594         else
5595                 table = RT6_TABLE_UNSPEC;
5596         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5597         if (nla_put_u32(skb, RTA_TABLE, table))
5598                 goto nla_put_failure;
5599
5600         rtm->rtm_type = rt->fib6_type;
5601         rtm->rtm_flags = 0;
5602         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5603         rtm->rtm_protocol = rt->fib6_protocol;
5604
5605         if (rt6_flags & RTF_CACHE)
5606                 rtm->rtm_flags |= RTM_F_CLONED;
5607
5608         if (dest) {
5609                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5610                         goto nla_put_failure;
5611                 rtm->rtm_dst_len = 128;
5612         } else if (rtm->rtm_dst_len)
5613                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5614                         goto nla_put_failure;
5615 #ifdef CONFIG_IPV6_SUBTREES
5616         if (src) {
5617                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5618                         goto nla_put_failure;
5619                 rtm->rtm_src_len = 128;
5620         } else if (rtm->rtm_src_len &&
5621                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5622                 goto nla_put_failure;
5623 #endif
5624         if (iif) {
5625 #ifdef CONFIG_IPV6_MROUTE
5626                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5627                         int err = ip6mr_get_route(net, skb, rtm, portid);
5628
5629                         if (err == 0)
5630                                 return 0;
5631                         if (err < 0)
5632                                 goto nla_put_failure;
5633                 } else
5634 #endif
5635                         if (nla_put_u32(skb, RTA_IIF, iif))
5636                                 goto nla_put_failure;
5637         } else if (dest) {
5638                 struct in6_addr saddr_buf;
5639                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5640                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5641                         goto nla_put_failure;
5642         }
5643
5644         if (rt->fib6_prefsrc.plen) {
5645                 struct in6_addr saddr_buf;
5646                 saddr_buf = rt->fib6_prefsrc.addr;
5647                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5648                         goto nla_put_failure;
5649         }
5650
5651         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5652         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5653                 goto nla_put_failure;
5654
5655         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5656                 goto nla_put_failure;
5657
5658         /* For multipath routes, walk the siblings list and add
5659          * each as a nexthop within RTA_MULTIPATH.
5660          */
5661         if (rt6) {
5662                 if (rt6_flags & RTF_GATEWAY &&
5663                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5664                         goto nla_put_failure;
5665
5666                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5667                         goto nla_put_failure;
5668
5669                 if (dst->lwtstate &&
5670                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5671                         goto nla_put_failure;
5672         } else if (rt->fib6_nsiblings) {
5673                 struct fib6_info *sibling, *next_sibling;
5674                 struct nlattr *mp;
5675
5676                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5677                 if (!mp)
5678                         goto nla_put_failure;
5679
5680                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5681                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5682                         goto nla_put_failure;
5683
5684                 list_for_each_entry_safe(sibling, next_sibling,
5685                                          &rt->fib6_siblings, fib6_siblings) {
5686                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5687                                             sibling->fib6_nh->fib_nh_weight,
5688                                             AF_INET6) < 0)
5689                                 goto nla_put_failure;
5690                 }
5691
5692                 nla_nest_end(skb, mp);
5693         } else if (rt->nh) {
5694                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5695                         goto nla_put_failure;
5696
5697                 if (nexthop_is_blackhole(rt->nh))
5698                         rtm->rtm_type = RTN_BLACKHOLE;
5699
5700                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5701                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5702                         goto nla_put_failure;
5703
5704                 rtm->rtm_flags |= nh_flags;
5705         } else {
5706                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5707                                      &nh_flags, false) < 0)
5708                         goto nla_put_failure;
5709
5710                 rtm->rtm_flags |= nh_flags;
5711         }
5712
5713         if (rt6_flags & RTF_EXPIRES) {
5714                 expires = dst ? dst->expires : rt->expires;
5715                 expires -= jiffies;
5716         }
5717
5718         if (!dst) {
5719                 if (rt->offload)
5720                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5721                 if (rt->trap)
5722                         rtm->rtm_flags |= RTM_F_TRAP;
5723                 if (rt->offload_failed)
5724                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5725         }
5726
5727         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5728                 goto nla_put_failure;
5729
5730         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5731                 goto nla_put_failure;
5732
5733
5734         nlmsg_end(skb, nlh);
5735         return 0;
5736
5737 nla_put_failure:
5738         nlmsg_cancel(skb, nlh);
5739         return -EMSGSIZE;
5740 }
5741
5742 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5743 {
5744         const struct net_device *dev = arg;
5745
5746         if (nh->fib_nh_dev == dev)
5747                 return 1;
5748
5749         return 0;
5750 }
5751
5752 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5753                                const struct net_device *dev)
5754 {
5755         if (f6i->nh) {
5756                 struct net_device *_dev = (struct net_device *)dev;
5757
5758                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5759                                                   fib6_info_nh_uses_dev,
5760                                                   _dev);
5761         }
5762
5763         if (f6i->fib6_nh->fib_nh_dev == dev)
5764                 return true;
5765
5766         if (f6i->fib6_nsiblings) {
5767                 struct fib6_info *sibling, *next_sibling;
5768
5769                 list_for_each_entry_safe(sibling, next_sibling,
5770                                          &f6i->fib6_siblings, fib6_siblings) {
5771                         if (sibling->fib6_nh->fib_nh_dev == dev)
5772                                 return true;
5773                 }
5774         }
5775
5776         return false;
5777 }
5778
5779 struct fib6_nh_exception_dump_walker {
5780         struct rt6_rtnl_dump_arg *dump;
5781         struct fib6_info *rt;
5782         unsigned int flags;
5783         unsigned int skip;
5784         unsigned int count;
5785 };
5786
5787 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5788 {
5789         struct fib6_nh_exception_dump_walker *w = arg;
5790         struct rt6_rtnl_dump_arg *dump = w->dump;
5791         struct rt6_exception_bucket *bucket;
5792         struct rt6_exception *rt6_ex;
5793         int i, err;
5794
5795         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5796         if (!bucket)
5797                 return 0;
5798
5799         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5800                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5801                         if (w->skip) {
5802                                 w->skip--;
5803                                 continue;
5804                         }
5805
5806                         /* Expiration of entries doesn't bump sernum, insertion
5807                          * does. Removal is triggered by insertion, so we can
5808                          * rely on the fact that if entries change between two
5809                          * partial dumps, this node is scanned again completely,
5810                          * see rt6_insert_exception() and fib6_dump_table().
5811                          *
5812                          * Count expired entries we go through as handled
5813                          * entries that we'll skip next time, in case of partial
5814                          * node dump. Otherwise, if entries expire meanwhile,
5815                          * we'll skip the wrong amount.
5816                          */
5817                         if (rt6_check_expired(rt6_ex->rt6i)) {
5818                                 w->count++;
5819                                 continue;
5820                         }
5821
5822                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5823                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5824                                             RTM_NEWROUTE,
5825                                             NETLINK_CB(dump->cb->skb).portid,
5826                                             dump->cb->nlh->nlmsg_seq, w->flags);
5827                         if (err)
5828                                 return err;
5829
5830                         w->count++;
5831                 }
5832                 bucket++;
5833         }
5834
5835         return 0;
5836 }
5837
5838 /* Return -1 if done with node, number of handled routes on partial dump */
5839 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5840 {
5841         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5842         struct fib_dump_filter *filter = &arg->filter;
5843         unsigned int flags = NLM_F_MULTI;
5844         struct net *net = arg->net;
5845         int count = 0;
5846
5847         if (rt == net->ipv6.fib6_null_entry)
5848                 return -1;
5849
5850         if ((filter->flags & RTM_F_PREFIX) &&
5851             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5852                 /* success since this is not a prefix route */
5853                 return -1;
5854         }
5855         if (filter->filter_set &&
5856             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5857              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5858              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5859                 return -1;
5860         }
5861
5862         if (filter->filter_set ||
5863             !filter->dump_routes || !filter->dump_exceptions) {
5864                 flags |= NLM_F_DUMP_FILTERED;
5865         }
5866
5867         if (filter->dump_routes) {
5868                 if (skip) {
5869                         skip--;
5870                 } else {
5871                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5872                                           0, RTM_NEWROUTE,
5873                                           NETLINK_CB(arg->cb->skb).portid,
5874                                           arg->cb->nlh->nlmsg_seq, flags)) {
5875                                 return 0;
5876                         }
5877                         count++;
5878                 }
5879         }
5880
5881         if (filter->dump_exceptions) {
5882                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5883                                                            .rt = rt,
5884                                                            .flags = flags,
5885                                                            .skip = skip,
5886                                                            .count = 0 };
5887                 int err;
5888
5889                 rcu_read_lock();
5890                 if (rt->nh) {
5891                         err = nexthop_for_each_fib6_nh(rt->nh,
5892                                                        rt6_nh_dump_exceptions,
5893                                                        &w);
5894                 } else {
5895                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5896                 }
5897                 rcu_read_unlock();
5898
5899                 if (err)
5900                         return count += w.count;
5901         }
5902
5903         return -1;
5904 }
5905
5906 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5907                                         const struct nlmsghdr *nlh,
5908                                         struct nlattr **tb,
5909                                         struct netlink_ext_ack *extack)
5910 {
5911         struct rtmsg *rtm;
5912         int i, err;
5913
5914         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5915                 NL_SET_ERR_MSG_MOD(extack,
5916                                    "Invalid header for get route request");
5917                 return -EINVAL;
5918         }
5919
5920         if (!netlink_strict_get_check(skb))
5921                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5922                                               rtm_ipv6_policy, extack);
5923
5924         rtm = nlmsg_data(nlh);
5925         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5926             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5927             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5928             rtm->rtm_type) {
5929                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5930                 return -EINVAL;
5931         }
5932         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5933                 NL_SET_ERR_MSG_MOD(extack,
5934                                    "Invalid flags for get route request");
5935                 return -EINVAL;
5936         }
5937
5938         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5939                                             rtm_ipv6_policy, extack);
5940         if (err)
5941                 return err;
5942
5943         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5944             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5945                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5946                 return -EINVAL;
5947         }
5948
5949         for (i = 0; i <= RTA_MAX; i++) {
5950                 if (!tb[i])
5951                         continue;
5952
5953                 switch (i) {
5954                 case RTA_SRC:
5955                 case RTA_DST:
5956                 case RTA_IIF:
5957                 case RTA_OIF:
5958                 case RTA_MARK:
5959                 case RTA_UID:
5960                 case RTA_SPORT:
5961                 case RTA_DPORT:
5962                 case RTA_IP_PROTO:
5963                         break;
5964                 default:
5965                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5966                         return -EINVAL;
5967                 }
5968         }
5969
5970         return 0;
5971 }
5972
5973 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5974                               struct netlink_ext_ack *extack)
5975 {
5976         struct net *net = sock_net(in_skb->sk);
5977         struct nlattr *tb[RTA_MAX+1];
5978         int err, iif = 0, oif = 0;
5979         struct fib6_info *from;
5980         struct dst_entry *dst;
5981         struct rt6_info *rt;
5982         struct sk_buff *skb;
5983         struct rtmsg *rtm;
5984         struct flowi6 fl6 = {};
5985         bool fibmatch;
5986
5987         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5988         if (err < 0)
5989                 goto errout;
5990
5991         err = -EINVAL;
5992         rtm = nlmsg_data(nlh);
5993         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5994         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5995
5996         if (tb[RTA_SRC]) {
5997                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5998                         goto errout;
5999
6000                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6001         }
6002
6003         if (tb[RTA_DST]) {
6004                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6005                         goto errout;
6006
6007                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6008         }
6009
6010         if (tb[RTA_IIF])
6011                 iif = nla_get_u32(tb[RTA_IIF]);
6012
6013         if (tb[RTA_OIF])
6014                 oif = nla_get_u32(tb[RTA_OIF]);
6015
6016         if (tb[RTA_MARK])
6017                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6018
6019         if (tb[RTA_UID])
6020                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6021                                            nla_get_u32(tb[RTA_UID]));
6022         else
6023                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6024
6025         if (tb[RTA_SPORT])
6026                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6027
6028         if (tb[RTA_DPORT])
6029                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6030
6031         if (tb[RTA_IP_PROTO]) {
6032                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6033                                                   &fl6.flowi6_proto, AF_INET6,
6034                                                   extack);
6035                 if (err)
6036                         goto errout;
6037         }
6038
6039         if (iif) {
6040                 struct net_device *dev;
6041                 int flags = 0;
6042
6043                 rcu_read_lock();
6044
6045                 dev = dev_get_by_index_rcu(net, iif);
6046                 if (!dev) {
6047                         rcu_read_unlock();
6048                         err = -ENODEV;
6049                         goto errout;
6050                 }
6051
6052                 fl6.flowi6_iif = iif;
6053
6054                 if (!ipv6_addr_any(&fl6.saddr))
6055                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6056
6057                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6058
6059                 rcu_read_unlock();
6060         } else {
6061                 fl6.flowi6_oif = oif;
6062
6063                 dst = ip6_route_output(net, NULL, &fl6);
6064         }
6065
6066
6067         rt = container_of(dst, struct rt6_info, dst);
6068         if (rt->dst.error) {
6069                 err = rt->dst.error;
6070                 ip6_rt_put(rt);
6071                 goto errout;
6072         }
6073
6074         if (rt == net->ipv6.ip6_null_entry) {
6075                 err = rt->dst.error;
6076                 ip6_rt_put(rt);
6077                 goto errout;
6078         }
6079
6080         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6081         if (!skb) {
6082                 ip6_rt_put(rt);
6083                 err = -ENOBUFS;
6084                 goto errout;
6085         }
6086
6087         skb_dst_set(skb, &rt->dst);
6088
6089         rcu_read_lock();
6090         from = rcu_dereference(rt->from);
6091         if (from) {
6092                 if (fibmatch)
6093                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6094                                             iif, RTM_NEWROUTE,
6095                                             NETLINK_CB(in_skb).portid,
6096                                             nlh->nlmsg_seq, 0);
6097                 else
6098                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6099                                             &fl6.saddr, iif, RTM_NEWROUTE,
6100                                             NETLINK_CB(in_skb).portid,
6101                                             nlh->nlmsg_seq, 0);
6102         } else {
6103                 err = -ENETUNREACH;
6104         }
6105         rcu_read_unlock();
6106
6107         if (err < 0) {
6108                 kfree_skb(skb);
6109                 goto errout;
6110         }
6111
6112         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6113 errout:
6114         return err;
6115 }
6116
6117 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6118                      unsigned int nlm_flags)
6119 {
6120         struct sk_buff *skb;
6121         struct net *net = info->nl_net;
6122         u32 seq;
6123         int err;
6124
6125         err = -ENOBUFS;
6126         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6127
6128         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6129         if (!skb)
6130                 goto errout;
6131
6132         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6133                             event, info->portid, seq, nlm_flags);
6134         if (err < 0) {
6135                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6136                 WARN_ON(err == -EMSGSIZE);
6137                 kfree_skb(skb);
6138                 goto errout;
6139         }
6140         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6141                     info->nlh, gfp_any());
6142         return;
6143 errout:
6144         if (err < 0)
6145                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6146 }
6147
6148 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6149                     struct nl_info *info)
6150 {
6151         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6152         struct sk_buff *skb;
6153         int err = -ENOBUFS;
6154
6155         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6156         if (!skb)
6157                 goto errout;
6158
6159         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6160                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6161         if (err < 0) {
6162                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6163                 WARN_ON(err == -EMSGSIZE);
6164                 kfree_skb(skb);
6165                 goto errout;
6166         }
6167         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6168                     info->nlh, gfp_any());
6169         return;
6170 errout:
6171         if (err < 0)
6172                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6173 }
6174
6175 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6176                             bool offload, bool trap, bool offload_failed)
6177 {
6178         struct sk_buff *skb;
6179         int err;
6180
6181         if (f6i->offload == offload && f6i->trap == trap &&
6182             f6i->offload_failed == offload_failed)
6183                 return;
6184
6185         f6i->offload = offload;
6186         f6i->trap = trap;
6187
6188         /* 2 means send notifications only if offload_failed was changed. */
6189         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6190             f6i->offload_failed == offload_failed)
6191                 return;
6192
6193         f6i->offload_failed = offload_failed;
6194
6195         if (!rcu_access_pointer(f6i->fib6_node))
6196                 /* The route was removed from the tree, do not send
6197                  * notification.
6198                  */
6199                 return;
6200
6201         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6202                 return;
6203
6204         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6205         if (!skb) {
6206                 err = -ENOBUFS;
6207                 goto errout;
6208         }
6209
6210         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6211                             0, 0);
6212         if (err < 0) {
6213                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6214                 WARN_ON(err == -EMSGSIZE);
6215                 kfree_skb(skb);
6216                 goto errout;
6217         }
6218
6219         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6220         return;
6221
6222 errout:
6223         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6224 }
6225 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6226
6227 static int ip6_route_dev_notify(struct notifier_block *this,
6228                                 unsigned long event, void *ptr)
6229 {
6230         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6231         struct net *net = dev_net(dev);
6232
6233         if (!(dev->flags & IFF_LOOPBACK))
6234                 return NOTIFY_OK;
6235
6236         if (event == NETDEV_REGISTER) {
6237                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6238                 net->ipv6.ip6_null_entry->dst.dev = dev;
6239                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6240 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6241                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6242                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6243                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6244                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6245 #endif
6246          } else if (event == NETDEV_UNREGISTER &&
6247                     dev->reg_state != NETREG_UNREGISTERED) {
6248                 /* NETDEV_UNREGISTER could be fired for multiple times by
6249                  * netdev_wait_allrefs(). Make sure we only call this once.
6250                  */
6251                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6252 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6253                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6254                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6255 #endif
6256         }
6257
6258         return NOTIFY_OK;
6259 }
6260
6261 /*
6262  *      /proc
6263  */
6264
6265 #ifdef CONFIG_PROC_FS
6266 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6267 {
6268         struct net *net = (struct net *)seq->private;
6269         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6270                    net->ipv6.rt6_stats->fib_nodes,
6271                    net->ipv6.rt6_stats->fib_route_nodes,
6272                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6273                    net->ipv6.rt6_stats->fib_rt_entries,
6274                    net->ipv6.rt6_stats->fib_rt_cache,
6275                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6276                    net->ipv6.rt6_stats->fib_discarded_routes);
6277
6278         return 0;
6279 }
6280 #endif  /* CONFIG_PROC_FS */
6281
6282 #ifdef CONFIG_SYSCTL
6283
6284 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6285                               void *buffer, size_t *lenp, loff_t *ppos)
6286 {
6287         struct net *net;
6288         int delay;
6289         int ret;
6290         if (!write)
6291                 return -EINVAL;
6292
6293         net = (struct net *)ctl->extra1;
6294         delay = net->ipv6.sysctl.flush_delay;
6295         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6296         if (ret)
6297                 return ret;
6298
6299         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6300         return 0;
6301 }
6302
6303 static struct ctl_table ipv6_route_table_template[] = {
6304         {
6305                 .procname       =       "flush",
6306                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6307                 .maxlen         =       sizeof(int),
6308                 .mode           =       0200,
6309                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6310         },
6311         {
6312                 .procname       =       "gc_thresh",
6313                 .data           =       &ip6_dst_ops_template.gc_thresh,
6314                 .maxlen         =       sizeof(int),
6315                 .mode           =       0644,
6316                 .proc_handler   =       proc_dointvec,
6317         },
6318         {
6319                 .procname       =       "max_size",
6320                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6321                 .maxlen         =       sizeof(int),
6322                 .mode           =       0644,
6323                 .proc_handler   =       proc_dointvec,
6324         },
6325         {
6326                 .procname       =       "gc_min_interval",
6327                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6328                 .maxlen         =       sizeof(int),
6329                 .mode           =       0644,
6330                 .proc_handler   =       proc_dointvec_jiffies,
6331         },
6332         {
6333                 .procname       =       "gc_timeout",
6334                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6335                 .maxlen         =       sizeof(int),
6336                 .mode           =       0644,
6337                 .proc_handler   =       proc_dointvec_jiffies,
6338         },
6339         {
6340                 .procname       =       "gc_interval",
6341                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6342                 .maxlen         =       sizeof(int),
6343                 .mode           =       0644,
6344                 .proc_handler   =       proc_dointvec_jiffies,
6345         },
6346         {
6347                 .procname       =       "gc_elasticity",
6348                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6349                 .maxlen         =       sizeof(int),
6350                 .mode           =       0644,
6351                 .proc_handler   =       proc_dointvec,
6352         },
6353         {
6354                 .procname       =       "mtu_expires",
6355                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6356                 .maxlen         =       sizeof(int),
6357                 .mode           =       0644,
6358                 .proc_handler   =       proc_dointvec_jiffies,
6359         },
6360         {
6361                 .procname       =       "min_adv_mss",
6362                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6363                 .maxlen         =       sizeof(int),
6364                 .mode           =       0644,
6365                 .proc_handler   =       proc_dointvec,
6366         },
6367         {
6368                 .procname       =       "gc_min_interval_ms",
6369                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6370                 .maxlen         =       sizeof(int),
6371                 .mode           =       0644,
6372                 .proc_handler   =       proc_dointvec_ms_jiffies,
6373         },
6374         {
6375                 .procname       =       "skip_notify_on_dev_down",
6376                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6377                 .maxlen         =       sizeof(int),
6378                 .mode           =       0644,
6379                 .proc_handler   =       proc_dointvec_minmax,
6380                 .extra1         =       SYSCTL_ZERO,
6381                 .extra2         =       SYSCTL_ONE,
6382         },
6383         { }
6384 };
6385
6386 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6387 {
6388         struct ctl_table *table;
6389
6390         table = kmemdup(ipv6_route_table_template,
6391                         sizeof(ipv6_route_table_template),
6392                         GFP_KERNEL);
6393
6394         if (table) {
6395                 table[0].data = &net->ipv6.sysctl.flush_delay;
6396                 table[0].extra1 = net;
6397                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6398                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6399                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6400                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6401                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6402                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6403                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6404                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6405                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6406                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6407
6408                 /* Don't export sysctls to unprivileged users */
6409                 if (net->user_ns != &init_user_ns)
6410                         table[0].procname = NULL;
6411         }
6412
6413         return table;
6414 }
6415 #endif
6416
6417 static int __net_init ip6_route_net_init(struct net *net)
6418 {
6419         int ret = -ENOMEM;
6420
6421         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6422                sizeof(net->ipv6.ip6_dst_ops));
6423
6424         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6425                 goto out_ip6_dst_ops;
6426
6427         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6428         if (!net->ipv6.fib6_null_entry)
6429                 goto out_ip6_dst_entries;
6430         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6431                sizeof(*net->ipv6.fib6_null_entry));
6432
6433         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6434                                            sizeof(*net->ipv6.ip6_null_entry),
6435                                            GFP_KERNEL);
6436         if (!net->ipv6.ip6_null_entry)
6437                 goto out_fib6_null_entry;
6438         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6439         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6440                          ip6_template_metrics, true);
6441         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6442
6443 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6444         net->ipv6.fib6_has_custom_rules = false;
6445         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6446                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6447                                                GFP_KERNEL);
6448         if (!net->ipv6.ip6_prohibit_entry)
6449                 goto out_ip6_null_entry;
6450         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6451         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6452                          ip6_template_metrics, true);
6453         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6454
6455         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6456                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6457                                                GFP_KERNEL);
6458         if (!net->ipv6.ip6_blk_hole_entry)
6459                 goto out_ip6_prohibit_entry;
6460         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6461         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6462                          ip6_template_metrics, true);
6463         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6464 #ifdef CONFIG_IPV6_SUBTREES
6465         net->ipv6.fib6_routes_require_src = 0;
6466 #endif
6467 #endif
6468
6469         net->ipv6.sysctl.flush_delay = 0;
6470         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6471         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6472         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6473         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6474         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6475         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6476         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6477         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6478
6479         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6480
6481         ret = 0;
6482 out:
6483         return ret;
6484
6485 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6486 out_ip6_prohibit_entry:
6487         kfree(net->ipv6.ip6_prohibit_entry);
6488 out_ip6_null_entry:
6489         kfree(net->ipv6.ip6_null_entry);
6490 #endif
6491 out_fib6_null_entry:
6492         kfree(net->ipv6.fib6_null_entry);
6493 out_ip6_dst_entries:
6494         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6495 out_ip6_dst_ops:
6496         goto out;
6497 }
6498
6499 static void __net_exit ip6_route_net_exit(struct net *net)
6500 {
6501         kfree(net->ipv6.fib6_null_entry);
6502         kfree(net->ipv6.ip6_null_entry);
6503 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6504         kfree(net->ipv6.ip6_prohibit_entry);
6505         kfree(net->ipv6.ip6_blk_hole_entry);
6506 #endif
6507         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6508 }
6509
6510 static int __net_init ip6_route_net_init_late(struct net *net)
6511 {
6512 #ifdef CONFIG_PROC_FS
6513         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6514                         sizeof(struct ipv6_route_iter));
6515         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6516                         rt6_stats_seq_show, NULL);
6517 #endif
6518         return 0;
6519 }
6520
6521 static void __net_exit ip6_route_net_exit_late(struct net *net)
6522 {
6523 #ifdef CONFIG_PROC_FS
6524         remove_proc_entry("ipv6_route", net->proc_net);
6525         remove_proc_entry("rt6_stats", net->proc_net);
6526 #endif
6527 }
6528
6529 static struct pernet_operations ip6_route_net_ops = {
6530         .init = ip6_route_net_init,
6531         .exit = ip6_route_net_exit,
6532 };
6533
6534 static int __net_init ipv6_inetpeer_init(struct net *net)
6535 {
6536         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6537
6538         if (!bp)
6539                 return -ENOMEM;
6540         inet_peer_base_init(bp);
6541         net->ipv6.peers = bp;
6542         return 0;
6543 }
6544
6545 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6546 {
6547         struct inet_peer_base *bp = net->ipv6.peers;
6548
6549         net->ipv6.peers = NULL;
6550         inetpeer_invalidate_tree(bp);
6551         kfree(bp);
6552 }
6553
6554 static struct pernet_operations ipv6_inetpeer_ops = {
6555         .init   =       ipv6_inetpeer_init,
6556         .exit   =       ipv6_inetpeer_exit,
6557 };
6558
6559 static struct pernet_operations ip6_route_net_late_ops = {
6560         .init = ip6_route_net_init_late,
6561         .exit = ip6_route_net_exit_late,
6562 };
6563
6564 static struct notifier_block ip6_route_dev_notifier = {
6565         .notifier_call = ip6_route_dev_notify,
6566         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6567 };
6568
6569 void __init ip6_route_init_special_entries(void)
6570 {
6571         /* Registering of the loopback is done before this portion of code,
6572          * the loopback reference in rt6_info will not be taken, do it
6573          * manually for init_net */
6574         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6575         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6576         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6577   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6578         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6579         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6580         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6581         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6582   #endif
6583 }
6584
6585 #if IS_BUILTIN(CONFIG_IPV6)
6586 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6587 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6588
6589 BTF_ID_LIST(btf_fib6_info_id)
6590 BTF_ID(struct, fib6_info)
6591
6592 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6593         .seq_ops                = &ipv6_route_seq_ops,
6594         .init_seq_private       = bpf_iter_init_seq_net,
6595         .fini_seq_private       = bpf_iter_fini_seq_net,
6596         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6597 };
6598
6599 static struct bpf_iter_reg ipv6_route_reg_info = {
6600         .target                 = "ipv6_route",
6601         .ctx_arg_info_size      = 1,
6602         .ctx_arg_info           = {
6603                 { offsetof(struct bpf_iter__ipv6_route, rt),
6604                   PTR_TO_BTF_ID_OR_NULL },
6605         },
6606         .seq_info               = &ipv6_route_seq_info,
6607 };
6608
6609 static int __init bpf_iter_register(void)
6610 {
6611         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6612         return bpf_iter_reg_target(&ipv6_route_reg_info);
6613 }
6614
6615 static void bpf_iter_unregister(void)
6616 {
6617         bpf_iter_unreg_target(&ipv6_route_reg_info);
6618 }
6619 #endif
6620 #endif
6621
6622 int __init ip6_route_init(void)
6623 {
6624         int ret;
6625         int cpu;
6626
6627         ret = -ENOMEM;
6628         ip6_dst_ops_template.kmem_cachep =
6629                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6630                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6631         if (!ip6_dst_ops_template.kmem_cachep)
6632                 goto out;
6633
6634         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6635         if (ret)
6636                 goto out_kmem_cache;
6637
6638         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6639         if (ret)
6640                 goto out_dst_entries;
6641
6642         ret = register_pernet_subsys(&ip6_route_net_ops);
6643         if (ret)
6644                 goto out_register_inetpeer;
6645
6646         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6647
6648         ret = fib6_init();
6649         if (ret)
6650                 goto out_register_subsys;
6651
6652         ret = xfrm6_init();
6653         if (ret)
6654                 goto out_fib6_init;
6655
6656         ret = fib6_rules_init();
6657         if (ret)
6658                 goto xfrm6_init;
6659
6660         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6661         if (ret)
6662                 goto fib6_rules_init;
6663
6664         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6665                                    inet6_rtm_newroute, NULL, 0);
6666         if (ret < 0)
6667                 goto out_register_late_subsys;
6668
6669         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6670                                    inet6_rtm_delroute, NULL, 0);
6671         if (ret < 0)
6672                 goto out_register_late_subsys;
6673
6674         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6675                                    inet6_rtm_getroute, NULL,
6676                                    RTNL_FLAG_DOIT_UNLOCKED);
6677         if (ret < 0)
6678                 goto out_register_late_subsys;
6679
6680         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6681         if (ret)
6682                 goto out_register_late_subsys;
6683
6684 #if IS_BUILTIN(CONFIG_IPV6)
6685 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6686         ret = bpf_iter_register();
6687         if (ret)
6688                 goto out_register_late_subsys;
6689 #endif
6690 #endif
6691
6692         for_each_possible_cpu(cpu) {
6693                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6694
6695                 INIT_LIST_HEAD(&ul->head);
6696                 spin_lock_init(&ul->lock);
6697         }
6698
6699 out:
6700         return ret;
6701
6702 out_register_late_subsys:
6703         rtnl_unregister_all(PF_INET6);
6704         unregister_pernet_subsys(&ip6_route_net_late_ops);
6705 fib6_rules_init:
6706         fib6_rules_cleanup();
6707 xfrm6_init:
6708         xfrm6_fini();
6709 out_fib6_init:
6710         fib6_gc_cleanup();
6711 out_register_subsys:
6712         unregister_pernet_subsys(&ip6_route_net_ops);
6713 out_register_inetpeer:
6714         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6715 out_dst_entries:
6716         dst_entries_destroy(&ip6_dst_blackhole_ops);
6717 out_kmem_cache:
6718         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6719         goto out;
6720 }
6721
6722 void ip6_route_cleanup(void)
6723 {
6724 #if IS_BUILTIN(CONFIG_IPV6)
6725 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6726         bpf_iter_unregister();
6727 #endif
6728 #endif
6729         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6730         unregister_pernet_subsys(&ip6_route_net_late_ops);
6731         fib6_rules_cleanup();
6732         xfrm6_fini();
6733         fib6_gc_cleanup();
6734         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6735         unregister_pernet_subsys(&ip6_route_net_ops);
6736         dst_entries_destroy(&ip6_dst_blackhole_ops);
6737         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6738 }