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 void              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         struct list_head        quarantine;
 134 };
 135
 136 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 137
 138 void rt6_uncached_list_add(struct rt6_info *rt)
 139 {
 140         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 141
 142         rt->dst.rt_uncached_list = ul;
 143
 144         spin_lock_bh(&ul->lock);
 145         list_add_tail(&rt->dst.rt_uncached, &ul->head);
 146         spin_unlock_bh(&ul->lock);
 147 }
 148
 149 void rt6_uncached_list_del(struct rt6_info *rt)
 150 {
 151         if (!list_empty(&rt->dst.rt_uncached)) {
 152                 struct uncached_list *ul = rt->dst.rt_uncached_list;
 153
 154                 spin_lock_bh(&ul->lock);
 155                 list_del_init(&rt->dst.rt_uncached);
 156                 spin_unlock_bh(&ul->lock);
 157         }
 158 }
 159
 160 static void rt6_uncached_list_flush_dev(struct net_device *dev)
 161 {
 162         int cpu;
 163
 164         for_each_possible_cpu(cpu) {
 165                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 166                 struct rt6_info *rt, *safe;
 167
 168                 if (list_empty(&ul->head))
 169                         continue;
 170
 171                 spin_lock_bh(&ul->lock);
 172                 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
 173                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 174                         struct net_device *rt_dev = rt->dst.dev;
 175                         bool handled = false;
 176
 177                         if (rt_idev->dev == dev) {
 178                                 rt->rt6i_idev = in6_dev_get(blackhole_netdev);
 179                                 in6_dev_put(rt_idev);
 180                                 handled = true;
 181                         }
 182
 183                         if (rt_dev == dev) {
 184                                 rt->dst.dev = blackhole_netdev;
 185                                 netdev_ref_replace(rt_dev, blackhole_netdev,
 186                                                    &rt->dst.dev_tracker,
 187                                                    GFP_ATOMIC);
 188                                 handled = true;
 189                         }
 190                         if (handled)
 191                                 list_move(&rt->dst.rt_uncached,
 192                                           &ul->quarantine);
 193                 }
 194                 spin_unlock_bh(&ul->lock);
 195         }
 196 }
 197
 198 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 199                                              struct sk_buff *skb,
 200                                              const void *daddr)
 201 {
 202         if (!ipv6_addr_any(p))
 203                 return (const void *) p;
 204         else if (skb)
 205                 return &ipv6_hdr(skb)->daddr;
 206         return daddr;
 207 }
 208
 209 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 210                                    struct net_device *dev,
 211                                    struct sk_buff *skb,
 212                                    const void *daddr)
 213 {
 214         struct neighbour *n;
 215
 216         daddr = choose_neigh_daddr(gw, skb, daddr);
 217         n = __ipv6_neigh_lookup(dev, daddr);
 218         if (n)
 219                 return n;
 220
 221         n = neigh_create(&nd_tbl, daddr, dev);
 222         return IS_ERR(n) ? NULL : n;
 223 }
 224
 225 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 226                                               struct sk_buff *skb,
 227                                               const void *daddr)
 228 {
 229         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 230
 231         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 232                                 dst->dev, skb, daddr);
 233 }
 234
 235 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 236 {
 237         struct net_device *dev = dst->dev;
 238         struct rt6_info *rt = (struct rt6_info *)dst;
 239
 240         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 241         if (!daddr)
 242                 return;
 243         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 244                 return;
 245         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 246                 return;
 247         __ipv6_confirm_neigh(dev, daddr);
 248 }
 249
 250 static struct dst_ops ip6_dst_ops_template = {
 251         .family                 =       AF_INET6,
 252         .gc                     =       ip6_dst_gc,
 253         .gc_thresh              =       1024,
 254         .check                  =       ip6_dst_check,
 255         .default_advmss         =       ip6_default_advmss,
 256         .mtu                    =       ip6_mtu,
 257         .cow_metrics            =       dst_cow_metrics_generic,
 258         .destroy                =       ip6_dst_destroy,
 259         .ifdown                 =       ip6_dst_ifdown,
 260         .negative_advice        =       ip6_negative_advice,
 261         .link_failure           =       ip6_link_failure,
 262         .update_pmtu            =       ip6_rt_update_pmtu,
 263         .redirect               =       rt6_do_redirect,
 264         .local_out              =       __ip6_local_out,
 265         .neigh_lookup           =       ip6_dst_neigh_lookup,
 266         .confirm_neigh          =       ip6_confirm_neigh,
 267 };
 268
 269 static struct dst_ops ip6_dst_blackhole_ops = {
 270         .family                 = AF_INET6,
 271         .default_advmss         = ip6_default_advmss,
 272         .neigh_lookup           = ip6_dst_neigh_lookup,
 273         .check                  = ip6_dst_check,
 274         .destroy                = ip6_dst_destroy,
 275         .cow_metrics            = dst_cow_metrics_generic,
 276         .update_pmtu            = dst_blackhole_update_pmtu,
 277         .redirect               = dst_blackhole_redirect,
 278         .mtu                    = dst_blackhole_mtu,
 279 };
 280
 281 static const u32 ip6_template_metrics[RTAX_MAX] = {
 282         [RTAX_HOPLIMIT - 1] = 0,
 283 };
 284
 285 static const struct fib6_info fib6_null_entry_template = {
 286         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 287         .fib6_protocol  = RTPROT_KERNEL,
 288         .fib6_metric    = ~(u32)0,
 289         .fib6_ref       = REFCOUNT_INIT(1),
 290         .fib6_type      = RTN_UNREACHABLE,
 291         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 292 };
 293
 294 static const struct rt6_info ip6_null_entry_template = {
 295         .dst = {
 296                 .__rcuref       = RCUREF_INIT(1),
 297                 .__use          = 1,
 298                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 299                 .error          = -ENETUNREACH,
 300                 .input          = ip6_pkt_discard,
 301                 .output         = ip6_pkt_discard_out,
 302         },
 303         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 304 };
 305
 306 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 307
 308 static const struct rt6_info ip6_prohibit_entry_template = {
 309         .dst = {
 310                 .__rcuref       = RCUREF_INIT(1),
 311                 .__use          = 1,
 312                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 313                 .error          = -EACCES,
 314                 .input          = ip6_pkt_prohibit,
 315                 .output         = ip6_pkt_prohibit_out,
 316         },
 317         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 318 };
 319
 320 static const struct rt6_info ip6_blk_hole_entry_template = {
 321         .dst = {
 322                 .__rcuref       = RCUREF_INIT(1),
 323                 .__use          = 1,
 324                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 325                 .error          = -EINVAL,
 326                 .input          = dst_discard,
 327                 .output         = dst_discard_out,
 328         },
 329         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 330 };
 331
 332 #endif
 333
 334 static void rt6_info_init(struct rt6_info *rt)
 335 {
 336         memset_after(rt, 0, dst);
 337 }
 338
 339 /* allocate dst with ip6_dst_ops */
 340 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 341                                int flags)
 342 {
 343         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 344                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 345
 346         if (rt) {
 347                 rt6_info_init(rt);
 348                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 349         }
 350
 351         return rt;
 352 }
 353 EXPORT_SYMBOL(ip6_dst_alloc);
 354
 355 static void ip6_dst_destroy(struct dst_entry *dst)
 356 {
 357         struct rt6_info *rt = (struct rt6_info *)dst;
 358         struct fib6_info *from;
 359         struct inet6_dev *idev;
 360
 361         ip_dst_metrics_put(dst);
 362         rt6_uncached_list_del(rt);
 363
 364         idev = rt->rt6i_idev;
 365         if (idev) {
 366                 rt->rt6i_idev = NULL;
 367                 in6_dev_put(idev);
 368         }
 369
 370         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 371         fib6_info_release(from);
 372 }
 373
 374 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 375                            int how)
 376 {
 377         struct rt6_info *rt = (struct rt6_info *)dst;
 378         struct inet6_dev *idev = rt->rt6i_idev;
 379
 380         if (idev && idev->dev != blackhole_netdev) {
 381                 struct inet6_dev *blackhole_idev = in6_dev_get(blackhole_netdev);
 382
 383                 if (blackhole_idev) {
 384                         rt->rt6i_idev = blackhole_idev;
 385                         in6_dev_put(idev);
 386                 }
 387         }
 388 }
 389
 390 static bool __rt6_check_expired(const struct rt6_info *rt)
 391 {
 392         if (rt->rt6i_flags & RTF_EXPIRES)
 393                 return time_after(jiffies, rt->dst.expires);
 394         else
 395                 return false;
 396 }
 397
 398 static bool rt6_check_expired(const struct rt6_info *rt)
 399 {
 400         struct fib6_info *from;
 401
 402         from = rcu_dereference(rt->from);
 403
 404         if (rt->rt6i_flags & RTF_EXPIRES) {
 405                 if (time_after(jiffies, rt->dst.expires))
 406                         return true;
 407         } else if (from) {
 408                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 409                         fib6_check_expired(from);
 410         }
 411         return false;
 412 }
 413
 414 void fib6_select_path(const struct net *net, struct fib6_result *res,
 415                       struct flowi6 *fl6, int oif, bool have_oif_match,
 416                       const struct sk_buff *skb, int strict)
 417 {
 418         struct fib6_info *sibling, *next_sibling;
 419         struct fib6_info *match = res->f6i;
 420
 421         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 422                 goto out;
 423
 424         if (match->nh && have_oif_match && res->nh)
 425                 return;
 426
 427         /* We might have already computed the hash for ICMPv6 errors. In such
 428          * case it will always be non-zero. Otherwise now is the time to do it.
 429          */
 430         if (!fl6->mp_hash &&
 431             (!match->nh || nexthop_is_multipath(match->nh)))
 432                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 433
 434         if (unlikely(match->nh)) {
 435                 nexthop_path_fib6_result(res, fl6->mp_hash);
 436                 return;
 437         }
 438
 439         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 440                 goto out;
 441
 442         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 443                                  fib6_siblings) {
 444                 const struct fib6_nh *nh = sibling->fib6_nh;
 445                 int nh_upper_bound;
 446
 447                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 448                 if (fl6->mp_hash > nh_upper_bound)
 449                         continue;
 450                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 451                         break;
 452                 match = sibling;
 453                 break;
 454         }
 455
 456 out:
 457         res->f6i = match;
 458         res->nh = match->fib6_nh;
 459 }
 460
 461 /*
 462  *      Route lookup. rcu_read_lock() should be held.
 463  */
 464
 465 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 466                                const struct in6_addr *saddr, int oif, int flags)
 467 {
 468         const struct net_device *dev;
 469
 470         if (nh->fib_nh_flags & RTNH_F_DEAD)
 471                 return false;
 472
 473         dev = nh->fib_nh_dev;
 474         if (oif) {
 475                 if (dev->ifindex == oif)
 476                         return true;
 477         } else {
 478                 if (ipv6_chk_addr(net, saddr, dev,
 479                                   flags & RT6_LOOKUP_F_IFACE))
 480                         return true;
 481         }
 482
 483         return false;
 484 }
 485
 486 struct fib6_nh_dm_arg {
 487         struct net              *net;
 488         const struct in6_addr   *saddr;
 489         int                     oif;
 490         int                     flags;
 491         struct fib6_nh          *nh;
 492 };
 493
 494 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 495 {
 496         struct fib6_nh_dm_arg *arg = _arg;
 497
 498         arg->nh = nh;
 499         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 500                                   arg->flags);
 501 }
 502
 503 /* returns fib6_nh from nexthop or NULL */
 504 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 505                                         struct fib6_result *res,
 506                                         const struct in6_addr *saddr,
 507                                         int oif, int flags)
 508 {
 509         struct fib6_nh_dm_arg arg = {
 510                 .net   = net,
 511                 .saddr = saddr,
 512                 .oif   = oif,
 513                 .flags = flags,
 514         };
 515
 516         if (nexthop_is_blackhole(nh))
 517                 return NULL;
 518
 519         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 520                 return arg.nh;
 521
 522         return NULL;
 523 }
 524
 525 static void rt6_device_match(struct net *net, struct fib6_result *res,
 526                              const struct in6_addr *saddr, int oif, int flags)
 527 {
 528         struct fib6_info *f6i = res->f6i;
 529         struct fib6_info *spf6i;
 530         struct fib6_nh *nh;
 531
 532         if (!oif && ipv6_addr_any(saddr)) {
 533                 if (unlikely(f6i->nh)) {
 534                         nh = nexthop_fib6_nh(f6i->nh);
 535                         if (nexthop_is_blackhole(f6i->nh))
 536                                 goto out_blackhole;
 537                 } else {
 538                         nh = f6i->fib6_nh;
 539                 }
 540                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 541                         goto out;
 542         }
 543
 544         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 545                 bool matched = false;
 546
 547                 if (unlikely(spf6i->nh)) {
 548                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 549                                               oif, flags);
 550                         if (nh)
 551                                 matched = true;
 552                 } else {
 553                         nh = spf6i->fib6_nh;
 554                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 555                                 matched = true;
 556                 }
 557                 if (matched) {
 558                         res->f6i = spf6i;
 559                         goto out;
 560                 }
 561         }
 562
 563         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 564                 res->f6i = net->ipv6.fib6_null_entry;
 565                 nh = res->f6i->fib6_nh;
 566                 goto out;
 567         }
 568
 569         if (unlikely(f6i->nh)) {
 570                 nh = nexthop_fib6_nh(f6i->nh);
 571                 if (nexthop_is_blackhole(f6i->nh))
 572                         goto out_blackhole;
 573         } else {
 574                 nh = f6i->fib6_nh;
 575         }
 576
 577         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 578                 res->f6i = net->ipv6.fib6_null_entry;
 579                 nh = res->f6i->fib6_nh;
 580         }
 581 out:
 582         res->nh = nh;
 583         res->fib6_type = res->f6i->fib6_type;
 584         res->fib6_flags = res->f6i->fib6_flags;
 585         return;
 586
 587 out_blackhole:
 588         res->fib6_flags |= RTF_REJECT;
 589         res->fib6_type = RTN_BLACKHOLE;
 590         res->nh = nh;
 591 }
 592
 593 #ifdef CONFIG_IPV6_ROUTER_PREF
 594 struct __rt6_probe_work {
 595         struct work_struct work;
 596         struct in6_addr target;
 597         struct net_device *dev;
 598         netdevice_tracker dev_tracker;
 599 };
 600
 601 static void rt6_probe_deferred(struct work_struct *w)
 602 {
 603         struct in6_addr mcaddr;
 604         struct __rt6_probe_work *work =
 605                 container_of(w, struct __rt6_probe_work, work);
 606
 607         addrconf_addr_solict_mult(&work->target, &mcaddr);
 608         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 609         netdev_put(work->dev, &work->dev_tracker);
 610         kfree(work);
 611 }
 612
 613 static void rt6_probe(struct fib6_nh *fib6_nh)
 614 {
 615         struct __rt6_probe_work *work = NULL;
 616         const struct in6_addr *nh_gw;
 617         unsigned long last_probe;
 618         struct neighbour *neigh;
 619         struct net_device *dev;
 620         struct inet6_dev *idev;
 621
 622         /*
 623          * Okay, this does not seem to be appropriate
 624          * for now, however, we need to check if it
 625          * is really so; aka Router Reachability Probing.
 626          *
 627          * Router Reachability Probe MUST be rate-limited
 628          * to no more than one per minute.
 629          */
 630         if (!fib6_nh->fib_nh_gw_family)
 631                 return;
 632
 633         nh_gw = &fib6_nh->fib_nh_gw6;
 634         dev = fib6_nh->fib_nh_dev;
 635         rcu_read_lock();
 636         last_probe = READ_ONCE(fib6_nh->last_probe);
 637         idev = __in6_dev_get(dev);
 638         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 639         if (neigh) {
 640                 if (READ_ONCE(neigh->nud_state) & NUD_VALID)
 641                         goto out;
 642
 643                 write_lock_bh(&neigh->lock);
 644                 if (!(neigh->nud_state & NUD_VALID) &&
 645                     time_after(jiffies,
 646                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 647                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 648                         if (work)
 649                                 __neigh_set_probe_once(neigh);
 650                 }
 651                 write_unlock_bh(&neigh->lock);
 652         } else if (time_after(jiffies, last_probe +
 653                                        idev->cnf.rtr_probe_interval)) {
 654                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 655         }
 656
 657         if (!work || cmpxchg(&fib6_nh->last_probe,
 658                              last_probe, jiffies) != last_probe) {
 659                 kfree(work);
 660         } else {
 661                 INIT_WORK(&work->work, rt6_probe_deferred);
 662                 work->target = *nh_gw;
 663                 netdev_hold(dev, &work->dev_tracker, GFP_ATOMIC);
 664                 work->dev = dev;
 665                 schedule_work(&work->work);
 666         }
 667
 668 out:
 669         rcu_read_unlock();
 670 }
 671 #else
 672 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 673 {
 674 }
 675 #endif
 676
 677 /*
 678  * Default Router Selection (RFC 2461 6.3.6)
 679  */
 680 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 681 {
 682         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 683         struct neighbour *neigh;
 684
 685         rcu_read_lock();
 686         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 687                                           &fib6_nh->fib_nh_gw6);
 688         if (neigh) {
 689                 u8 nud_state = READ_ONCE(neigh->nud_state);
 690
 691                 if (nud_state & NUD_VALID)
 692                         ret = RT6_NUD_SUCCEED;
 693 #ifdef CONFIG_IPV6_ROUTER_PREF
 694                 else if (!(nud_state & NUD_FAILED))
 695                         ret = RT6_NUD_SUCCEED;
 696                 else
 697                         ret = RT6_NUD_FAIL_PROBE;
 698 #endif
 699         } else {
 700                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 701                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 702         }
 703         rcu_read_unlock();
 704
 705         return ret;
 706 }
 707
 708 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 709                            int strict)
 710 {
 711         int m = 0;
 712
 713         if (!oif || nh->fib_nh_dev->ifindex == oif)
 714                 m = 2;
 715
 716         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 717                 return RT6_NUD_FAIL_HARD;
 718 #ifdef CONFIG_IPV6_ROUTER_PREF
 719         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 720 #endif
 721         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 722             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 723                 int n = rt6_check_neigh(nh);
 724                 if (n < 0)
 725                         return n;
 726         }
 727         return m;
 728 }
 729
 730 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 731                        int oif, int strict, int *mpri, bool *do_rr)
 732 {
 733         bool match_do_rr = false;
 734         bool rc = false;
 735         int m;
 736
 737         if (nh->fib_nh_flags & RTNH_F_DEAD)
 738                 goto out;
 739
 740         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 741             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 742             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 743                 goto out;
 744
 745         m = rt6_score_route(nh, fib6_flags, oif, strict);
 746         if (m == RT6_NUD_FAIL_DO_RR) {
 747                 match_do_rr = true;
 748                 m = 0; /* lowest valid score */
 749         } else if (m == RT6_NUD_FAIL_HARD) {
 750                 goto out;
 751         }
 752
 753         if (strict & RT6_LOOKUP_F_REACHABLE)
 754                 rt6_probe(nh);
 755
 756         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 757         if (m > *mpri) {
 758                 *do_rr = match_do_rr;
 759                 *mpri = m;
 760                 rc = true;
 761         }
 762 out:
 763         return rc;
 764 }
 765
 766 struct fib6_nh_frl_arg {
 767         u32             flags;
 768         int             oif;
 769         int             strict;
 770         int             *mpri;
 771         bool            *do_rr;
 772         struct fib6_nh  *nh;
 773 };
 774
 775 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 776 {
 777         struct fib6_nh_frl_arg *arg = _arg;
 778
 779         arg->nh = nh;
 780         return find_match(nh, arg->flags, arg->oif, arg->strict,
 781                           arg->mpri, arg->do_rr);
 782 }
 783
 784 static void __find_rr_leaf(struct fib6_info *f6i_start,
 785                            struct fib6_info *nomatch, u32 metric,
 786                            struct fib6_result *res, struct fib6_info **cont,
 787                            int oif, int strict, bool *do_rr, int *mpri)
 788 {
 789         struct fib6_info *f6i;
 790
 791         for (f6i = f6i_start;
 792              f6i && f6i != nomatch;
 793              f6i = rcu_dereference(f6i->fib6_next)) {
 794                 bool matched = false;
 795                 struct fib6_nh *nh;
 796
 797                 if (cont && f6i->fib6_metric != metric) {
 798                         *cont = f6i;
 799                         return;
 800                 }
 801
 802                 if (fib6_check_expired(f6i))
 803                         continue;
 804
 805                 if (unlikely(f6i->nh)) {
 806                         struct fib6_nh_frl_arg arg = {
 807                                 .flags  = f6i->fib6_flags,
 808                                 .oif    = oif,
 809                                 .strict = strict,
 810                                 .mpri   = mpri,
 811                                 .do_rr  = do_rr
 812                         };
 813
 814                         if (nexthop_is_blackhole(f6i->nh)) {
 815                                 res->fib6_flags = RTF_REJECT;
 816                                 res->fib6_type = RTN_BLACKHOLE;
 817                                 res->f6i = f6i;
 818                                 res->nh = nexthop_fib6_nh(f6i->nh);
 819                                 return;
 820                         }
 821                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 822                                                      &arg)) {
 823                                 matched = true;
 824                                 nh = arg.nh;
 825                         }
 826                 } else {
 827                         nh = f6i->fib6_nh;
 828                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 829                                        mpri, do_rr))
 830                                 matched = true;
 831                 }
 832                 if (matched) {
 833                         res->f6i = f6i;
 834                         res->nh = nh;
 835                         res->fib6_flags = f6i->fib6_flags;
 836                         res->fib6_type = f6i->fib6_type;
 837                 }
 838         }
 839 }
 840
 841 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 842                          struct fib6_info *rr_head, int oif, int strict,
 843                          bool *do_rr, struct fib6_result *res)
 844 {
 845         u32 metric = rr_head->fib6_metric;
 846         struct fib6_info *cont = NULL;
 847         int mpri = -1;
 848
 849         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 850                        oif, strict, do_rr, &mpri);
 851
 852         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 853                        oif, strict, do_rr, &mpri);
 854
 855         if (res->f6i || !cont)
 856                 return;
 857
 858         __find_rr_leaf(cont, NULL, metric, res, NULL,
 859                        oif, strict, do_rr, &mpri);
 860 }
 861
 862 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 863                        struct fib6_result *res, int strict)
 864 {
 865         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 866         struct fib6_info *rt0;
 867         bool do_rr = false;
 868         int key_plen;
 869
 870         /* make sure this function or its helpers sets f6i */
 871         res->f6i = NULL;
 872
 873         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 874                 goto out;
 875
 876         rt0 = rcu_dereference(fn->rr_ptr);
 877         if (!rt0)
 878                 rt0 = leaf;
 879
 880         /* Double check to make sure fn is not an intermediate node
 881          * and fn->leaf does not points to its child's leaf
 882          * (This might happen if all routes under fn are deleted from
 883          * the tree and fib6_repair_tree() is called on the node.)
 884          */
 885         key_plen = rt0->fib6_dst.plen;
 886 #ifdef CONFIG_IPV6_SUBTREES
 887         if (rt0->fib6_src.plen)
 888                 key_plen = rt0->fib6_src.plen;
 889 #endif
 890         if (fn->fn_bit != key_plen)
 891                 goto out;
 892
 893         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 894         if (do_rr) {
 895                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 896
 897                 /* no entries matched; do round-robin */
 898                 if (!next || next->fib6_metric != rt0->fib6_metric)
 899                         next = leaf;
 900
 901                 if (next != rt0) {
 902                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 903                         /* make sure next is not being deleted from the tree */
 904                         if (next->fib6_node)
 905                                 rcu_assign_pointer(fn->rr_ptr, next);
 906                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 907                 }
 908         }
 909
 910 out:
 911         if (!res->f6i) {
 912                 res->f6i = net->ipv6.fib6_null_entry;
 913                 res->nh = res->f6i->fib6_nh;
 914                 res->fib6_flags = res->f6i->fib6_flags;
 915                 res->fib6_type = res->f6i->fib6_type;
 916         }
 917 }
 918
 919 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 920 {
 921         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 922                res->nh->fib_nh_gw_family;
 923 }
 924
 925 #ifdef CONFIG_IPV6_ROUTE_INFO
 926 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 927                   const struct in6_addr *gwaddr)
 928 {
 929         struct net *net = dev_net(dev);
 930         struct route_info *rinfo = (struct route_info *) opt;
 931         struct in6_addr prefix_buf, *prefix;
 932         unsigned int pref;
 933         unsigned long lifetime;
 934         struct fib6_info *rt;
 935
 936         if (len < sizeof(struct route_info)) {
 937                 return -EINVAL;
 938         }
 939
 940         /* Sanity check for prefix_len and length */
 941         if (rinfo->length > 3) {
 942                 return -EINVAL;
 943         } else if (rinfo->prefix_len > 128) {
 944                 return -EINVAL;
 945         } else if (rinfo->prefix_len > 64) {
 946                 if (rinfo->length < 2) {
 947                         return -EINVAL;
 948                 }
 949         } else if (rinfo->prefix_len > 0) {
 950                 if (rinfo->length < 1) {
 951                         return -EINVAL;
 952                 }
 953         }
 954
 955         pref = rinfo->route_pref;
 956         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 957                 return -EINVAL;
 958
 959         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 960
 961         if (rinfo->length == 3)
 962                 prefix = (struct in6_addr *)rinfo->prefix;
 963         else {
 964                 /* this function is safe */
 965                 ipv6_addr_prefix(&prefix_buf,
 966                                  (struct in6_addr *)rinfo->prefix,
 967                                  rinfo->prefix_len);
 968                 prefix = &prefix_buf;
 969         }
 970
 971         if (rinfo->prefix_len == 0)
 972                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 973         else
 974                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 975                                         gwaddr, dev);
 976
 977         if (rt && !lifetime) {
 978                 ip6_del_rt(net, rt, false);
 979                 rt = NULL;
 980         }
 981
 982         if (!rt && lifetime)
 983                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 984                                         dev, pref);
 985         else if (rt)
 986                 rt->fib6_flags = RTF_ROUTEINFO |
 987                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 988
 989         if (rt) {
 990                 if (!addrconf_finite_timeout(lifetime))
 991                         fib6_clean_expires(rt);
 992                 else
 993                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 994
 995                 fib6_info_release(rt);
 996         }
 997         return 0;
 998 }
 999 #endif
1000
1001 /*
1002  *      Misc support functions
1003  */
1004
1005 /* called with rcu_lock held */
1006 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1007 {
1008         struct net_device *dev = res->nh->fib_nh_dev;
1009
1010         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1011                 /* for copies of local routes, dst->dev needs to be the
1012                  * device if it is a master device, the master device if
1013                  * device is enslaved, and the loopback as the default
1014                  */
1015                 if (netif_is_l3_slave(dev) &&
1016                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1017                         dev = l3mdev_master_dev_rcu(dev);
1018                 else if (!netif_is_l3_master(dev))
1019                         dev = dev_net(dev)->loopback_dev;
1020                 /* last case is netif_is_l3_master(dev) is true in which
1021                  * case we want dev returned to be dev
1022                  */
1023         }
1024
1025         return dev;
1026 }
1027
1028 static const int fib6_prop[RTN_MAX + 1] = {
1029         [RTN_UNSPEC]    = 0,
1030         [RTN_UNICAST]   = 0,
1031         [RTN_LOCAL]     = 0,
1032         [RTN_BROADCAST] = 0,
1033         [RTN_ANYCAST]   = 0,
1034         [RTN_MULTICAST] = 0,
1035         [RTN_BLACKHOLE] = -EINVAL,
1036         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1037         [RTN_PROHIBIT]  = -EACCES,
1038         [RTN_THROW]     = -EAGAIN,
1039         [RTN_NAT]       = -EINVAL,
1040         [RTN_XRESOLVE]  = -EINVAL,
1041 };
1042
1043 static int ip6_rt_type_to_error(u8 fib6_type)
1044 {
1045         return fib6_prop[fib6_type];
1046 }
1047
1048 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1049 {
1050         unsigned short flags = 0;
1051
1052         if (rt->dst_nocount)
1053                 flags |= DST_NOCOUNT;
1054         if (rt->dst_nopolicy)
1055                 flags |= DST_NOPOLICY;
1056
1057         return flags;
1058 }
1059
1060 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1061 {
1062         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1063
1064         switch (fib6_type) {
1065         case RTN_BLACKHOLE:
1066                 rt->dst.output = dst_discard_out;
1067                 rt->dst.input = dst_discard;
1068                 break;
1069         case RTN_PROHIBIT:
1070                 rt->dst.output = ip6_pkt_prohibit_out;
1071                 rt->dst.input = ip6_pkt_prohibit;
1072                 break;
1073         case RTN_THROW:
1074         case RTN_UNREACHABLE:
1075         default:
1076                 rt->dst.output = ip6_pkt_discard_out;
1077                 rt->dst.input = ip6_pkt_discard;
1078                 break;
1079         }
1080 }
1081
1082 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1083 {
1084         struct fib6_info *f6i = res->f6i;
1085
1086         if (res->fib6_flags & RTF_REJECT) {
1087                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1088                 return;
1089         }
1090
1091         rt->dst.error = 0;
1092         rt->dst.output = ip6_output;
1093
1094         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1095                 rt->dst.input = ip6_input;
1096         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1097                 rt->dst.input = ip6_mc_input;
1098         } else {
1099                 rt->dst.input = ip6_forward;
1100         }
1101
1102         if (res->nh->fib_nh_lws) {
1103                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1104                 lwtunnel_set_redirect(&rt->dst);
1105         }
1106
1107         rt->dst.lastuse = jiffies;
1108 }
1109
1110 /* Caller must already hold reference to @from */
1111 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1112 {
1113         rt->rt6i_flags &= ~RTF_EXPIRES;
1114         rcu_assign_pointer(rt->from, from);
1115         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1116 }
1117
1118 /* Caller must already hold reference to f6i in result */
1119 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1120 {
1121         const struct fib6_nh *nh = res->nh;
1122         const struct net_device *dev = nh->fib_nh_dev;
1123         struct fib6_info *f6i = res->f6i;
1124
1125         ip6_rt_init_dst(rt, res);
1126
1127         rt->rt6i_dst = f6i->fib6_dst;
1128         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1129         rt->rt6i_flags = res->fib6_flags;
1130         if (nh->fib_nh_gw_family) {
1131                 rt->rt6i_gateway = nh->fib_nh_gw6;
1132                 rt->rt6i_flags |= RTF_GATEWAY;
1133         }
1134         rt6_set_from(rt, f6i);
1135 #ifdef CONFIG_IPV6_SUBTREES
1136         rt->rt6i_src = f6i->fib6_src;
1137 #endif
1138 }
1139
1140 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1141                                         struct in6_addr *saddr)
1142 {
1143         struct fib6_node *pn, *sn;
1144         while (1) {
1145                 if (fn->fn_flags & RTN_TL_ROOT)
1146                         return NULL;
1147                 pn = rcu_dereference(fn->parent);
1148                 sn = FIB6_SUBTREE(pn);
1149                 if (sn && sn != fn)
1150                         fn = fib6_node_lookup(sn, NULL, saddr);
1151                 else
1152                         fn = pn;
1153                 if (fn->fn_flags & RTN_RTINFO)
1154                         return fn;
1155         }
1156 }
1157
1158 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1159 {
1160         struct rt6_info *rt = *prt;
1161
1162         if (dst_hold_safe(&rt->dst))
1163                 return true;
1164         if (net) {
1165                 rt = net->ipv6.ip6_null_entry;
1166                 dst_hold(&rt->dst);
1167         } else {
1168                 rt = NULL;
1169         }
1170         *prt = rt;
1171         return false;
1172 }
1173
1174 /* called with rcu_lock held */
1175 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1176 {
1177         struct net_device *dev = res->nh->fib_nh_dev;
1178         struct fib6_info *f6i = res->f6i;
1179         unsigned short flags;
1180         struct rt6_info *nrt;
1181
1182         if (!fib6_info_hold_safe(f6i))
1183                 goto fallback;
1184
1185         flags = fib6_info_dst_flags(f6i);
1186         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1187         if (!nrt) {
1188                 fib6_info_release(f6i);
1189                 goto fallback;
1190         }
1191
1192         ip6_rt_copy_init(nrt, res);
1193         return nrt;
1194
1195 fallback:
1196         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1197         dst_hold(&nrt->dst);
1198         return nrt;
1199 }
1200
1201 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1202                                              struct fib6_table *table,
1203                                              struct flowi6 *fl6,
1204                                              const struct sk_buff *skb,
1205                                              int flags)
1206 {
1207         struct fib6_result res = {};
1208         struct fib6_node *fn;
1209         struct rt6_info *rt;
1210
1211         rcu_read_lock();
1212         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1213 restart:
1214         res.f6i = rcu_dereference(fn->leaf);
1215         if (!res.f6i)
1216                 res.f6i = net->ipv6.fib6_null_entry;
1217         else
1218                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1219                                  flags);
1220
1221         if (res.f6i == net->ipv6.fib6_null_entry) {
1222                 fn = fib6_backtrack(fn, &fl6->saddr);
1223                 if (fn)
1224                         goto restart;
1225
1226                 rt = net->ipv6.ip6_null_entry;
1227                 dst_hold(&rt->dst);
1228                 goto out;
1229         } else if (res.fib6_flags & RTF_REJECT) {
1230                 goto do_create;
1231         }
1232
1233         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1234                          fl6->flowi6_oif != 0, skb, flags);
1235
1236         /* Search through exception table */
1237         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1238         if (rt) {
1239                 if (ip6_hold_safe(net, &rt))
1240                         dst_use_noref(&rt->dst, jiffies);
1241         } else {
1242 do_create:
1243                 rt = ip6_create_rt_rcu(&res);
1244         }
1245
1246 out:
1247         trace_fib6_table_lookup(net, &res, table, fl6);
1248
1249         rcu_read_unlock();
1250
1251         return rt;
1252 }
1253
1254 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1255                                    const struct sk_buff *skb, int flags)
1256 {
1257         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1258 }
1259 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1260
1261 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1262                             const struct in6_addr *saddr, int oif,
1263                             const struct sk_buff *skb, int strict)
1264 {
1265         struct flowi6 fl6 = {
1266                 .flowi6_oif = oif,
1267                 .daddr = *daddr,
1268         };
1269         struct dst_entry *dst;
1270         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1271
1272         if (saddr) {
1273                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1274                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1275         }
1276
1277         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1278         if (dst->error == 0)
1279                 return (struct rt6_info *) dst;
1280
1281         dst_release(dst);
1282
1283         return NULL;
1284 }
1285 EXPORT_SYMBOL(rt6_lookup);
1286
1287 /* ip6_ins_rt is called with FREE table->tb6_lock.
1288  * It takes new route entry, the addition fails by any reason the
1289  * route is released.
1290  * Caller must hold dst before calling it.
1291  */
1292
1293 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1294                         struct netlink_ext_ack *extack)
1295 {
1296         int err;
1297         struct fib6_table *table;
1298
1299         table = rt->fib6_table;
1300         spin_lock_bh(&table->tb6_lock);
1301         err = fib6_add(&table->tb6_root, rt, info, extack);
1302         spin_unlock_bh(&table->tb6_lock);
1303
1304         return err;
1305 }
1306
1307 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1308 {
1309         struct nl_info info = { .nl_net = net, };
1310
1311         return __ip6_ins_rt(rt, &info, NULL);
1312 }
1313
1314 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1315                                            const struct in6_addr *daddr,
1316                                            const struct in6_addr *saddr)
1317 {
1318         struct fib6_info *f6i = res->f6i;
1319         struct net_device *dev;
1320         struct rt6_info *rt;
1321
1322         /*
1323          *      Clone the route.
1324          */
1325
1326         if (!fib6_info_hold_safe(f6i))
1327                 return NULL;
1328
1329         dev = ip6_rt_get_dev_rcu(res);
1330         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1331         if (!rt) {
1332                 fib6_info_release(f6i);
1333                 return NULL;
1334         }
1335
1336         ip6_rt_copy_init(rt, res);
1337         rt->rt6i_flags |= RTF_CACHE;
1338         rt->rt6i_dst.addr = *daddr;
1339         rt->rt6i_dst.plen = 128;
1340
1341         if (!rt6_is_gw_or_nonexthop(res)) {
1342                 if (f6i->fib6_dst.plen != 128 &&
1343                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1344                         rt->rt6i_flags |= RTF_ANYCAST;
1345 #ifdef CONFIG_IPV6_SUBTREES
1346                 if (rt->rt6i_src.plen && saddr) {
1347                         rt->rt6i_src.addr = *saddr;
1348                         rt->rt6i_src.plen = 128;
1349                 }
1350 #endif
1351         }
1352
1353         return rt;
1354 }
1355
1356 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1357 {
1358         struct fib6_info *f6i = res->f6i;
1359         unsigned short flags = fib6_info_dst_flags(f6i);
1360         struct net_device *dev;
1361         struct rt6_info *pcpu_rt;
1362
1363         if (!fib6_info_hold_safe(f6i))
1364                 return NULL;
1365
1366         rcu_read_lock();
1367         dev = ip6_rt_get_dev_rcu(res);
1368         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1369         rcu_read_unlock();
1370         if (!pcpu_rt) {
1371                 fib6_info_release(f6i);
1372                 return NULL;
1373         }
1374         ip6_rt_copy_init(pcpu_rt, res);
1375         pcpu_rt->rt6i_flags |= RTF_PCPU;
1376
1377         if (f6i->nh)
1378                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1379
1380         return pcpu_rt;
1381 }
1382
1383 static bool rt6_is_valid(const struct rt6_info *rt6)
1384 {
1385         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1386 }
1387
1388 /* It should be called with rcu_read_lock() acquired */
1389 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1390 {
1391         struct rt6_info *pcpu_rt;
1392
1393         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1394
1395         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1396                 struct rt6_info *prev, **p;
1397
1398                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1399                 prev = xchg(p, NULL);
1400                 if (prev) {
1401                         dst_dev_put(&prev->dst);
1402                         dst_release(&prev->dst);
1403                 }
1404
1405                 pcpu_rt = NULL;
1406         }
1407
1408         return pcpu_rt;
1409 }
1410
1411 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1412                                             const struct fib6_result *res)
1413 {
1414         struct rt6_info *pcpu_rt, *prev, **p;
1415
1416         pcpu_rt = ip6_rt_pcpu_alloc(res);
1417         if (!pcpu_rt)
1418                 return NULL;
1419
1420         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1421         prev = cmpxchg(p, NULL, pcpu_rt);
1422         BUG_ON(prev);
1423
1424         if (res->f6i->fib6_destroying) {
1425                 struct fib6_info *from;
1426
1427                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1428                 fib6_info_release(from);
1429         }
1430
1431         return pcpu_rt;
1432 }
1433
1434 /* exception hash table implementation
1435  */
1436 static DEFINE_SPINLOCK(rt6_exception_lock);
1437
1438 /* Remove rt6_ex from hash table and free the memory
1439  * Caller must hold rt6_exception_lock
1440  */
1441 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1442                                  struct rt6_exception *rt6_ex)
1443 {
1444         struct fib6_info *from;
1445         struct net *net;
1446
1447         if (!bucket || !rt6_ex)
1448                 return;
1449
1450         net = dev_net(rt6_ex->rt6i->dst.dev);
1451         net->ipv6.rt6_stats->fib_rt_cache--;
1452
1453         /* purge completely the exception to allow releasing the held resources:
1454          * some [sk] cache may keep the dst around for unlimited time
1455          */
1456         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1457         fib6_info_release(from);
1458         dst_dev_put(&rt6_ex->rt6i->dst);
1459
1460         hlist_del_rcu(&rt6_ex->hlist);
1461         dst_release(&rt6_ex->rt6i->dst);
1462         kfree_rcu(rt6_ex, rcu);
1463         WARN_ON_ONCE(!bucket->depth);
1464         bucket->depth--;
1465 }
1466
1467 /* Remove oldest rt6_ex in bucket and free the memory
1468  * Caller must hold rt6_exception_lock
1469  */
1470 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1471 {
1472         struct rt6_exception *rt6_ex, *oldest = NULL;
1473
1474         if (!bucket)
1475                 return;
1476
1477         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1478                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1479                         oldest = rt6_ex;
1480         }
1481         rt6_remove_exception(bucket, oldest);
1482 }
1483
1484 static u32 rt6_exception_hash(const struct in6_addr *dst,
1485                               const struct in6_addr *src)
1486 {
1487         static siphash_aligned_key_t rt6_exception_key;
1488         struct {
1489                 struct in6_addr dst;
1490                 struct in6_addr src;
1491         } __aligned(SIPHASH_ALIGNMENT) combined = {
1492                 .dst = *dst,
1493         };
1494         u64 val;
1495
1496         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1497
1498 #ifdef CONFIG_IPV6_SUBTREES
1499         if (src)
1500                 combined.src = *src;
1501 #endif
1502         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1503
1504         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1505 }
1506
1507 /* Helper function to find the cached rt in the hash table
1508  * and update bucket pointer to point to the bucket for this
1509  * (daddr, saddr) pair
1510  * Caller must hold rt6_exception_lock
1511  */
1512 static struct rt6_exception *
1513 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1514                               const struct in6_addr *daddr,
1515                               const struct in6_addr *saddr)
1516 {
1517         struct rt6_exception *rt6_ex;
1518         u32 hval;
1519
1520         if (!(*bucket) || !daddr)
1521                 return NULL;
1522
1523         hval = rt6_exception_hash(daddr, saddr);
1524         *bucket += hval;
1525
1526         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1527                 struct rt6_info *rt6 = rt6_ex->rt6i;
1528                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1529
1530 #ifdef CONFIG_IPV6_SUBTREES
1531                 if (matched && saddr)
1532                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1533 #endif
1534                 if (matched)
1535                         return rt6_ex;
1536         }
1537         return NULL;
1538 }
1539
1540 /* Helper function to find the cached rt in the hash table
1541  * and update bucket pointer to point to the bucket for this
1542  * (daddr, saddr) pair
1543  * Caller must hold rcu_read_lock()
1544  */
1545 static struct rt6_exception *
1546 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1547                          const struct in6_addr *daddr,
1548                          const struct in6_addr *saddr)
1549 {
1550         struct rt6_exception *rt6_ex;
1551         u32 hval;
1552
1553         WARN_ON_ONCE(!rcu_read_lock_held());
1554
1555         if (!(*bucket) || !daddr)
1556                 return NULL;
1557
1558         hval = rt6_exception_hash(daddr, saddr);
1559         *bucket += hval;
1560
1561         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1562                 struct rt6_info *rt6 = rt6_ex->rt6i;
1563                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1564
1565 #ifdef CONFIG_IPV6_SUBTREES
1566                 if (matched && saddr)
1567                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1568 #endif
1569                 if (matched)
1570                         return rt6_ex;
1571         }
1572         return NULL;
1573 }
1574
1575 static unsigned int fib6_mtu(const struct fib6_result *res)
1576 {
1577         const struct fib6_nh *nh = res->nh;
1578         unsigned int mtu;
1579
1580         if (res->f6i->fib6_pmtu) {
1581                 mtu = res->f6i->fib6_pmtu;
1582         } else {
1583                 struct net_device *dev = nh->fib_nh_dev;
1584                 struct inet6_dev *idev;
1585
1586                 rcu_read_lock();
1587                 idev = __in6_dev_get(dev);
1588                 mtu = idev->cnf.mtu6;
1589                 rcu_read_unlock();
1590         }
1591
1592         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1593
1594         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1595 }
1596
1597 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1598
1599 /* used when the flushed bit is not relevant, only access to the bucket
1600  * (ie., all bucket users except rt6_insert_exception);
1601  *
1602  * called under rcu lock; sometimes called with rt6_exception_lock held
1603  */
1604 static
1605 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1606                                                        spinlock_t *lock)
1607 {
1608         struct rt6_exception_bucket *bucket;
1609
1610         if (lock)
1611                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1612                                                    lockdep_is_held(lock));
1613         else
1614                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1615
1616         /* remove bucket flushed bit if set */
1617         if (bucket) {
1618                 unsigned long p = (unsigned long)bucket;
1619
1620                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1621                 bucket = (struct rt6_exception_bucket *)p;
1622         }
1623
1624         return bucket;
1625 }
1626
1627 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1628 {
1629         unsigned long p = (unsigned long)bucket;
1630
1631         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1632 }
1633
1634 /* called with rt6_exception_lock held */
1635 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1636                                               spinlock_t *lock)
1637 {
1638         struct rt6_exception_bucket *bucket;
1639         unsigned long p;
1640
1641         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1642                                            lockdep_is_held(lock));
1643
1644         p = (unsigned long)bucket;
1645         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1646         bucket = (struct rt6_exception_bucket *)p;
1647         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1648 }
1649
1650 static int rt6_insert_exception(struct rt6_info *nrt,
1651                                 const struct fib6_result *res)
1652 {
1653         struct net *net = dev_net(nrt->dst.dev);
1654         struct rt6_exception_bucket *bucket;
1655         struct fib6_info *f6i = res->f6i;
1656         struct in6_addr *src_key = NULL;
1657         struct rt6_exception *rt6_ex;
1658         struct fib6_nh *nh = res->nh;
1659         int max_depth;
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         /* Randomize max depth to avoid some side channels attacks. */
1715         max_depth = FIB6_MAX_DEPTH + get_random_u32_below(FIB6_MAX_DEPTH);
1716         while (bucket->depth > max_depth)
1717                 rt6_exception_remove_oldest(bucket);
1718
1719 out:
1720         spin_unlock_bh(&rt6_exception_lock);
1721
1722         /* Update fn->fn_sernum to invalidate all cached dst */
1723         if (!err) {
1724                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1725                 fib6_update_sernum(net, f6i);
1726                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1727                 fib6_force_start_gc(net);
1728         }
1729
1730         return err;
1731 }
1732
1733 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1734 {
1735         struct rt6_exception_bucket *bucket;
1736         struct rt6_exception *rt6_ex;
1737         struct hlist_node *tmp;
1738         int i;
1739
1740         spin_lock_bh(&rt6_exception_lock);
1741
1742         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1743         if (!bucket)
1744                 goto out;
1745
1746         /* Prevent rt6_insert_exception() to recreate the bucket list */
1747         if (!from)
1748                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1749
1750         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1751                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1752                         if (!from ||
1753                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1754                                 rt6_remove_exception(bucket, rt6_ex);
1755                 }
1756                 WARN_ON_ONCE(!from && bucket->depth);
1757                 bucket++;
1758         }
1759 out:
1760         spin_unlock_bh(&rt6_exception_lock);
1761 }
1762
1763 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1764 {
1765         struct fib6_info *f6i = arg;
1766
1767         fib6_nh_flush_exceptions(nh, f6i);
1768
1769         return 0;
1770 }
1771
1772 void rt6_flush_exceptions(struct fib6_info *f6i)
1773 {
1774         if (f6i->nh)
1775                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1776                                          f6i);
1777         else
1778                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1779 }
1780
1781 /* Find cached rt in the hash table inside passed in rt
1782  * Caller has to hold rcu_read_lock()
1783  */
1784 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1785                                            const struct in6_addr *daddr,
1786                                            const struct in6_addr *saddr)
1787 {
1788         const struct in6_addr *src_key = NULL;
1789         struct rt6_exception_bucket *bucket;
1790         struct rt6_exception *rt6_ex;
1791         struct rt6_info *ret = NULL;
1792
1793 #ifdef CONFIG_IPV6_SUBTREES
1794         /* fib6i_src.plen != 0 indicates f6i is in subtree
1795          * and exception table is indexed by a hash of
1796          * both fib6_dst and fib6_src.
1797          * However, the src addr used to create the hash
1798          * might not be exactly the passed in saddr which
1799          * is a /128 addr from the flow.
1800          * So we need to use f6i->fib6_src to redo lookup
1801          * if the passed in saddr does not find anything.
1802          * (See the logic in ip6_rt_cache_alloc() on how
1803          * rt->rt6i_src is updated.)
1804          */
1805         if (res->f6i->fib6_src.plen)
1806                 src_key = saddr;
1807 find_ex:
1808 #endif
1809         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1810         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1811
1812         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1813                 ret = rt6_ex->rt6i;
1814
1815 #ifdef CONFIG_IPV6_SUBTREES
1816         /* Use fib6_src as src_key and redo lookup */
1817         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1818                 src_key = &res->f6i->fib6_src.addr;
1819                 goto find_ex;
1820         }
1821 #endif
1822
1823         return ret;
1824 }
1825
1826 /* Remove the passed in cached rt from the hash table that contains it */
1827 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1828                                     const struct rt6_info *rt)
1829 {
1830         const struct in6_addr *src_key = NULL;
1831         struct rt6_exception_bucket *bucket;
1832         struct rt6_exception *rt6_ex;
1833         int err;
1834
1835         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1836                 return -ENOENT;
1837
1838         spin_lock_bh(&rt6_exception_lock);
1839         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1840
1841 #ifdef CONFIG_IPV6_SUBTREES
1842         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1843          * and exception table is indexed by a hash of
1844          * both rt6i_dst and rt6i_src.
1845          * Otherwise, the exception table is indexed by
1846          * a hash of only rt6i_dst.
1847          */
1848         if (plen)
1849                 src_key = &rt->rt6i_src.addr;
1850 #endif
1851         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1852                                                &rt->rt6i_dst.addr,
1853                                                src_key);
1854         if (rt6_ex) {
1855                 rt6_remove_exception(bucket, rt6_ex);
1856                 err = 0;
1857         } else {
1858                 err = -ENOENT;
1859         }
1860
1861         spin_unlock_bh(&rt6_exception_lock);
1862         return err;
1863 }
1864
1865 struct fib6_nh_excptn_arg {
1866         struct rt6_info *rt;
1867         int             plen;
1868 };
1869
1870 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1871 {
1872         struct fib6_nh_excptn_arg *arg = _arg;
1873         int err;
1874
1875         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1876         if (err == 0)
1877                 return 1;
1878
1879         return 0;
1880 }
1881
1882 static int rt6_remove_exception_rt(struct rt6_info *rt)
1883 {
1884         struct fib6_info *from;
1885
1886         from = rcu_dereference(rt->from);
1887         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1888                 return -EINVAL;
1889
1890         if (from->nh) {
1891                 struct fib6_nh_excptn_arg arg = {
1892                         .rt = rt,
1893                         .plen = from->fib6_src.plen
1894                 };
1895                 int rc;
1896
1897                 /* rc = 1 means an entry was found */
1898                 rc = nexthop_for_each_fib6_nh(from->nh,
1899                                               rt6_nh_remove_exception_rt,
1900                                               &arg);
1901                 return rc ? 0 : -ENOENT;
1902         }
1903
1904         return fib6_nh_remove_exception(from->fib6_nh,
1905                                         from->fib6_src.plen, rt);
1906 }
1907
1908 /* Find rt6_ex which contains the passed in rt cache and
1909  * refresh its stamp
1910  */
1911 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1912                                      const struct rt6_info *rt)
1913 {
1914         const struct in6_addr *src_key = NULL;
1915         struct rt6_exception_bucket *bucket;
1916         struct rt6_exception *rt6_ex;
1917
1918         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1919 #ifdef CONFIG_IPV6_SUBTREES
1920         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1921          * and exception table is indexed by a hash of
1922          * both rt6i_dst and rt6i_src.
1923          * Otherwise, the exception table is indexed by
1924          * a hash of only rt6i_dst.
1925          */
1926         if (plen)
1927                 src_key = &rt->rt6i_src.addr;
1928 #endif
1929         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1930         if (rt6_ex)
1931                 rt6_ex->stamp = jiffies;
1932 }
1933
1934 struct fib6_nh_match_arg {
1935         const struct net_device *dev;
1936         const struct in6_addr   *gw;
1937         struct fib6_nh          *match;
1938 };
1939
1940 /* determine if fib6_nh has given device and gateway */
1941 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1942 {
1943         struct fib6_nh_match_arg *arg = _arg;
1944
1945         if (arg->dev != nh->fib_nh_dev ||
1946             (arg->gw && !nh->fib_nh_gw_family) ||
1947             (!arg->gw && nh->fib_nh_gw_family) ||
1948             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1949                 return 0;
1950
1951         arg->match = nh;
1952
1953         /* found a match, break the loop */
1954         return 1;
1955 }
1956
1957 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1958 {
1959         struct fib6_info *from;
1960         struct fib6_nh *fib6_nh;
1961
1962         rcu_read_lock();
1963
1964         from = rcu_dereference(rt->from);
1965         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1966                 goto unlock;
1967
1968         if (from->nh) {
1969                 struct fib6_nh_match_arg arg = {
1970                         .dev = rt->dst.dev,
1971                         .gw = &rt->rt6i_gateway,
1972                 };
1973
1974                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1975
1976                 if (!arg.match)
1977                         goto unlock;
1978                 fib6_nh = arg.match;
1979         } else {
1980                 fib6_nh = from->fib6_nh;
1981         }
1982         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1983 unlock:
1984         rcu_read_unlock();
1985 }
1986
1987 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1988                                          struct rt6_info *rt, int mtu)
1989 {
1990         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1991          * lowest MTU in the path: always allow updating the route PMTU to
1992          * reflect PMTU decreases.
1993          *
1994          * If the new MTU is higher, and the route PMTU is equal to the local
1995          * MTU, this means the old MTU is the lowest in the path, so allow
1996          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1997          * handle this.
1998          */
1999
2000         if (dst_mtu(&rt->dst) >= mtu)
2001                 return true;
2002
2003         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2004                 return true;
2005
2006         return false;
2007 }
2008
2009 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2010                                        const struct fib6_nh *nh, int mtu)
2011 {
2012         struct rt6_exception_bucket *bucket;
2013         struct rt6_exception *rt6_ex;
2014         int i;
2015
2016         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2017         if (!bucket)
2018                 return;
2019
2020         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2021                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2022                         struct rt6_info *entry = rt6_ex->rt6i;
2023
2024                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2025                          * route), the metrics of its rt->from have already
2026                          * been updated.
2027                          */
2028                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2029                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2030                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2031                 }
2032                 bucket++;
2033         }
2034 }
2035
2036 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2037
2038 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2039                                             const struct in6_addr *gateway)
2040 {
2041         struct rt6_exception_bucket *bucket;
2042         struct rt6_exception *rt6_ex;
2043         struct hlist_node *tmp;
2044         int i;
2045
2046         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2047                 return;
2048
2049         spin_lock_bh(&rt6_exception_lock);
2050         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2051         if (bucket) {
2052                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2053                         hlist_for_each_entry_safe(rt6_ex, tmp,
2054                                                   &bucket->chain, hlist) {
2055                                 struct rt6_info *entry = rt6_ex->rt6i;
2056
2057                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2058                                     RTF_CACHE_GATEWAY &&
2059                                     ipv6_addr_equal(gateway,
2060                                                     &entry->rt6i_gateway)) {
2061                                         rt6_remove_exception(bucket, rt6_ex);
2062                                 }
2063                         }
2064                         bucket++;
2065                 }
2066         }
2067
2068         spin_unlock_bh(&rt6_exception_lock);
2069 }
2070
2071 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2072                                       struct rt6_exception *rt6_ex,
2073                                       struct fib6_gc_args *gc_args,
2074                                       unsigned long now)
2075 {
2076         struct rt6_info *rt = rt6_ex->rt6i;
2077
2078         /* we are pruning and obsoleting aged-out and non gateway exceptions
2079          * even if others have still references to them, so that on next
2080          * dst_check() such references can be dropped.
2081          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2082          * expired, independently from their aging, as per RFC 8201 section 4
2083          */
2084         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2085                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2086                         RT6_TRACE("aging clone %p\n", rt);
2087                         rt6_remove_exception(bucket, rt6_ex);
2088                         return;
2089                 }
2090         } else if (time_after(jiffies, rt->dst.expires)) {
2091                 RT6_TRACE("purging expired route %p\n", rt);
2092                 rt6_remove_exception(bucket, rt6_ex);
2093                 return;
2094         }
2095
2096         if (rt->rt6i_flags & RTF_GATEWAY) {
2097                 struct neighbour *neigh;
2098
2099                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2100
2101                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2102                         RT6_TRACE("purging route %p via non-router but gateway\n",
2103                                   rt);
2104                         rt6_remove_exception(bucket, rt6_ex);
2105                         return;
2106                 }
2107         }
2108
2109         gc_args->more++;
2110 }
2111
2112 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2113                                    struct fib6_gc_args *gc_args,
2114                                    unsigned long now)
2115 {
2116         struct rt6_exception_bucket *bucket;
2117         struct rt6_exception *rt6_ex;
2118         struct hlist_node *tmp;
2119         int i;
2120
2121         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2122                 return;
2123
2124         rcu_read_lock_bh();
2125         spin_lock(&rt6_exception_lock);
2126         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2127         if (bucket) {
2128                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2129                         hlist_for_each_entry_safe(rt6_ex, tmp,
2130                                                   &bucket->chain, hlist) {
2131                                 rt6_age_examine_exception(bucket, rt6_ex,
2132                                                           gc_args, now);
2133                         }
2134                         bucket++;
2135                 }
2136         }
2137         spin_unlock(&rt6_exception_lock);
2138         rcu_read_unlock_bh();
2139 }
2140
2141 struct fib6_nh_age_excptn_arg {
2142         struct fib6_gc_args     *gc_args;
2143         unsigned long           now;
2144 };
2145
2146 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2147 {
2148         struct fib6_nh_age_excptn_arg *arg = _arg;
2149
2150         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2151         return 0;
2152 }
2153
2154 void rt6_age_exceptions(struct fib6_info *f6i,
2155                         struct fib6_gc_args *gc_args,
2156                         unsigned long now)
2157 {
2158         if (f6i->nh) {
2159                 struct fib6_nh_age_excptn_arg arg = {
2160                         .gc_args = gc_args,
2161                         .now = now
2162                 };
2163
2164                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2165                                          &arg);
2166         } else {
2167                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2168         }
2169 }
2170
2171 /* must be called with rcu lock held */
2172 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2173                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2174 {
2175         struct fib6_node *fn, *saved_fn;
2176
2177         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2178         saved_fn = fn;
2179
2180 redo_rt6_select:
2181         rt6_select(net, fn, oif, res, strict);
2182         if (res->f6i == net->ipv6.fib6_null_entry) {
2183                 fn = fib6_backtrack(fn, &fl6->saddr);
2184                 if (fn)
2185                         goto redo_rt6_select;
2186                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2187                         /* also consider unreachable route */
2188                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2189                         fn = saved_fn;
2190                         goto redo_rt6_select;
2191                 }
2192         }
2193
2194         trace_fib6_table_lookup(net, res, table, fl6);
2195
2196         return 0;
2197 }
2198
2199 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2200                                int oif, struct flowi6 *fl6,
2201                                const struct sk_buff *skb, int flags)
2202 {
2203         struct fib6_result res = {};
2204         struct rt6_info *rt = NULL;
2205         int strict = 0;
2206
2207         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2208                      !rcu_read_lock_held());
2209
2210         strict |= flags & RT6_LOOKUP_F_IFACE;
2211         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2212         if (net->ipv6.devconf_all->forwarding == 0)
2213                 strict |= RT6_LOOKUP_F_REACHABLE;
2214
2215         rcu_read_lock();
2216
2217         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2218         if (res.f6i == net->ipv6.fib6_null_entry)
2219                 goto out;
2220
2221         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2222
2223         /*Search through exception table */
2224         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2225         if (rt) {
2226                 goto out;
2227         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2228                             !res.nh->fib_nh_gw_family)) {
2229                 /* Create a RTF_CACHE clone which will not be
2230                  * owned by the fib6 tree.  It is for the special case where
2231                  * the daddr in the skb during the neighbor look-up is different
2232                  * from the fl6->daddr used to look-up route here.
2233                  */
2234                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2235
2236                 if (rt) {
2237                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2238                          * As rt6_uncached_list_add() does not consume refcnt,
2239                          * this refcnt is always returned to the caller even
2240                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2241                          */
2242                         rt6_uncached_list_add(rt);
2243                         rcu_read_unlock();
2244
2245                         return rt;
2246                 }
2247         } else {
2248                 /* Get a percpu copy */
2249                 local_bh_disable();
2250                 rt = rt6_get_pcpu_route(&res);
2251
2252                 if (!rt)
2253                         rt = rt6_make_pcpu_route(net, &res);
2254
2255                 local_bh_enable();
2256         }
2257 out:
2258         if (!rt)
2259                 rt = net->ipv6.ip6_null_entry;
2260         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2261                 ip6_hold_safe(net, &rt);
2262         rcu_read_unlock();
2263
2264         return rt;
2265 }
2266 EXPORT_SYMBOL_GPL(ip6_pol_route);
2267
2268 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2269                                             struct fib6_table *table,
2270                                             struct flowi6 *fl6,
2271                                             const struct sk_buff *skb,
2272                                             int flags)
2273 {
2274         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2275 }
2276
2277 struct dst_entry *ip6_route_input_lookup(struct net *net,
2278                                          struct net_device *dev,
2279                                          struct flowi6 *fl6,
2280                                          const struct sk_buff *skb,
2281                                          int flags)
2282 {
2283         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2284                 flags |= RT6_LOOKUP_F_IFACE;
2285
2286         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2287 }
2288 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2289
2290 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2291                                   struct flow_keys *keys,
2292                                   struct flow_keys *flkeys)
2293 {
2294         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2295         const struct ipv6hdr *key_iph = outer_iph;
2296         struct flow_keys *_flkeys = flkeys;
2297         const struct ipv6hdr *inner_iph;
2298         const struct icmp6hdr *icmph;
2299         struct ipv6hdr _inner_iph;
2300         struct icmp6hdr _icmph;
2301
2302         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2303                 goto out;
2304
2305         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2306                                    sizeof(_icmph), &_icmph);
2307         if (!icmph)
2308                 goto out;
2309
2310         if (!icmpv6_is_err(icmph->icmp6_type))
2311                 goto out;
2312
2313         inner_iph = skb_header_pointer(skb,
2314                                        skb_transport_offset(skb) + sizeof(*icmph),
2315                                        sizeof(_inner_iph), &_inner_iph);
2316         if (!inner_iph)
2317                 goto out;
2318
2319         key_iph = inner_iph;
2320         _flkeys = NULL;
2321 out:
2322         if (_flkeys) {
2323                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2324                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2325                 keys->tags.flow_label = _flkeys->tags.flow_label;
2326                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2327         } else {
2328                 keys->addrs.v6addrs.src = key_iph->saddr;
2329                 keys->addrs.v6addrs.dst = key_iph->daddr;
2330                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2331                 keys->basic.ip_proto = key_iph->nexthdr;
2332         }
2333 }
2334
2335 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2336                                            const struct sk_buff *skb,
2337                                            bool *p_has_inner)
2338 {
2339         u32 hash_fields = ip6_multipath_hash_fields(net);
2340         struct flow_keys keys, hash_keys;
2341
2342         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2343                 return 0;
2344
2345         memset(&hash_keys, 0, sizeof(hash_keys));
2346         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2347
2348         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2349         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2350                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2351         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2352                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2353         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2354                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2355         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2356                 hash_keys.tags.flow_label = keys.tags.flow_label;
2357         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2358                 hash_keys.ports.src = keys.ports.src;
2359         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2360                 hash_keys.ports.dst = keys.ports.dst;
2361
2362         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2363         return flow_hash_from_keys(&hash_keys);
2364 }
2365
2366 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2367                                            const struct sk_buff *skb,
2368                                            bool has_inner)
2369 {
2370         u32 hash_fields = ip6_multipath_hash_fields(net);
2371         struct flow_keys keys, hash_keys;
2372
2373         /* We assume the packet carries an encapsulation, but if none was
2374          * encountered during dissection of the outer flow, then there is no
2375          * point in calling the flow dissector again.
2376          */
2377         if (!has_inner)
2378                 return 0;
2379
2380         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2381                 return 0;
2382
2383         memset(&hash_keys, 0, sizeof(hash_keys));
2384         skb_flow_dissect_flow_keys(skb, &keys, 0);
2385
2386         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2387                 return 0;
2388
2389         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2390                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2391                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2392                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2393                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2394                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2395         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2396                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2397                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2398                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2399                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2400                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2401                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2402                         hash_keys.tags.flow_label = keys.tags.flow_label;
2403         }
2404
2405         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2406                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2407         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2408                 hash_keys.ports.src = keys.ports.src;
2409         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2410                 hash_keys.ports.dst = keys.ports.dst;
2411
2412         return flow_hash_from_keys(&hash_keys);
2413 }
2414
2415 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2416                                          const struct sk_buff *skb)
2417 {
2418         u32 mhash, mhash_inner;
2419         bool has_inner = true;
2420
2421         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2422         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2423
2424         return jhash_2words(mhash, mhash_inner, 0);
2425 }
2426
2427 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2428                                          const struct flowi6 *fl6)
2429 {
2430         u32 hash_fields = ip6_multipath_hash_fields(net);
2431         struct flow_keys hash_keys;
2432
2433         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2434                 return 0;
2435
2436         memset(&hash_keys, 0, sizeof(hash_keys));
2437         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2438         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2439                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2440         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2441                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2442         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2443                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2444         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2445                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2446         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2447                 hash_keys.ports.src = fl6->fl6_sport;
2448         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2449                 hash_keys.ports.dst = fl6->fl6_dport;
2450
2451         return flow_hash_from_keys(&hash_keys);
2452 }
2453
2454 /* if skb is set it will be used and fl6 can be NULL */
2455 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2456                        const struct sk_buff *skb, struct flow_keys *flkeys)
2457 {
2458         struct flow_keys hash_keys;
2459         u32 mhash = 0;
2460
2461         switch (ip6_multipath_hash_policy(net)) {
2462         case 0:
2463                 memset(&hash_keys, 0, sizeof(hash_keys));
2464                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2465                 if (skb) {
2466                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2467                 } else {
2468                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2469                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2470                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2471                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2472                 }
2473                 mhash = flow_hash_from_keys(&hash_keys);
2474                 break;
2475         case 1:
2476                 if (skb) {
2477                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2478                         struct flow_keys keys;
2479
2480                         /* short-circuit if we already have L4 hash present */
2481                         if (skb->l4_hash)
2482                                 return skb_get_hash_raw(skb) >> 1;
2483
2484                         memset(&hash_keys, 0, sizeof(hash_keys));
2485
2486                         if (!flkeys) {
2487                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2488                                 flkeys = &keys;
2489                         }
2490                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2491                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2492                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2493                         hash_keys.ports.src = flkeys->ports.src;
2494                         hash_keys.ports.dst = flkeys->ports.dst;
2495                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2496                 } else {
2497                         memset(&hash_keys, 0, sizeof(hash_keys));
2498                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2499                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2500                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2501                         hash_keys.ports.src = fl6->fl6_sport;
2502                         hash_keys.ports.dst = fl6->fl6_dport;
2503                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2504                 }
2505                 mhash = flow_hash_from_keys(&hash_keys);
2506                 break;
2507         case 2:
2508                 memset(&hash_keys, 0, sizeof(hash_keys));
2509                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2510                 if (skb) {
2511                         struct flow_keys keys;
2512
2513                         if (!flkeys) {
2514                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2515                                 flkeys = &keys;
2516                         }
2517
2518                         /* Inner can be v4 or v6 */
2519                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2520                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2521                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2522                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2523                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2524                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2525                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2526                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2527                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2528                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2529                         } else {
2530                                 /* Same as case 0 */
2531                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2532                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2533                         }
2534                 } else {
2535                         /* Same as case 0 */
2536                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2537                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2538                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2539                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2540                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2541                 }
2542                 mhash = flow_hash_from_keys(&hash_keys);
2543                 break;
2544         case 3:
2545                 if (skb)
2546                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2547                 else
2548                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2549                 break;
2550         }
2551
2552         return mhash >> 1;
2553 }
2554
2555 /* Called with rcu held */
2556 void ip6_route_input(struct sk_buff *skb)
2557 {
2558         const struct ipv6hdr *iph = ipv6_hdr(skb);
2559         struct net *net = dev_net(skb->dev);
2560         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2561         struct ip_tunnel_info *tun_info;
2562         struct flowi6 fl6 = {
2563                 .flowi6_iif = skb->dev->ifindex,
2564                 .daddr = iph->daddr,
2565                 .saddr = iph->saddr,
2566                 .flowlabel = ip6_flowinfo(iph),
2567                 .flowi6_mark = skb->mark,
2568                 .flowi6_proto = iph->nexthdr,
2569         };
2570         struct flow_keys *flkeys = NULL, _flkeys;
2571
2572         tun_info = skb_tunnel_info(skb);
2573         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2574                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2575
2576         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2577                 flkeys = &_flkeys;
2578
2579         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2580                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2581         skb_dst_drop(skb);
2582         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2583                                                       &fl6, skb, flags));
2584 }
2585
2586 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2587                                              struct fib6_table *table,
2588                                              struct flowi6 *fl6,
2589                                              const struct sk_buff *skb,
2590                                              int flags)
2591 {
2592         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2593 }
2594
2595 static struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2596                                                       const struct sock *sk,
2597                                                       struct flowi6 *fl6,
2598                                                       int flags)
2599 {
2600         bool any_src;
2601
2602         if (ipv6_addr_type(&fl6->daddr) &
2603             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2604                 struct dst_entry *dst;
2605
2606                 /* This function does not take refcnt on the dst */
2607                 dst = l3mdev_link_scope_lookup(net, fl6);
2608                 if (dst)
2609                         return dst;
2610         }
2611
2612         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2613
2614         flags |= RT6_LOOKUP_F_DST_NOREF;
2615         any_src = ipv6_addr_any(&fl6->saddr);
2616         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2617             (fl6->flowi6_oif && any_src))
2618                 flags |= RT6_LOOKUP_F_IFACE;
2619
2620         if (!any_src)
2621                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2622         else if (sk)
2623                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2624
2625         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2626 }
2627
2628 struct dst_entry *ip6_route_output_flags(struct net *net,
2629                                          const struct sock *sk,
2630                                          struct flowi6 *fl6,
2631                                          int flags)
2632 {
2633         struct dst_entry *dst;
2634         struct rt6_info *rt6;
2635
2636         rcu_read_lock();
2637         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2638         rt6 = (struct rt6_info *)dst;
2639         /* For dst cached in uncached_list, refcnt is already taken. */
2640         if (list_empty(&rt6->dst.rt_uncached) && !dst_hold_safe(dst)) {
2641                 dst = &net->ipv6.ip6_null_entry->dst;
2642                 dst_hold(dst);
2643         }
2644         rcu_read_unlock();
2645
2646         return dst;
2647 }
2648 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2649
2650 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2651 {
2652         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2653         struct net_device *loopback_dev = net->loopback_dev;
2654         struct dst_entry *new = NULL;
2655
2656         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2657                        DST_OBSOLETE_DEAD, 0);
2658         if (rt) {
2659                 rt6_info_init(rt);
2660                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2661
2662                 new = &rt->dst;
2663                 new->__use = 1;
2664                 new->input = dst_discard;
2665                 new->output = dst_discard_out;
2666
2667                 dst_copy_metrics(new, &ort->dst);
2668
2669                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2670                 rt->rt6i_gateway = ort->rt6i_gateway;
2671                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2672
2673                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2674 #ifdef CONFIG_IPV6_SUBTREES
2675                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2676 #endif
2677         }
2678
2679         dst_release(dst_orig);
2680         return new ? new : ERR_PTR(-ENOMEM);
2681 }
2682
2683 /*
2684  *      Destination cache support functions
2685  */
2686
2687 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2688 {
2689         u32 rt_cookie = 0;
2690
2691         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2692                 return false;
2693
2694         if (fib6_check_expired(f6i))
2695                 return false;
2696
2697         return true;
2698 }
2699
2700 static struct dst_entry *rt6_check(struct rt6_info *rt,
2701                                    struct fib6_info *from,
2702                                    u32 cookie)
2703 {
2704         u32 rt_cookie = 0;
2705
2706         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2707             rt_cookie != cookie)
2708                 return NULL;
2709
2710         if (rt6_check_expired(rt))
2711                 return NULL;
2712
2713         return &rt->dst;
2714 }
2715
2716 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2717                                             struct fib6_info *from,
2718                                             u32 cookie)
2719 {
2720         if (!__rt6_check_expired(rt) &&
2721             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2722             fib6_check(from, cookie))
2723                 return &rt->dst;
2724         else
2725                 return NULL;
2726 }
2727
2728 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2729                                                         u32 cookie)
2730 {
2731         struct dst_entry *dst_ret;
2732         struct fib6_info *from;
2733         struct rt6_info *rt;
2734
2735         rt = container_of(dst, struct rt6_info, dst);
2736
2737         if (rt->sernum)
2738                 return rt6_is_valid(rt) ? dst : NULL;
2739
2740         rcu_read_lock();
2741
2742         /* All IPV6 dsts are created with ->obsolete set to the value
2743          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2744          * into this function always.
2745          */
2746
2747         from = rcu_dereference(rt->from);
2748
2749         if (from && (rt->rt6i_flags & RTF_PCPU ||
2750             unlikely(!list_empty(&rt->dst.rt_uncached))))
2751                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2752         else
2753                 dst_ret = rt6_check(rt, from, cookie);
2754
2755         rcu_read_unlock();
2756
2757         return dst_ret;
2758 }
2759 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2760
2761 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2762 {
2763         struct rt6_info *rt = (struct rt6_info *) dst;
2764
2765         if (rt) {
2766                 if (rt->rt6i_flags & RTF_CACHE) {
2767                         rcu_read_lock();
2768                         if (rt6_check_expired(rt)) {
2769                                 rt6_remove_exception_rt(rt);
2770                                 dst = NULL;
2771                         }
2772                         rcu_read_unlock();
2773                 } else {
2774                         dst_release(dst);
2775                         dst = NULL;
2776                 }
2777         }
2778         return dst;
2779 }
2780
2781 static void ip6_link_failure(struct sk_buff *skb)
2782 {
2783         struct rt6_info *rt;
2784
2785         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2786
2787         rt = (struct rt6_info *) skb_dst(skb);
2788         if (rt) {
2789                 rcu_read_lock();
2790                 if (rt->rt6i_flags & RTF_CACHE) {
2791                         rt6_remove_exception_rt(rt);
2792                 } else {
2793                         struct fib6_info *from;
2794                         struct fib6_node *fn;
2795
2796                         from = rcu_dereference(rt->from);
2797                         if (from) {
2798                                 fn = rcu_dereference(from->fib6_node);
2799                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2800                                         WRITE_ONCE(fn->fn_sernum, -1);
2801                         }
2802                 }
2803                 rcu_read_unlock();
2804         }
2805 }
2806
2807 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2808 {
2809         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2810                 struct fib6_info *from;
2811
2812                 rcu_read_lock();
2813                 from = rcu_dereference(rt0->from);
2814                 if (from)
2815                         rt0->dst.expires = from->expires;
2816                 rcu_read_unlock();
2817         }
2818
2819         dst_set_expires(&rt0->dst, timeout);
2820         rt0->rt6i_flags |= RTF_EXPIRES;
2821 }
2822
2823 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2824 {
2825         struct net *net = dev_net(rt->dst.dev);
2826
2827         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2828         rt->rt6i_flags |= RTF_MODIFIED;
2829         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2830 }
2831
2832 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2833 {
2834         return !(rt->rt6i_flags & RTF_CACHE) &&
2835                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2836 }
2837
2838 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2839                                  const struct ipv6hdr *iph, u32 mtu,
2840                                  bool confirm_neigh)
2841 {
2842         const struct in6_addr *daddr, *saddr;
2843         struct rt6_info *rt6 = (struct rt6_info *)dst;
2844
2845         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2846          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2847          * [see also comment in rt6_mtu_change_route()]
2848          */
2849
2850         if (iph) {
2851                 daddr = &iph->daddr;
2852                 saddr = &iph->saddr;
2853         } else if (sk) {
2854                 daddr = &sk->sk_v6_daddr;
2855                 saddr = &inet6_sk(sk)->saddr;
2856         } else {
2857                 daddr = NULL;
2858                 saddr = NULL;
2859         }
2860
2861         if (confirm_neigh)
2862                 dst_confirm_neigh(dst, daddr);
2863
2864         if (mtu < IPV6_MIN_MTU)
2865                 return;
2866         if (mtu >= dst_mtu(dst))
2867                 return;
2868
2869         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2870                 rt6_do_update_pmtu(rt6, mtu);
2871                 /* update rt6_ex->stamp for cache */
2872                 if (rt6->rt6i_flags & RTF_CACHE)
2873                         rt6_update_exception_stamp_rt(rt6);
2874         } else if (daddr) {
2875                 struct fib6_result res = {};
2876                 struct rt6_info *nrt6;
2877
2878                 rcu_read_lock();
2879                 res.f6i = rcu_dereference(rt6->from);
2880                 if (!res.f6i)
2881                         goto out_unlock;
2882
2883                 res.fib6_flags = res.f6i->fib6_flags;
2884                 res.fib6_type = res.f6i->fib6_type;
2885
2886                 if (res.f6i->nh) {
2887                         struct fib6_nh_match_arg arg = {
2888                                 .dev = dst->dev,
2889                                 .gw = &rt6->rt6i_gateway,
2890                         };
2891
2892                         nexthop_for_each_fib6_nh(res.f6i->nh,
2893                                                  fib6_nh_find_match, &arg);
2894
2895                         /* fib6_info uses a nexthop that does not have fib6_nh
2896                          * using the dst->dev + gw. Should be impossible.
2897                          */
2898                         if (!arg.match)
2899                                 goto out_unlock;
2900
2901                         res.nh = arg.match;
2902                 } else {
2903                         res.nh = res.f6i->fib6_nh;
2904                 }
2905
2906                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2907                 if (nrt6) {
2908                         rt6_do_update_pmtu(nrt6, mtu);
2909                         if (rt6_insert_exception(nrt6, &res))
2910                                 dst_release_immediate(&nrt6->dst);
2911                 }
2912 out_unlock:
2913                 rcu_read_unlock();
2914         }
2915 }
2916
2917 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2918                                struct sk_buff *skb, u32 mtu,
2919                                bool confirm_neigh)
2920 {
2921         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2922                              confirm_neigh);
2923 }
2924
2925 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2926                      int oif, u32 mark, kuid_t uid)
2927 {
2928         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2929         struct dst_entry *dst;
2930         struct flowi6 fl6 = {
2931                 .flowi6_oif = oif,
2932                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2933                 .daddr = iph->daddr,
2934                 .saddr = iph->saddr,
2935                 .flowlabel = ip6_flowinfo(iph),
2936                 .flowi6_uid = uid,
2937         };
2938
2939         dst = ip6_route_output(net, NULL, &fl6);
2940         if (!dst->error)
2941                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2942         dst_release(dst);
2943 }
2944 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2945
2946 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2947 {
2948         int oif = sk->sk_bound_dev_if;
2949         struct dst_entry *dst;
2950
2951         if (!oif && skb->dev)
2952                 oif = l3mdev_master_ifindex(skb->dev);
2953
2954         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2955
2956         dst = __sk_dst_get(sk);
2957         if (!dst || !dst->obsolete ||
2958             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2959                 return;
2960
2961         bh_lock_sock(sk);
2962         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2963                 ip6_datagram_dst_update(sk, false);
2964         bh_unlock_sock(sk);
2965 }
2966 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2967
2968 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2969                            const struct flowi6 *fl6)
2970 {
2971 #ifdef CONFIG_IPV6_SUBTREES
2972         struct ipv6_pinfo *np = inet6_sk(sk);
2973 #endif
2974
2975         ip6_dst_store(sk, dst,
2976                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2977                       &sk->sk_v6_daddr : NULL,
2978 #ifdef CONFIG_IPV6_SUBTREES
2979                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2980                       &np->saddr :
2981 #endif
2982                       NULL);
2983 }
2984
2985 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2986                                   struct flowi6 *fl6,
2987                                   const struct in6_addr *gw,
2988                                   struct rt6_info **ret)
2989 {
2990         const struct fib6_nh *nh = res->nh;
2991
2992         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2993             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2994                 return false;
2995
2996         /* rt_cache's gateway might be different from its 'parent'
2997          * in the case of an ip redirect.
2998          * So we keep searching in the exception table if the gateway
2999          * is different.
3000          */
3001         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3002                 struct rt6_info *rt_cache;
3003
3004                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3005                 if (rt_cache &&
3006                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3007                         *ret = rt_cache;
3008                         return true;
3009                 }
3010                 return false;
3011         }
3012         return true;
3013 }
3014
3015 struct fib6_nh_rd_arg {
3016         struct fib6_result      *res;
3017         struct flowi6           *fl6;
3018         const struct in6_addr   *gw;
3019         struct rt6_info         **ret;
3020 };
3021
3022 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3023 {
3024         struct fib6_nh_rd_arg *arg = _arg;
3025
3026         arg->res->nh = nh;
3027         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3028 }
3029
3030 /* Handle redirects */
3031 struct ip6rd_flowi {
3032         struct flowi6 fl6;
3033         struct in6_addr gateway;
3034 };
3035
3036 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3037                                              struct fib6_table *table,
3038                                              struct flowi6 *fl6,
3039                                              const struct sk_buff *skb,
3040                                              int flags)
3041 {
3042         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3043         struct rt6_info *ret = NULL;
3044         struct fib6_result res = {};
3045         struct fib6_nh_rd_arg arg = {
3046                 .res = &res,
3047                 .fl6 = fl6,
3048                 .gw  = &rdfl->gateway,
3049                 .ret = &ret
3050         };
3051         struct fib6_info *rt;
3052         struct fib6_node *fn;
3053
3054         /* Get the "current" route for this destination and
3055          * check if the redirect has come from appropriate router.
3056          *
3057          * RFC 4861 specifies that redirects should only be
3058          * accepted if they come from the nexthop to the target.
3059          * Due to the way the routes are chosen, this notion
3060          * is a bit fuzzy and one might need to check all possible
3061          * routes.
3062          */
3063
3064         rcu_read_lock();
3065         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3066 restart:
3067         for_each_fib6_node_rt_rcu(fn) {
3068                 res.f6i = rt;
3069                 if (fib6_check_expired(rt))
3070                         continue;
3071                 if (rt->fib6_flags & RTF_REJECT)
3072                         break;
3073                 if (unlikely(rt->nh)) {
3074                         if (nexthop_is_blackhole(rt->nh))
3075                                 continue;
3076                         /* on match, res->nh is filled in and potentially ret */
3077                         if (nexthop_for_each_fib6_nh(rt->nh,
3078                                                      fib6_nh_redirect_match,
3079                                                      &arg))
3080                                 goto out;
3081                 } else {
3082                         res.nh = rt->fib6_nh;
3083                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3084                                                   &ret))
3085                                 goto out;
3086                 }
3087         }
3088
3089         if (!rt)
3090                 rt = net->ipv6.fib6_null_entry;
3091         else if (rt->fib6_flags & RTF_REJECT) {
3092                 ret = net->ipv6.ip6_null_entry;
3093                 goto out;
3094         }
3095
3096         if (rt == net->ipv6.fib6_null_entry) {
3097                 fn = fib6_backtrack(fn, &fl6->saddr);
3098                 if (fn)
3099                         goto restart;
3100         }
3101
3102         res.f6i = rt;
3103         res.nh = rt->fib6_nh;
3104 out:
3105         if (ret) {
3106                 ip6_hold_safe(net, &ret);
3107         } else {
3108                 res.fib6_flags = res.f6i->fib6_flags;
3109                 res.fib6_type = res.f6i->fib6_type;
3110                 ret = ip6_create_rt_rcu(&res);
3111         }
3112
3113         rcu_read_unlock();
3114
3115         trace_fib6_table_lookup(net, &res, table, fl6);
3116         return ret;
3117 };
3118
3119 static struct dst_entry *ip6_route_redirect(struct net *net,
3120                                             const struct flowi6 *fl6,
3121                                             const struct sk_buff *skb,
3122                                             const struct in6_addr *gateway)
3123 {
3124         int flags = RT6_LOOKUP_F_HAS_SADDR;
3125         struct ip6rd_flowi rdfl;
3126
3127         rdfl.fl6 = *fl6;
3128         rdfl.gateway = *gateway;
3129
3130         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3131                                 flags, __ip6_route_redirect);
3132 }
3133
3134 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3135                   kuid_t uid)
3136 {
3137         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3138         struct dst_entry *dst;
3139         struct flowi6 fl6 = {
3140                 .flowi6_iif = LOOPBACK_IFINDEX,
3141                 .flowi6_oif = oif,
3142                 .flowi6_mark = mark,
3143                 .daddr = iph->daddr,
3144                 .saddr = iph->saddr,
3145                 .flowlabel = ip6_flowinfo(iph),
3146                 .flowi6_uid = uid,
3147         };
3148
3149         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3150         rt6_do_redirect(dst, NULL, skb);
3151         dst_release(dst);
3152 }
3153 EXPORT_SYMBOL_GPL(ip6_redirect);
3154
3155 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3156 {
3157         const struct ipv6hdr *iph = ipv6_hdr(skb);
3158         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3159         struct dst_entry *dst;
3160         struct flowi6 fl6 = {
3161                 .flowi6_iif = LOOPBACK_IFINDEX,
3162                 .flowi6_oif = oif,
3163                 .daddr = msg->dest,
3164                 .saddr = iph->daddr,
3165                 .flowi6_uid = sock_net_uid(net, NULL),
3166         };
3167
3168         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3169         rt6_do_redirect(dst, NULL, skb);
3170         dst_release(dst);
3171 }
3172
3173 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3174 {
3175         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3176                      sk->sk_uid);
3177 }
3178 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3179
3180 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3181 {
3182         struct net_device *dev = dst->dev;
3183         unsigned int mtu = dst_mtu(dst);
3184         struct net *net = dev_net(dev);
3185
3186         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3187
3188         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3189                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3190
3191         /*
3192          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3193          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3194          * IPV6_MAXPLEN is also valid and means: "any MSS,
3195          * rely only on pmtu discovery"
3196          */
3197         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3198                 mtu = IPV6_MAXPLEN;
3199         return mtu;
3200 }
3201
3202 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3203 {
3204         return ip6_dst_mtu_maybe_forward(dst, false);
3205 }
3206 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3207
3208 /* MTU selection:
3209  * 1. mtu on route is locked - use it
3210  * 2. mtu from nexthop exception
3211  * 3. mtu from egress device
3212  *
3213  * based on ip6_dst_mtu_forward and exception logic of
3214  * rt6_find_cached_rt; called with rcu_read_lock
3215  */
3216 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3217                       const struct in6_addr *daddr,
3218                       const struct in6_addr *saddr)
3219 {
3220         const struct fib6_nh *nh = res->nh;
3221         struct fib6_info *f6i = res->f6i;
3222         struct inet6_dev *idev;
3223         struct rt6_info *rt;
3224         u32 mtu = 0;
3225
3226         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3227                 mtu = f6i->fib6_pmtu;
3228                 if (mtu)
3229                         goto out;
3230         }
3231
3232         rt = rt6_find_cached_rt(res, daddr, saddr);
3233         if (unlikely(rt)) {
3234                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3235         } else {
3236                 struct net_device *dev = nh->fib_nh_dev;
3237
3238                 mtu = IPV6_MIN_MTU;
3239                 idev = __in6_dev_get(dev);
3240                 if (idev && idev->cnf.mtu6 > mtu)
3241                         mtu = idev->cnf.mtu6;
3242         }
3243
3244         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3245 out:
3246         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3247 }
3248
3249 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3250                                   struct flowi6 *fl6)
3251 {
3252         struct dst_entry *dst;
3253         struct rt6_info *rt;
3254         struct inet6_dev *idev = in6_dev_get(dev);
3255         struct net *net = dev_net(dev);
3256
3257         if (unlikely(!idev))
3258                 return ERR_PTR(-ENODEV);
3259
3260         rt = ip6_dst_alloc(net, dev, 0);
3261         if (unlikely(!rt)) {
3262                 in6_dev_put(idev);
3263                 dst = ERR_PTR(-ENOMEM);
3264                 goto out;
3265         }
3266
3267         rt->dst.input = ip6_input;
3268         rt->dst.output  = ip6_output;
3269         rt->rt6i_gateway  = fl6->daddr;
3270         rt->rt6i_dst.addr = fl6->daddr;
3271         rt->rt6i_dst.plen = 128;
3272         rt->rt6i_idev     = idev;
3273         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3274
3275         /* Add this dst into uncached_list so that rt6_disable_ip() can
3276          * do proper release of the net_device
3277          */
3278         rt6_uncached_list_add(rt);
3279
3280         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3281
3282 out:
3283         return dst;
3284 }
3285
3286 static void ip6_dst_gc(struct dst_ops *ops)
3287 {
3288         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3289         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3290         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3291         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3292         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3293         unsigned int val;
3294         int entries;
3295
3296         if (time_after(rt_last_gc + rt_min_interval, jiffies))
3297                 goto out;
3298
3299         fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3300         entries = dst_entries_get_slow(ops);
3301         if (entries < ops->gc_thresh)
3302                 atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3303 out:
3304         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3305         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3306 }
3307
3308 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3309                                const struct in6_addr *gw_addr, u32 tbid,
3310                                int flags, struct fib6_result *res)
3311 {
3312         struct flowi6 fl6 = {
3313                 .flowi6_oif = cfg->fc_ifindex,
3314                 .daddr = *gw_addr,
3315                 .saddr = cfg->fc_prefsrc,
3316         };
3317         struct fib6_table *table;
3318         int err;
3319
3320         table = fib6_get_table(net, tbid);
3321         if (!table)
3322                 return -EINVAL;
3323
3324         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3325                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3326
3327         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3328
3329         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3330         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3331                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3332                                  cfg->fc_ifindex != 0, NULL, flags);
3333
3334         return err;
3335 }
3336
3337 static int ip6_route_check_nh_onlink(struct net *net,
3338                                      struct fib6_config *cfg,
3339                                      const struct net_device *dev,
3340                                      struct netlink_ext_ack *extack)
3341 {
3342         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3343         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3344         struct fib6_result res = {};
3345         int err;
3346
3347         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3348         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3349             /* ignore match if it is the default route */
3350             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3351             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3352                 NL_SET_ERR_MSG(extack,
3353                                "Nexthop has invalid gateway or device mismatch");
3354                 err = -EINVAL;
3355         }
3356
3357         return err;
3358 }
3359
3360 static int ip6_route_check_nh(struct net *net,
3361                               struct fib6_config *cfg,
3362                               struct net_device **_dev,
3363                               struct inet6_dev **idev)
3364 {
3365         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3366         struct net_device *dev = _dev ? *_dev : NULL;
3367         int flags = RT6_LOOKUP_F_IFACE;
3368         struct fib6_result res = {};
3369         int err = -EHOSTUNREACH;
3370
3371         if (cfg->fc_table) {
3372                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3373                                           cfg->fc_table, flags, &res);
3374                 /* gw_addr can not require a gateway or resolve to a reject
3375                  * route. If a device is given, it must match the result.
3376                  */
3377                 if (err || res.fib6_flags & RTF_REJECT ||
3378                     res.nh->fib_nh_gw_family ||
3379                     (dev && dev != res.nh->fib_nh_dev))
3380                         err = -EHOSTUNREACH;
3381         }
3382
3383         if (err < 0) {
3384                 struct flowi6 fl6 = {
3385                         .flowi6_oif = cfg->fc_ifindex,
3386                         .daddr = *gw_addr,
3387                 };
3388
3389                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3390                 if (err || res.fib6_flags & RTF_REJECT ||
3391                     res.nh->fib_nh_gw_family)
3392                         err = -EHOSTUNREACH;
3393
3394                 if (err)
3395                         return err;
3396
3397                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3398                                  cfg->fc_ifindex != 0, NULL, flags);
3399         }
3400
3401         err = 0;
3402         if (dev) {
3403                 if (dev != res.nh->fib_nh_dev)
3404                         err = -EHOSTUNREACH;
3405         } else {
3406                 *_dev = dev = res.nh->fib_nh_dev;
3407                 dev_hold(dev);
3408                 *idev = in6_dev_get(dev);
3409         }
3410
3411         return err;
3412 }
3413
3414 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3415                            struct net_device **_dev, struct inet6_dev **idev,
3416                            struct netlink_ext_ack *extack)
3417 {
3418         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3419         int gwa_type = ipv6_addr_type(gw_addr);
3420         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3421         const struct net_device *dev = *_dev;
3422         bool need_addr_check = !dev;
3423         int err = -EINVAL;
3424
3425         /* if gw_addr is local we will fail to detect this in case
3426          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3427          * will return already-added prefix route via interface that
3428          * prefix route was assigned to, which might be non-loopback.
3429          */
3430         if (dev &&
3431             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3432                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3433                 goto out;
3434         }
3435
3436         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3437                 /* IPv6 strictly inhibits using not link-local
3438                  * addresses as nexthop address.
3439                  * Otherwise, router will not able to send redirects.
3440                  * It is very good, but in some (rare!) circumstances
3441                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3442                  * some exceptions. --ANK
3443                  * We allow IPv4-mapped nexthops to support RFC4798-type
3444                  * addressing
3445                  */
3446                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3447                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3448                         goto out;
3449                 }
3450
3451                 rcu_read_lock();
3452
3453                 if (cfg->fc_flags & RTNH_F_ONLINK)
3454                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3455                 else
3456                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3457
3458                 rcu_read_unlock();
3459
3460                 if (err)
3461                         goto out;
3462         }
3463
3464         /* reload in case device was changed */
3465         dev = *_dev;
3466
3467         err = -EINVAL;
3468         if (!dev) {
3469                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3470                 goto out;
3471         } else if (dev->flags & IFF_LOOPBACK) {
3472                 NL_SET_ERR_MSG(extack,
3473                                "Egress device can not be loopback device for this route");
3474                 goto out;
3475         }
3476
3477         /* if we did not check gw_addr above, do so now that the
3478          * egress device has been resolved.
3479          */
3480         if (need_addr_check &&
3481             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3482                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3483                 goto out;
3484         }
3485
3486         err = 0;
3487 out:
3488         return err;
3489 }
3490
3491 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3492 {
3493         if ((flags & RTF_REJECT) ||
3494             (dev && (dev->flags & IFF_LOOPBACK) &&
3495              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3496              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3497                 return true;
3498
3499         return false;
3500 }
3501
3502 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3503                  struct fib6_config *cfg, gfp_t gfp_flags,
3504                  struct netlink_ext_ack *extack)
3505 {
3506         struct net_device *dev = NULL;
3507         struct inet6_dev *idev = NULL;
3508         int addr_type;
3509         int err;
3510
3511         fib6_nh->fib_nh_family = AF_INET6;
3512 #ifdef CONFIG_IPV6_ROUTER_PREF
3513         fib6_nh->last_probe = jiffies;
3514 #endif
3515         if (cfg->fc_is_fdb) {
3516                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3517                 fib6_nh->fib_nh_gw_family = AF_INET6;
3518                 return 0;
3519         }
3520
3521         err = -ENODEV;
3522         if (cfg->fc_ifindex) {
3523                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3524                 if (!dev)
3525                         goto out;
3526                 idev = in6_dev_get(dev);
3527                 if (!idev)
3528                         goto out;
3529         }
3530
3531         if (cfg->fc_flags & RTNH_F_ONLINK) {
3532                 if (!dev) {
3533                         NL_SET_ERR_MSG(extack,
3534                                        "Nexthop device required for onlink");
3535                         goto out;
3536                 }
3537
3538                 if (!(dev->flags & IFF_UP)) {
3539                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3540                         err = -ENETDOWN;
3541                         goto out;
3542                 }
3543
3544                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3545         }
3546
3547         fib6_nh->fib_nh_weight = 1;
3548
3549         /* We cannot add true routes via loopback here,
3550          * they would result in kernel looping; promote them to reject routes
3551          */
3552         addr_type = ipv6_addr_type(&cfg->fc_dst);
3553         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3554                 /* hold loopback dev/idev if we haven't done so. */
3555                 if (dev != net->loopback_dev) {
3556                         if (dev) {
3557                                 dev_put(dev);
3558                                 in6_dev_put(idev);
3559                         }
3560                         dev = net->loopback_dev;
3561                         dev_hold(dev);
3562                         idev = in6_dev_get(dev);
3563                         if (!idev) {
3564                                 err = -ENODEV;
3565                                 goto out;
3566                         }
3567                 }
3568                 goto pcpu_alloc;
3569         }
3570
3571         if (cfg->fc_flags & RTF_GATEWAY) {
3572                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3573                 if (err)
3574                         goto out;
3575
3576                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3577                 fib6_nh->fib_nh_gw_family = AF_INET6;
3578         }
3579
3580         err = -ENODEV;
3581         if (!dev)
3582                 goto out;
3583
3584         if (idev->cnf.disable_ipv6) {
3585                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3586                 err = -EACCES;
3587                 goto out;
3588         }
3589
3590         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3591                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3592                 err = -ENETDOWN;
3593                 goto out;
3594         }
3595
3596         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3597             !netif_carrier_ok(dev))
3598                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3599
3600         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3601                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3602         if (err)
3603                 goto out;
3604
3605 pcpu_alloc:
3606         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3607         if (!fib6_nh->rt6i_pcpu) {
3608                 err = -ENOMEM;
3609                 goto out;
3610         }
3611
3612         fib6_nh->fib_nh_dev = dev;
3613         netdev_tracker_alloc(dev, &fib6_nh->fib_nh_dev_tracker, gfp_flags);
3614
3615         fib6_nh->fib_nh_oif = dev->ifindex;
3616         err = 0;
3617 out:
3618         if (idev)
3619                 in6_dev_put(idev);
3620
3621         if (err) {
3622                 lwtstate_put(fib6_nh->fib_nh_lws);
3623                 fib6_nh->fib_nh_lws = NULL;
3624                 dev_put(dev);
3625         }
3626
3627         return err;
3628 }
3629
3630 void fib6_nh_release(struct fib6_nh *fib6_nh)
3631 {
3632         struct rt6_exception_bucket *bucket;
3633
3634         rcu_read_lock();
3635
3636         fib6_nh_flush_exceptions(fib6_nh, NULL);
3637         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3638         if (bucket) {
3639                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3640                 kfree(bucket);
3641         }
3642
3643         rcu_read_unlock();
3644
3645         fib6_nh_release_dsts(fib6_nh);
3646         free_percpu(fib6_nh->rt6i_pcpu);
3647
3648         fib_nh_common_release(&fib6_nh->nh_common);
3649 }
3650
3651 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3652 {
3653         int cpu;
3654
3655         if (!fib6_nh->rt6i_pcpu)
3656                 return;
3657
3658         for_each_possible_cpu(cpu) {
3659                 struct rt6_info *pcpu_rt, **ppcpu_rt;
3660
3661                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3662                 pcpu_rt = xchg(ppcpu_rt, NULL);
3663                 if (pcpu_rt) {
3664                         dst_dev_put(&pcpu_rt->dst);
3665                         dst_release(&pcpu_rt->dst);
3666                 }
3667         }
3668 }
3669
3670 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3671                                               gfp_t gfp_flags,
3672                                               struct netlink_ext_ack *extack)
3673 {
3674         struct net *net = cfg->fc_nlinfo.nl_net;
3675         struct fib6_info *rt = NULL;
3676         struct nexthop *nh = NULL;
3677         struct fib6_table *table;
3678         struct fib6_nh *fib6_nh;
3679         int err = -EINVAL;
3680         int addr_type;
3681
3682         /* RTF_PCPU is an internal flag; can not be set by userspace */
3683         if (cfg->fc_flags & RTF_PCPU) {
3684                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3685                 goto out;
3686         }
3687
3688         /* RTF_CACHE is an internal flag; can not be set by userspace */
3689         if (cfg->fc_flags & RTF_CACHE) {
3690                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3691                 goto out;
3692         }
3693
3694         if (cfg->fc_type > RTN_MAX) {
3695                 NL_SET_ERR_MSG(extack, "Invalid route type");
3696                 goto out;
3697         }
3698
3699         if (cfg->fc_dst_len > 128) {
3700                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3701                 goto out;
3702         }
3703         if (cfg->fc_src_len > 128) {
3704                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3705                 goto out;
3706         }
3707 #ifndef CONFIG_IPV6_SUBTREES
3708         if (cfg->fc_src_len) {
3709                 NL_SET_ERR_MSG(extack,
3710                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3711                 goto out;
3712         }
3713 #endif
3714         if (cfg->fc_nh_id) {
3715                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3716                 if (!nh) {
3717                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3718                         goto out;
3719                 }
3720                 err = fib6_check_nexthop(nh, cfg, extack);
3721                 if (err)
3722                         goto out;
3723         }
3724
3725         err = -ENOBUFS;
3726         if (cfg->fc_nlinfo.nlh &&
3727             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3728                 table = fib6_get_table(net, cfg->fc_table);
3729                 if (!table) {
3730                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3731                         table = fib6_new_table(net, cfg->fc_table);
3732                 }
3733         } else {
3734                 table = fib6_new_table(net, cfg->fc_table);
3735         }
3736
3737         if (!table)
3738                 goto out;
3739
3740         err = -ENOMEM;
3741         rt = fib6_info_alloc(gfp_flags, !nh);
3742         if (!rt)
3743                 goto out;
3744
3745         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3746                                                extack);
3747         if (IS_ERR(rt->fib6_metrics)) {
3748                 err = PTR_ERR(rt->fib6_metrics);
3749                 /* Do not leave garbage there. */
3750                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3751                 goto out_free;
3752         }
3753
3754         if (cfg->fc_flags & RTF_ADDRCONF)
3755                 rt->dst_nocount = true;
3756
3757         if (cfg->fc_flags & RTF_EXPIRES)
3758                 fib6_set_expires(rt, jiffies +
3759                                 clock_t_to_jiffies(cfg->fc_expires));
3760         else
3761                 fib6_clean_expires(rt);
3762
3763         if (cfg->fc_protocol == RTPROT_UNSPEC)
3764                 cfg->fc_protocol = RTPROT_BOOT;
3765         rt->fib6_protocol = cfg->fc_protocol;
3766
3767         rt->fib6_table = table;
3768         rt->fib6_metric = cfg->fc_metric;
3769         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3770         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3771
3772         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3773         rt->fib6_dst.plen = cfg->fc_dst_len;
3774
3775 #ifdef CONFIG_IPV6_SUBTREES
3776         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3777         rt->fib6_src.plen = cfg->fc_src_len;
3778 #endif
3779         if (nh) {
3780                 if (rt->fib6_src.plen) {
3781                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3782                         goto out_free;
3783                 }
3784                 if (!nexthop_get(nh)) {
3785                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3786                         goto out_free;
3787                 }
3788                 rt->nh = nh;
3789                 fib6_nh = nexthop_fib6_nh(rt->nh);
3790         } else {
3791                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3792                 if (err)
3793                         goto out;
3794
3795                 fib6_nh = rt->fib6_nh;
3796
3797                 /* We cannot add true routes via loopback here, they would
3798                  * result in kernel looping; promote them to reject routes
3799                  */
3800                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3801                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3802                                    addr_type))
3803                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3804         }
3805
3806         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3807                 struct net_device *dev = fib6_nh->fib_nh_dev;
3808
3809                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3810                         NL_SET_ERR_MSG(extack, "Invalid source address");
3811                         err = -EINVAL;
3812                         goto out;
3813                 }
3814                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3815                 rt->fib6_prefsrc.plen = 128;
3816         } else
3817                 rt->fib6_prefsrc.plen = 0;
3818
3819         return rt;
3820 out:
3821         fib6_info_release(rt);
3822         return ERR_PTR(err);
3823 out_free:
3824         ip_fib_metrics_put(rt->fib6_metrics);
3825         kfree(rt);
3826         return ERR_PTR(err);
3827 }
3828
3829 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3830                   struct netlink_ext_ack *extack)
3831 {
3832         struct fib6_info *rt;
3833         int err;
3834
3835         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3836         if (IS_ERR(rt))
3837                 return PTR_ERR(rt);
3838
3839         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3840         fib6_info_release(rt);
3841
3842         return err;
3843 }
3844
3845 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3846 {
3847         struct net *net = info->nl_net;
3848         struct fib6_table *table;
3849         int err;
3850
3851         if (rt == net->ipv6.fib6_null_entry) {
3852                 err = -ENOENT;
3853                 goto out;
3854         }
3855
3856         table = rt->fib6_table;
3857         spin_lock_bh(&table->tb6_lock);
3858         err = fib6_del(rt, info);
3859         spin_unlock_bh(&table->tb6_lock);
3860
3861 out:
3862         fib6_info_release(rt);
3863         return err;
3864 }
3865
3866 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3867 {
3868         struct nl_info info = {
3869                 .nl_net = net,
3870                 .skip_notify = skip_notify
3871         };
3872
3873         return __ip6_del_rt(rt, &info);
3874 }
3875
3876 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3877 {
3878         struct nl_info *info = &cfg->fc_nlinfo;
3879         struct net *net = info->nl_net;
3880         struct sk_buff *skb = NULL;
3881         struct fib6_table *table;
3882         int err = -ENOENT;
3883
3884         if (rt == net->ipv6.fib6_null_entry)
3885                 goto out_put;
3886         table = rt->fib6_table;
3887         spin_lock_bh(&table->tb6_lock);
3888
3889         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3890                 struct fib6_info *sibling, *next_sibling;
3891                 struct fib6_node *fn;
3892
3893                 /* prefer to send a single notification with all hops */
3894                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3895                 if (skb) {
3896                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3897
3898                         if (rt6_fill_node(net, skb, rt, NULL,
3899                                           NULL, NULL, 0, RTM_DELROUTE,
3900                                           info->portid, seq, 0) < 0) {
3901                                 kfree_skb(skb);
3902                                 skb = NULL;
3903                         } else
3904                                 info->skip_notify = 1;
3905                 }
3906
3907                 /* 'rt' points to the first sibling route. If it is not the
3908                  * leaf, then we do not need to send a notification. Otherwise,
3909                  * we need to check if the last sibling has a next route or not
3910                  * and emit a replace or delete notification, respectively.
3911                  */
3912                 info->skip_notify_kernel = 1;
3913                 fn = rcu_dereference_protected(rt->fib6_node,
3914                                             lockdep_is_held(&table->tb6_lock));
3915                 if (rcu_access_pointer(fn->leaf) == rt) {
3916                         struct fib6_info *last_sibling, *replace_rt;
3917
3918                         last_sibling = list_last_entry(&rt->fib6_siblings,
3919                                                        struct fib6_info,
3920                                                        fib6_siblings);
3921                         replace_rt = rcu_dereference_protected(
3922                                             last_sibling->fib6_next,
3923                                             lockdep_is_held(&table->tb6_lock));
3924                         if (replace_rt)
3925                                 call_fib6_entry_notifiers_replace(net,
3926                                                                   replace_rt);
3927                         else
3928                                 call_fib6_multipath_entry_notifiers(net,
3929                                                        FIB_EVENT_ENTRY_DEL,
3930                                                        rt, rt->fib6_nsiblings,
3931                                                        NULL);
3932                 }
3933                 list_for_each_entry_safe(sibling, next_sibling,
3934                                          &rt->fib6_siblings,
3935                                          fib6_siblings) {
3936                         err = fib6_del(sibling, info);
3937                         if (err)
3938                                 goto out_unlock;
3939                 }
3940         }
3941
3942         err = fib6_del(rt, info);
3943 out_unlock:
3944         spin_unlock_bh(&table->tb6_lock);
3945 out_put:
3946         fib6_info_release(rt);
3947
3948         if (skb) {
3949                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3950                             info->nlh, gfp_any());
3951         }
3952         return err;
3953 }
3954
3955 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3956 {
3957         int rc = -ESRCH;
3958
3959         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3960                 goto out;
3961
3962         if (cfg->fc_flags & RTF_GATEWAY &&
3963             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3964                 goto out;
3965
3966         rc = rt6_remove_exception_rt(rt);
3967 out:
3968         return rc;
3969 }
3970
3971 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3972                              struct fib6_nh *nh)
3973 {
3974         struct fib6_result res = {
3975                 .f6i = rt,
3976                 .nh = nh,
3977         };
3978         struct rt6_info *rt_cache;
3979
3980         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3981         if (rt_cache)
3982                 return __ip6_del_cached_rt(rt_cache, cfg);
3983
3984         return 0;
3985 }
3986
3987 struct fib6_nh_del_cached_rt_arg {
3988         struct fib6_config *cfg;
3989         struct fib6_info *f6i;
3990 };
3991
3992 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3993 {
3994         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3995         int rc;
3996
3997         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3998         return rc != -ESRCH ? rc : 0;
3999 }
4000
4001 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4002 {
4003         struct fib6_nh_del_cached_rt_arg arg = {
4004                 .cfg = cfg,
4005                 .f6i = f6i
4006         };
4007
4008         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4009 }
4010
4011 static int ip6_route_del(struct fib6_config *cfg,
4012                          struct netlink_ext_ack *extack)
4013 {
4014         struct fib6_table *table;
4015         struct fib6_info *rt;
4016         struct fib6_node *fn;
4017         int err = -ESRCH;
4018
4019         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4020         if (!table) {
4021                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4022                 return err;
4023         }
4024
4025         rcu_read_lock();
4026
4027         fn = fib6_locate(&table->tb6_root,
4028                          &cfg->fc_dst, cfg->fc_dst_len,
4029                          &cfg->fc_src, cfg->fc_src_len,
4030                          !(cfg->fc_flags & RTF_CACHE));
4031
4032         if (fn) {
4033                 for_each_fib6_node_rt_rcu(fn) {
4034                         struct fib6_nh *nh;
4035
4036                         if (rt->nh && cfg->fc_nh_id &&
4037                             rt->nh->id != cfg->fc_nh_id)
4038                                 continue;
4039
4040                         if (cfg->fc_flags & RTF_CACHE) {
4041                                 int rc = 0;
4042
4043                                 if (rt->nh) {
4044                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4045                                 } else if (cfg->fc_nh_id) {
4046                                         continue;
4047                                 } else {
4048                                         nh = rt->fib6_nh;
4049                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4050                                 }
4051                                 if (rc != -ESRCH) {
4052                                         rcu_read_unlock();
4053                                         return rc;
4054                                 }
4055                                 continue;
4056                         }
4057
4058                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4059                                 continue;
4060                         if (cfg->fc_protocol &&
4061                             cfg->fc_protocol != rt->fib6_protocol)
4062                                 continue;
4063
4064                         if (rt->nh) {
4065                                 if (!fib6_info_hold_safe(rt))
4066                                         continue;
4067                                 rcu_read_unlock();
4068
4069                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4070                         }
4071                         if (cfg->fc_nh_id)
4072                                 continue;
4073
4074                         nh = rt->fib6_nh;
4075                         if (cfg->fc_ifindex &&
4076                             (!nh->fib_nh_dev ||
4077                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4078                                 continue;
4079                         if (cfg->fc_flags & RTF_GATEWAY &&
4080                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4081                                 continue;
4082                         if (!fib6_info_hold_safe(rt))
4083                                 continue;
4084                         rcu_read_unlock();
4085
4086                         /* if gateway was specified only delete the one hop */
4087                         if (cfg->fc_flags & RTF_GATEWAY)
4088                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4089
4090                         return __ip6_del_rt_siblings(rt, cfg);
4091                 }
4092         }
4093         rcu_read_unlock();
4094
4095         return err;
4096 }
4097
4098 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4099 {
4100         struct netevent_redirect netevent;
4101         struct rt6_info *rt, *nrt = NULL;
4102         struct fib6_result res = {};
4103         struct ndisc_options ndopts;
4104         struct inet6_dev *in6_dev;
4105         struct neighbour *neigh;
4106         struct rd_msg *msg;
4107         int optlen, on_link;
4108         u8 *lladdr;
4109
4110         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4111         optlen -= sizeof(*msg);
4112
4113         if (optlen < 0) {
4114                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4115                 return;
4116         }
4117
4118         msg = (struct rd_msg *)icmp6_hdr(skb);
4119
4120         if (ipv6_addr_is_multicast(&msg->dest)) {
4121                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4122                 return;
4123         }
4124
4125         on_link = 0;
4126         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4127                 on_link = 1;
4128         } else if (ipv6_addr_type(&msg->target) !=
4129                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4130                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4131                 return;
4132         }
4133
4134         in6_dev = __in6_dev_get(skb->dev);
4135         if (!in6_dev)
4136                 return;
4137         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4138                 return;
4139
4140         /* RFC2461 8.1:
4141          *      The IP source address of the Redirect MUST be the same as the current
4142          *      first-hop router for the specified ICMP Destination Address.
4143          */
4144
4145         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4146                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4147                 return;
4148         }
4149
4150         lladdr = NULL;
4151         if (ndopts.nd_opts_tgt_lladdr) {
4152                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4153                                              skb->dev);
4154                 if (!lladdr) {
4155                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4156                         return;
4157                 }
4158         }
4159
4160         rt = (struct rt6_info *) dst;
4161         if (rt->rt6i_flags & RTF_REJECT) {
4162                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4163                 return;
4164         }
4165
4166         /* Redirect received -> path was valid.
4167          * Look, redirects are sent only in response to data packets,
4168          * so that this nexthop apparently is reachable. --ANK
4169          */
4170         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4171
4172         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4173         if (!neigh)
4174                 return;
4175
4176         /*
4177          *      We have finally decided to accept it.
4178          */
4179
4180         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4181                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4182                      NEIGH_UPDATE_F_OVERRIDE|
4183                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4184                                      NEIGH_UPDATE_F_ISROUTER)),
4185                      NDISC_REDIRECT, &ndopts);
4186
4187         rcu_read_lock();
4188         res.f6i = rcu_dereference(rt->from);
4189         if (!res.f6i)
4190                 goto out;
4191
4192         if (res.f6i->nh) {
4193                 struct fib6_nh_match_arg arg = {
4194                         .dev = dst->dev,
4195                         .gw = &rt->rt6i_gateway,
4196                 };
4197
4198                 nexthop_for_each_fib6_nh(res.f6i->nh,
4199                                          fib6_nh_find_match, &arg);
4200
4201                 /* fib6_info uses a nexthop that does not have fib6_nh
4202                  * using the dst->dev. Should be impossible
4203                  */
4204                 if (!arg.match)
4205                         goto out;
4206                 res.nh = arg.match;
4207         } else {
4208                 res.nh = res.f6i->fib6_nh;
4209         }
4210
4211         res.fib6_flags = res.f6i->fib6_flags;
4212         res.fib6_type = res.f6i->fib6_type;
4213         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4214         if (!nrt)
4215                 goto out;
4216
4217         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4218         if (on_link)
4219                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4220
4221         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4222
4223         /* rt6_insert_exception() will take care of duplicated exceptions */
4224         if (rt6_insert_exception(nrt, &res)) {
4225                 dst_release_immediate(&nrt->dst);
4226                 goto out;
4227         }
4228
4229         netevent.old = &rt->dst;
4230         netevent.new = &nrt->dst;
4231         netevent.daddr = &msg->dest;
4232         netevent.neigh = neigh;
4233         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4234
4235 out:
4236         rcu_read_unlock();
4237         neigh_release(neigh);
4238 }
4239
4240 #ifdef CONFIG_IPV6_ROUTE_INFO
4241 static struct fib6_info *rt6_get_route_info(struct net *net,
4242                                            const struct in6_addr *prefix, int prefixlen,
4243                                            const struct in6_addr *gwaddr,
4244                                            struct net_device *dev)
4245 {
4246         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4247         int ifindex = dev->ifindex;
4248         struct fib6_node *fn;
4249         struct fib6_info *rt = NULL;
4250         struct fib6_table *table;
4251
4252         table = fib6_get_table(net, tb_id);
4253         if (!table)
4254                 return NULL;
4255
4256         rcu_read_lock();
4257         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4258         if (!fn)
4259                 goto out;
4260
4261         for_each_fib6_node_rt_rcu(fn) {
4262                 /* these routes do not use nexthops */
4263                 if (rt->nh)
4264                         continue;
4265                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4266                         continue;
4267                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4268                     !rt->fib6_nh->fib_nh_gw_family)
4269                         continue;
4270                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4271                         continue;
4272                 if (!fib6_info_hold_safe(rt))
4273                         continue;
4274                 break;
4275         }
4276 out:
4277         rcu_read_unlock();
4278         return rt;
4279 }
4280
4281 static struct fib6_info *rt6_add_route_info(struct net *net,
4282                                            const struct in6_addr *prefix, int prefixlen,
4283                                            const struct in6_addr *gwaddr,
4284                                            struct net_device *dev,
4285                                            unsigned int pref)
4286 {
4287         struct fib6_config cfg = {
4288                 .fc_metric      = IP6_RT_PRIO_USER,
4289                 .fc_ifindex     = dev->ifindex,
4290                 .fc_dst_len     = prefixlen,
4291                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4292                                   RTF_UP | RTF_PREF(pref),
4293                 .fc_protocol = RTPROT_RA,
4294                 .fc_type = RTN_UNICAST,
4295                 .fc_nlinfo.portid = 0,
4296                 .fc_nlinfo.nlh = NULL,
4297                 .fc_nlinfo.nl_net = net,
4298         };
4299
4300         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4301         cfg.fc_dst = *prefix;
4302         cfg.fc_gateway = *gwaddr;
4303
4304         /* We should treat it as a default route if prefix length is 0. */
4305         if (!prefixlen)
4306                 cfg.fc_flags |= RTF_DEFAULT;
4307
4308         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4309
4310         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4311 }
4312 #endif
4313
4314 struct fib6_info *rt6_get_dflt_router(struct net *net,
4315                                      const struct in6_addr *addr,
4316                                      struct net_device *dev)
4317 {
4318         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4319         struct fib6_info *rt;
4320         struct fib6_table *table;
4321
4322         table = fib6_get_table(net, tb_id);
4323         if (!table)
4324                 return NULL;
4325
4326         rcu_read_lock();
4327         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4328                 struct fib6_nh *nh;
4329
4330                 /* RA routes do not use nexthops */
4331                 if (rt->nh)
4332                         continue;
4333
4334                 nh = rt->fib6_nh;
4335                 if (dev == nh->fib_nh_dev &&
4336                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4337                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4338                         break;
4339         }
4340         if (rt && !fib6_info_hold_safe(rt))
4341                 rt = NULL;
4342         rcu_read_unlock();
4343         return rt;
4344 }
4345
4346 struct fib6_info *rt6_add_dflt_router(struct net *net,
4347                                      const struct in6_addr *gwaddr,
4348                                      struct net_device *dev,
4349                                      unsigned int pref,
4350                                      u32 defrtr_usr_metric)
4351 {
4352         struct fib6_config cfg = {
4353                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4354                 .fc_metric      = defrtr_usr_metric,
4355                 .fc_ifindex     = dev->ifindex,
4356                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4357                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4358                 .fc_protocol = RTPROT_RA,
4359                 .fc_type = RTN_UNICAST,
4360                 .fc_nlinfo.portid = 0,
4361                 .fc_nlinfo.nlh = NULL,
4362                 .fc_nlinfo.nl_net = net,
4363         };
4364
4365         cfg.fc_gateway = *gwaddr;
4366
4367         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4368                 struct fib6_table *table;
4369
4370                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4371                 if (table)
4372                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4373         }
4374
4375         return rt6_get_dflt_router(net, gwaddr, dev);
4376 }
4377
4378 static void __rt6_purge_dflt_routers(struct net *net,
4379                                      struct fib6_table *table)
4380 {
4381         struct fib6_info *rt;
4382
4383 restart:
4384         rcu_read_lock();
4385         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4386                 struct net_device *dev = fib6_info_nh_dev(rt);
4387                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4388
4389                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4390                     (!idev || idev->cnf.accept_ra != 2) &&
4391                     fib6_info_hold_safe(rt)) {
4392                         rcu_read_unlock();
4393                         ip6_del_rt(net, rt, false);
4394                         goto restart;
4395                 }
4396         }
4397         rcu_read_unlock();
4398
4399         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4400 }
4401
4402 void rt6_purge_dflt_routers(struct net *net)
4403 {
4404         struct fib6_table *table;
4405         struct hlist_head *head;
4406         unsigned int h;
4407
4408         rcu_read_lock();
4409
4410         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4411                 head = &net->ipv6.fib_table_hash[h];
4412                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4413                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4414                                 __rt6_purge_dflt_routers(net, table);
4415                 }
4416         }
4417
4418         rcu_read_unlock();
4419 }
4420
4421 static void rtmsg_to_fib6_config(struct net *net,
4422                                  struct in6_rtmsg *rtmsg,
4423                                  struct fib6_config *cfg)
4424 {
4425         *cfg = (struct fib6_config){
4426                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4427                          : RT6_TABLE_MAIN,
4428                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4429                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4430                 .fc_expires = rtmsg->rtmsg_info,
4431                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4432                 .fc_src_len = rtmsg->rtmsg_src_len,
4433                 .fc_flags = rtmsg->rtmsg_flags,
4434                 .fc_type = rtmsg->rtmsg_type,
4435
4436                 .fc_nlinfo.nl_net = net,
4437
4438                 .fc_dst = rtmsg->rtmsg_dst,
4439                 .fc_src = rtmsg->rtmsg_src,
4440                 .fc_gateway = rtmsg->rtmsg_gateway,
4441         };
4442 }
4443
4444 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4445 {
4446         struct fib6_config cfg;
4447         int err;
4448
4449         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4450                 return -EINVAL;
4451         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4452                 return -EPERM;
4453
4454         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4455
4456         rtnl_lock();
4457         switch (cmd) {
4458         case SIOCADDRT:
4459                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4460                 break;
4461         case SIOCDELRT:
4462                 err = ip6_route_del(&cfg, NULL);
4463                 break;
4464         }
4465         rtnl_unlock();
4466         return err;
4467 }
4468
4469 /*
4470  *      Drop the packet on the floor
4471  */
4472
4473 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4474 {
4475         struct dst_entry *dst = skb_dst(skb);
4476         struct net *net = dev_net(dst->dev);
4477         struct inet6_dev *idev;
4478         SKB_DR(reason);
4479         int type;
4480
4481         if (netif_is_l3_master(skb->dev) ||
4482             dst->dev == net->loopback_dev)
4483                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4484         else
4485                 idev = ip6_dst_idev(dst);
4486
4487         switch (ipstats_mib_noroutes) {
4488         case IPSTATS_MIB_INNOROUTES:
4489                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4490                 if (type == IPV6_ADDR_ANY) {
4491                         SKB_DR_SET(reason, IP_INADDRERRORS);
4492                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4493                         break;
4494                 }
4495                 SKB_DR_SET(reason, IP_INNOROUTES);
4496                 fallthrough;
4497         case IPSTATS_MIB_OUTNOROUTES:
4498                 SKB_DR_OR(reason, IP_OUTNOROUTES);
4499                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4500                 break;
4501         }
4502
4503         /* Start over by dropping the dst for l3mdev case */
4504         if (netif_is_l3_master(skb->dev))
4505                 skb_dst_drop(skb);
4506
4507         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4508         kfree_skb_reason(skb, reason);
4509         return 0;
4510 }
4511
4512 static int ip6_pkt_discard(struct sk_buff *skb)
4513 {
4514         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4515 }
4516
4517 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4518 {
4519         skb->dev = skb_dst(skb)->dev;
4520         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4521 }
4522
4523 static int ip6_pkt_prohibit(struct sk_buff *skb)
4524 {
4525         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4526 }
4527
4528 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4529 {
4530         skb->dev = skb_dst(skb)->dev;
4531         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4532 }
4533
4534 /*
4535  *      Allocate a dst for local (unicast / anycast) address.
4536  */
4537
4538 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4539                                      struct inet6_dev *idev,
4540                                      const struct in6_addr *addr,
4541                                      bool anycast, gfp_t gfp_flags)
4542 {
4543         struct fib6_config cfg = {
4544                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4545                 .fc_ifindex = idev->dev->ifindex,
4546                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4547                 .fc_dst = *addr,
4548                 .fc_dst_len = 128,
4549                 .fc_protocol = RTPROT_KERNEL,
4550                 .fc_nlinfo.nl_net = net,
4551                 .fc_ignore_dev_down = true,
4552         };
4553         struct fib6_info *f6i;
4554
4555         if (anycast) {
4556                 cfg.fc_type = RTN_ANYCAST;
4557                 cfg.fc_flags |= RTF_ANYCAST;
4558         } else {
4559                 cfg.fc_type = RTN_LOCAL;
4560                 cfg.fc_flags |= RTF_LOCAL;
4561         }
4562
4563         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4564         if (!IS_ERR(f6i)) {
4565                 f6i->dst_nocount = true;
4566
4567                 if (!anycast &&
4568                     (net->ipv6.devconf_all->disable_policy ||
4569                      idev->cnf.disable_policy))
4570                         f6i->dst_nopolicy = true;
4571         }
4572
4573         return f6i;
4574 }
4575
4576 /* remove deleted ip from prefsrc entries */
4577 struct arg_dev_net_ip {
4578         struct net_device *dev;
4579         struct net *net;
4580         struct in6_addr *addr;
4581 };
4582
4583 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4584 {
4585         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4586         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4587         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4588
4589         if (!rt->nh &&
4590             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4591             rt != net->ipv6.fib6_null_entry &&
4592             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4593                 spin_lock_bh(&rt6_exception_lock);
4594                 /* remove prefsrc entry */
4595                 rt->fib6_prefsrc.plen = 0;
4596                 spin_unlock_bh(&rt6_exception_lock);
4597         }
4598         return 0;
4599 }
4600
4601 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4602 {
4603         struct net *net = dev_net(ifp->idev->dev);
4604         struct arg_dev_net_ip adni = {
4605                 .dev = ifp->idev->dev,
4606                 .net = net,
4607                 .addr = &ifp->addr,
4608         };
4609         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4610 }
4611
4612 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4613
4614 /* Remove routers and update dst entries when gateway turn into host. */
4615 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4616 {
4617         struct in6_addr *gateway = (struct in6_addr *)arg;
4618         struct fib6_nh *nh;
4619
4620         /* RA routes do not use nexthops */
4621         if (rt->nh)
4622                 return 0;
4623
4624         nh = rt->fib6_nh;
4625         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4626             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4627                 return -1;
4628
4629         /* Further clean up cached routes in exception table.
4630          * This is needed because cached route may have a different
4631          * gateway than its 'parent' in the case of an ip redirect.
4632          */
4633         fib6_nh_exceptions_clean_tohost(nh, gateway);
4634
4635         return 0;
4636 }
4637
4638 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4639 {
4640         fib6_clean_all(net, fib6_clean_tohost, gateway);
4641 }
4642
4643 struct arg_netdev_event {
4644         const struct net_device *dev;
4645         union {
4646                 unsigned char nh_flags;
4647                 unsigned long event;
4648         };
4649 };
4650
4651 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4652 {
4653         struct fib6_info *iter;
4654         struct fib6_node *fn;
4655
4656         fn = rcu_dereference_protected(rt->fib6_node,
4657                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4658         iter = rcu_dereference_protected(fn->leaf,
4659                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4660         while (iter) {
4661                 if (iter->fib6_metric == rt->fib6_metric &&
4662                     rt6_qualify_for_ecmp(iter))
4663                         return iter;
4664                 iter = rcu_dereference_protected(iter->fib6_next,
4665                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4666         }
4667
4668         return NULL;
4669 }
4670
4671 /* only called for fib entries with builtin fib6_nh */
4672 static bool rt6_is_dead(const struct fib6_info *rt)
4673 {
4674         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4675             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4676              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4677                 return true;
4678
4679         return false;
4680 }
4681
4682 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4683 {
4684         struct fib6_info *iter;
4685         int total = 0;
4686
4687         if (!rt6_is_dead(rt))
4688                 total += rt->fib6_nh->fib_nh_weight;
4689
4690         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4691                 if (!rt6_is_dead(iter))
4692                         total += iter->fib6_nh->fib_nh_weight;
4693         }
4694
4695         return total;
4696 }
4697
4698 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4699 {
4700         int upper_bound = -1;
4701
4702         if (!rt6_is_dead(rt)) {
4703                 *weight += rt->fib6_nh->fib_nh_weight;
4704                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4705                                                     total) - 1;
4706         }
4707         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4708 }
4709
4710 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4711 {
4712         struct fib6_info *iter;
4713         int weight = 0;
4714
4715         rt6_upper_bound_set(rt, &weight, total);
4716
4717         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4718                 rt6_upper_bound_set(iter, &weight, total);
4719 }
4720
4721 void rt6_multipath_rebalance(struct fib6_info *rt)
4722 {
4723         struct fib6_info *first;
4724         int total;
4725
4726         /* In case the entire multipath route was marked for flushing,
4727          * then there is no need to rebalance upon the removal of every
4728          * sibling route.
4729          */
4730         if (!rt->fib6_nsiblings || rt->should_flush)
4731                 return;
4732
4733         /* During lookup routes are evaluated in order, so we need to
4734          * make sure upper bounds are assigned from the first sibling
4735          * onwards.
4736          */
4737         first = rt6_multipath_first_sibling(rt);
4738         if (WARN_ON_ONCE(!first))
4739                 return;
4740
4741         total = rt6_multipath_total_weight(first);
4742         rt6_multipath_upper_bound_set(first, total);
4743 }
4744
4745 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4746 {
4747         const struct arg_netdev_event *arg = p_arg;
4748         struct net *net = dev_net(arg->dev);
4749
4750         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4751             rt->fib6_nh->fib_nh_dev == arg->dev) {
4752                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4753                 fib6_update_sernum_upto_root(net, rt);
4754                 rt6_multipath_rebalance(rt);
4755         }
4756
4757         return 0;
4758 }
4759
4760 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4761 {
4762         struct arg_netdev_event arg = {
4763                 .dev = dev,
4764                 {
4765                         .nh_flags = nh_flags,
4766                 },
4767         };
4768
4769         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4770                 arg.nh_flags |= RTNH_F_LINKDOWN;
4771
4772         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4773 }
4774
4775 /* only called for fib entries with inline fib6_nh */
4776 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4777                                    const struct net_device *dev)
4778 {
4779         struct fib6_info *iter;
4780
4781         if (rt->fib6_nh->fib_nh_dev == dev)
4782                 return true;
4783         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4784                 if (iter->fib6_nh->fib_nh_dev == dev)
4785                         return true;
4786
4787         return false;
4788 }
4789
4790 static void rt6_multipath_flush(struct fib6_info *rt)
4791 {
4792         struct fib6_info *iter;
4793
4794         rt->should_flush = 1;
4795         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4796                 iter->should_flush = 1;
4797 }
4798
4799 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4800                                              const struct net_device *down_dev)
4801 {
4802         struct fib6_info *iter;
4803         unsigned int dead = 0;
4804
4805         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4806             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4807                 dead++;
4808         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4809                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4810                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4811                         dead++;
4812
4813         return dead;
4814 }
4815
4816 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4817                                        const struct net_device *dev,
4818                                        unsigned char nh_flags)
4819 {
4820         struct fib6_info *iter;
4821
4822         if (rt->fib6_nh->fib_nh_dev == dev)
4823                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4824         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4825                 if (iter->fib6_nh->fib_nh_dev == dev)
4826                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4827 }
4828
4829 /* called with write lock held for table with rt */
4830 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4831 {
4832         const struct arg_netdev_event *arg = p_arg;
4833         const struct net_device *dev = arg->dev;
4834         struct net *net = dev_net(dev);
4835
4836         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4837                 return 0;
4838
4839         switch (arg->event) {
4840         case NETDEV_UNREGISTER:
4841                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4842         case NETDEV_DOWN:
4843                 if (rt->should_flush)
4844                         return -1;
4845                 if (!rt->fib6_nsiblings)
4846                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4847                 if (rt6_multipath_uses_dev(rt, dev)) {
4848                         unsigned int count;
4849
4850                         count = rt6_multipath_dead_count(rt, dev);
4851                         if (rt->fib6_nsiblings + 1 == count) {
4852                                 rt6_multipath_flush(rt);
4853                                 return -1;
4854                         }
4855                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4856                                                    RTNH_F_LINKDOWN);
4857                         fib6_update_sernum(net, rt);
4858                         rt6_multipath_rebalance(rt);
4859                 }
4860                 return -2;
4861         case NETDEV_CHANGE:
4862                 if (rt->fib6_nh->fib_nh_dev != dev ||
4863                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4864                         break;
4865                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4866                 rt6_multipath_rebalance(rt);
4867                 break;
4868         }
4869
4870         return 0;
4871 }
4872
4873 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4874 {
4875         struct arg_netdev_event arg = {
4876                 .dev = dev,
4877                 {
4878                         .event = event,
4879                 },
4880         };
4881         struct net *net = dev_net(dev);
4882
4883         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4884                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4885         else
4886                 fib6_clean_all(net, fib6_ifdown, &arg);
4887 }
4888
4889 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4890 {
4891         rt6_sync_down_dev(dev, event);
4892         rt6_uncached_list_flush_dev(dev);
4893         neigh_ifdown(&nd_tbl, dev);
4894 }
4895
4896 struct rt6_mtu_change_arg {
4897         struct net_device *dev;
4898         unsigned int mtu;
4899         struct fib6_info *f6i;
4900 };
4901
4902 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4903 {
4904         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4905         struct fib6_info *f6i = arg->f6i;
4906
4907         /* For administrative MTU increase, there is no way to discover
4908          * IPv6 PMTU increase, so PMTU increase should be updated here.
4909          * Since RFC 1981 doesn't include administrative MTU increase
4910          * update PMTU increase is a MUST. (i.e. jumbo frame)
4911          */
4912         if (nh->fib_nh_dev == arg->dev) {
4913                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4914                 u32 mtu = f6i->fib6_pmtu;
4915
4916                 if (mtu >= arg->mtu ||
4917                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4918                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4919
4920                 spin_lock_bh(&rt6_exception_lock);
4921                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4922                 spin_unlock_bh(&rt6_exception_lock);
4923         }
4924
4925         return 0;
4926 }
4927
4928 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4929 {
4930         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4931         struct inet6_dev *idev;
4932
4933         /* In IPv6 pmtu discovery is not optional,
4934            so that RTAX_MTU lock cannot disable it.
4935            We still use this lock to block changes
4936            caused by addrconf/ndisc.
4937         */
4938
4939         idev = __in6_dev_get(arg->dev);
4940         if (!idev)
4941                 return 0;
4942
4943         if (fib6_metric_locked(f6i, RTAX_MTU))
4944                 return 0;
4945
4946         arg->f6i = f6i;
4947         if (f6i->nh) {
4948                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4949                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4950                                                 arg);
4951         }
4952
4953         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4954 }
4955
4956 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4957 {
4958         struct rt6_mtu_change_arg arg = {
4959                 .dev = dev,
4960                 .mtu = mtu,
4961         };
4962
4963         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4964 }
4965
4966 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4967         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4968         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4969         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4970         [RTA_OIF]               = { .type = NLA_U32 },
4971         [RTA_IIF]               = { .type = NLA_U32 },
4972         [RTA_PRIORITY]          = { .type = NLA_U32 },
4973         [RTA_METRICS]           = { .type = NLA_NESTED },
4974         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4975         [RTA_PREF]              = { .type = NLA_U8 },
4976         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4977         [RTA_ENCAP]             = { .type = NLA_NESTED },
4978         [RTA_EXPIRES]           = { .type = NLA_U32 },
4979         [RTA_UID]               = { .type = NLA_U32 },
4980         [RTA_MARK]              = { .type = NLA_U32 },
4981         [RTA_TABLE]             = { .type = NLA_U32 },
4982         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4983         [RTA_SPORT]             = { .type = NLA_U16 },
4984         [RTA_DPORT]             = { .type = NLA_U16 },
4985         [RTA_NH_ID]             = { .type = NLA_U32 },
4986 };
4987
4988 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4989                               struct fib6_config *cfg,
4990                               struct netlink_ext_ack *extack)
4991 {
4992         struct rtmsg *rtm;
4993         struct nlattr *tb[RTA_MAX+1];
4994         unsigned int pref;
4995         int err;
4996
4997         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4998                                      rtm_ipv6_policy, extack);
4999         if (err < 0)
5000                 goto errout;
5001
5002         err = -EINVAL;
5003         rtm = nlmsg_data(nlh);
5004
5005         if (rtm->rtm_tos) {
5006                 NL_SET_ERR_MSG(extack,
5007                                "Invalid dsfield (tos): option not available for IPv6");
5008                 goto errout;
5009         }
5010
5011         *cfg = (struct fib6_config){
5012                 .fc_table = rtm->rtm_table,
5013                 .fc_dst_len = rtm->rtm_dst_len,
5014                 .fc_src_len = rtm->rtm_src_len,
5015                 .fc_flags = RTF_UP,
5016                 .fc_protocol = rtm->rtm_protocol,
5017                 .fc_type = rtm->rtm_type,
5018
5019                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5020                 .fc_nlinfo.nlh = nlh,
5021                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5022         };
5023
5024         if (rtm->rtm_type == RTN_UNREACHABLE ||
5025             rtm->rtm_type == RTN_BLACKHOLE ||
5026             rtm->rtm_type == RTN_PROHIBIT ||
5027             rtm->rtm_type == RTN_THROW)
5028                 cfg->fc_flags |= RTF_REJECT;
5029
5030         if (rtm->rtm_type == RTN_LOCAL)
5031                 cfg->fc_flags |= RTF_LOCAL;
5032
5033         if (rtm->rtm_flags & RTM_F_CLONED)
5034                 cfg->fc_flags |= RTF_CACHE;
5035
5036         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5037
5038         if (tb[RTA_NH_ID]) {
5039                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5040                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5041                         NL_SET_ERR_MSG(extack,
5042                                        "Nexthop specification and nexthop id are mutually exclusive");
5043                         goto errout;
5044                 }
5045                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5046         }
5047
5048         if (tb[RTA_GATEWAY]) {
5049                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5050                 cfg->fc_flags |= RTF_GATEWAY;
5051         }
5052         if (tb[RTA_VIA]) {
5053                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5054                 goto errout;
5055         }
5056
5057         if (tb[RTA_DST]) {
5058                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5059
5060                 if (nla_len(tb[RTA_DST]) < plen)
5061                         goto errout;
5062
5063                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5064         }
5065
5066         if (tb[RTA_SRC]) {
5067                 int plen = (rtm->rtm_src_len + 7) >> 3;
5068
5069                 if (nla_len(tb[RTA_SRC]) < plen)
5070                         goto errout;
5071
5072                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5073         }
5074
5075         if (tb[RTA_PREFSRC])
5076                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5077
5078         if (tb[RTA_OIF])
5079                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5080
5081         if (tb[RTA_PRIORITY])
5082                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5083
5084         if (tb[RTA_METRICS]) {
5085                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5086                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5087         }
5088
5089         if (tb[RTA_TABLE])
5090                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5091
5092         if (tb[RTA_MULTIPATH]) {
5093                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5094                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5095
5096                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5097                                                      cfg->fc_mp_len, extack);
5098                 if (err < 0)
5099                         goto errout;
5100         }
5101
5102         if (tb[RTA_PREF]) {
5103                 pref = nla_get_u8(tb[RTA_PREF]);
5104                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5105                     pref != ICMPV6_ROUTER_PREF_HIGH)
5106                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5107                 cfg->fc_flags |= RTF_PREF(pref);
5108         }
5109
5110         if (tb[RTA_ENCAP])
5111                 cfg->fc_encap = tb[RTA_ENCAP];
5112
5113         if (tb[RTA_ENCAP_TYPE]) {
5114                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5115
5116                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5117                 if (err < 0)
5118                         goto errout;
5119         }
5120
5121         if (tb[RTA_EXPIRES]) {
5122                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5123
5124                 if (addrconf_finite_timeout(timeout)) {
5125                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5126                         cfg->fc_flags |= RTF_EXPIRES;
5127                 }
5128         }
5129
5130         err = 0;
5131 errout:
5132         return err;
5133 }
5134
5135 struct rt6_nh {
5136         struct fib6_info *fib6_info;
5137         struct fib6_config r_cfg;
5138         struct list_head next;
5139 };
5140
5141 static int ip6_route_info_append(struct net *net,
5142                                  struct list_head *rt6_nh_list,
5143                                  struct fib6_info *rt,
5144                                  struct fib6_config *r_cfg)
5145 {
5146         struct rt6_nh *nh;
5147         int err = -EEXIST;
5148
5149         list_for_each_entry(nh, rt6_nh_list, next) {
5150                 /* check if fib6_info already exists */
5151                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5152                         return err;
5153         }
5154
5155         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5156         if (!nh)
5157                 return -ENOMEM;
5158         nh->fib6_info = rt;
5159         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5160         list_add_tail(&nh->next, rt6_nh_list);
5161
5162         return 0;
5163 }
5164
5165 static void ip6_route_mpath_notify(struct fib6_info *rt,
5166                                    struct fib6_info *rt_last,
5167                                    struct nl_info *info,
5168                                    __u16 nlflags)
5169 {
5170         /* if this is an APPEND route, then rt points to the first route
5171          * inserted and rt_last points to last route inserted. Userspace
5172          * wants a consistent dump of the route which starts at the first
5173          * nexthop. Since sibling routes are always added at the end of
5174          * the list, find the first sibling of the last route appended
5175          */
5176         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5177                 rt = list_first_entry(&rt_last->fib6_siblings,
5178                                       struct fib6_info,
5179                                       fib6_siblings);
5180         }
5181
5182         if (rt)
5183                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5184 }
5185
5186 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5187 {
5188         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5189         bool should_notify = false;
5190         struct fib6_info *leaf;
5191         struct fib6_node *fn;
5192
5193         rcu_read_lock();
5194         fn = rcu_dereference(rt->fib6_node);
5195         if (!fn)
5196                 goto out;
5197
5198         leaf = rcu_dereference(fn->leaf);
5199         if (!leaf)
5200                 goto out;
5201
5202         if (rt == leaf ||
5203             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5204              rt6_qualify_for_ecmp(leaf)))
5205                 should_notify = true;
5206 out:
5207         rcu_read_unlock();
5208
5209         return should_notify;
5210 }
5211
5212 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5213                              struct netlink_ext_ack *extack)
5214 {
5215         if (nla_len(nla) < sizeof(*gw)) {
5216                 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5217                 return -EINVAL;
5218         }
5219
5220         *gw = nla_get_in6_addr(nla);
5221
5222         return 0;
5223 }
5224
5225 static int ip6_route_multipath_add(struct fib6_config *cfg,
5226                                    struct netlink_ext_ack *extack)
5227 {
5228         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5229         struct nl_info *info = &cfg->fc_nlinfo;
5230         struct fib6_config r_cfg;
5231         struct rtnexthop *rtnh;
5232         struct fib6_info *rt;
5233         struct rt6_nh *err_nh;
5234         struct rt6_nh *nh, *nh_safe;
5235         __u16 nlflags;
5236         int remaining;
5237         int attrlen;
5238         int err = 1;
5239         int nhn = 0;
5240         int replace = (cfg->fc_nlinfo.nlh &&
5241                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5242         LIST_HEAD(rt6_nh_list);
5243
5244         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5245         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5246                 nlflags |= NLM_F_APPEND;
5247
5248         remaining = cfg->fc_mp_len;
5249         rtnh = (struct rtnexthop *)cfg->fc_mp;
5250
5251         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5252          * fib6_info structs per nexthop
5253          */
5254         while (rtnh_ok(rtnh, remaining)) {
5255                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5256                 if (rtnh->rtnh_ifindex)
5257                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5258
5259                 attrlen = rtnh_attrlen(rtnh);
5260                 if (attrlen > 0) {
5261                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5262
5263                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5264                         if (nla) {
5265                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5266                                                         extack);
5267                                 if (err)
5268                                         goto cleanup;
5269
5270                                 r_cfg.fc_flags |= RTF_GATEWAY;
5271                         }
5272                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5273
5274                         /* RTA_ENCAP_TYPE length checked in
5275                          * lwtunnel_valid_encap_type_attr
5276                          */
5277                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5278                         if (nla)
5279                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5280                 }
5281
5282                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5283                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5284                 if (IS_ERR(rt)) {
5285                         err = PTR_ERR(rt);
5286                         rt = NULL;
5287                         goto cleanup;
5288                 }
5289                 if (!rt6_qualify_for_ecmp(rt)) {
5290                         err = -EINVAL;
5291                         NL_SET_ERR_MSG(extack,
5292                                        "Device only routes can not be added for IPv6 using the multipath API.");
5293                         fib6_info_release(rt);
5294                         goto cleanup;
5295                 }
5296
5297                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5298
5299                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5300                                             rt, &r_cfg);
5301                 if (err) {
5302                         fib6_info_release(rt);
5303                         goto cleanup;
5304                 }
5305
5306                 rtnh = rtnh_next(rtnh, &remaining);
5307         }
5308
5309         if (list_empty(&rt6_nh_list)) {
5310                 NL_SET_ERR_MSG(extack,
5311                                "Invalid nexthop configuration - no valid nexthops");
5312                 return -EINVAL;
5313         }
5314
5315         /* for add and replace send one notification with all nexthops.
5316          * Skip the notification in fib6_add_rt2node and send one with
5317          * the full route when done
5318          */
5319         info->skip_notify = 1;
5320
5321         /* For add and replace, send one notification with all nexthops. For
5322          * append, send one notification with all appended nexthops.
5323          */
5324         info->skip_notify_kernel = 1;
5325
5326         err_nh = NULL;
5327         list_for_each_entry(nh, &rt6_nh_list, next) {
5328                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5329                 fib6_info_release(nh->fib6_info);
5330
5331                 if (!err) {
5332                         /* save reference to last route successfully inserted */
5333                         rt_last = nh->fib6_info;
5334
5335                         /* save reference to first route for notification */
5336                         if (!rt_notif)
5337                                 rt_notif = nh->fib6_info;
5338                 }
5339
5340                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5341                 nh->fib6_info = NULL;
5342                 if (err) {
5343                         if (replace && nhn)
5344                                 NL_SET_ERR_MSG_MOD(extack,
5345                                                    "multipath route replace failed (check consistency of installed routes)");
5346                         err_nh = nh;
5347                         goto add_errout;
5348                 }
5349
5350                 /* Because each route is added like a single route we remove
5351                  * these flags after the first nexthop: if there is a collision,
5352                  * we have already failed to add the first nexthop:
5353                  * fib6_add_rt2node() has rejected it; when replacing, old
5354                  * nexthops have been replaced by first new, the rest should
5355                  * be added to it.
5356                  */
5357                 if (cfg->fc_nlinfo.nlh) {
5358                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5359                                                              NLM_F_REPLACE);
5360                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5361                 }
5362                 nhn++;
5363         }
5364
5365         /* An in-kernel notification should only be sent in case the new
5366          * multipath route is added as the first route in the node, or if
5367          * it was appended to it. We pass 'rt_notif' since it is the first
5368          * sibling and might allow us to skip some checks in the replace case.
5369          */
5370         if (ip6_route_mpath_should_notify(rt_notif)) {
5371                 enum fib_event_type fib_event;
5372
5373                 if (rt_notif->fib6_nsiblings != nhn - 1)
5374                         fib_event = FIB_EVENT_ENTRY_APPEND;
5375                 else
5376                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5377
5378                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5379                                                           fib_event, rt_notif,
5380                                                           nhn - 1, extack);
5381                 if (err) {
5382                         /* Delete all the siblings that were just added */
5383                         err_nh = NULL;
5384                         goto add_errout;
5385                 }
5386         }
5387
5388         /* success ... tell user about new route */
5389         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5390         goto cleanup;
5391
5392 add_errout:
5393         /* send notification for routes that were added so that
5394          * the delete notifications sent by ip6_route_del are
5395          * coherent
5396          */
5397         if (rt_notif)
5398                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5399
5400         /* Delete routes that were already added */
5401         list_for_each_entry(nh, &rt6_nh_list, next) {
5402                 if (err_nh == nh)
5403                         break;
5404                 ip6_route_del(&nh->r_cfg, extack);
5405         }
5406
5407 cleanup:
5408         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5409                 if (nh->fib6_info)
5410                         fib6_info_release(nh->fib6_info);
5411                 list_del(&nh->next);
5412                 kfree(nh);
5413         }
5414
5415         return err;
5416 }
5417
5418 static int ip6_route_multipath_del(struct fib6_config *cfg,
5419                                    struct netlink_ext_ack *extack)
5420 {
5421         struct fib6_config r_cfg;
5422         struct rtnexthop *rtnh;
5423         int last_err = 0;
5424         int remaining;
5425         int attrlen;
5426         int err;
5427
5428         remaining = cfg->fc_mp_len;
5429         rtnh = (struct rtnexthop *)cfg->fc_mp;
5430
5431         /* Parse a Multipath Entry */
5432         while (rtnh_ok(rtnh, remaining)) {
5433                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5434                 if (rtnh->rtnh_ifindex)
5435                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5436
5437                 attrlen = rtnh_attrlen(rtnh);
5438                 if (attrlen > 0) {
5439                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5440
5441                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5442                         if (nla) {
5443                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5444                                                         extack);
5445                                 if (err) {
5446                                         last_err = err;
5447                                         goto next_rtnh;
5448                                 }
5449
5450                                 r_cfg.fc_flags |= RTF_GATEWAY;
5451                         }
5452                 }
5453                 err = ip6_route_del(&r_cfg, extack);
5454                 if (err)
5455                         last_err = err;
5456
5457 next_rtnh:
5458                 rtnh = rtnh_next(rtnh, &remaining);
5459         }
5460
5461         return last_err;
5462 }
5463
5464 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5465                               struct netlink_ext_ack *extack)
5466 {
5467         struct fib6_config cfg;
5468         int err;
5469
5470         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5471         if (err < 0)
5472                 return err;
5473
5474         if (cfg.fc_nh_id &&
5475             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5476                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5477                 return -EINVAL;
5478         }
5479
5480         if (cfg.fc_mp)
5481                 return ip6_route_multipath_del(&cfg, extack);
5482         else {
5483                 cfg.fc_delete_all_nh = 1;
5484                 return ip6_route_del(&cfg, extack);
5485         }
5486 }
5487
5488 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5489                               struct netlink_ext_ack *extack)
5490 {
5491         struct fib6_config cfg;
5492         int err;
5493
5494         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5495         if (err < 0)
5496                 return err;
5497
5498         if (cfg.fc_metric == 0)
5499                 cfg.fc_metric = IP6_RT_PRIO_USER;
5500
5501         if (cfg.fc_mp)
5502                 return ip6_route_multipath_add(&cfg, extack);
5503         else
5504                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5505 }
5506
5507 /* add the overhead of this fib6_nh to nexthop_len */
5508 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5509 {
5510         int *nexthop_len = arg;
5511
5512         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5513                      + NLA_ALIGN(sizeof(struct rtnexthop))
5514                      + nla_total_size(16); /* RTA_GATEWAY */
5515
5516         if (nh->fib_nh_lws) {
5517                 /* RTA_ENCAP_TYPE */
5518                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5519                 /* RTA_ENCAP */
5520                 *nexthop_len += nla_total_size(2);
5521         }
5522
5523         return 0;
5524 }
5525
5526 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5527 {
5528         int nexthop_len;
5529
5530         if (f6i->nh) {
5531                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5532                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5533                                          &nexthop_len);
5534         } else {
5535                 struct fib6_info *sibling, *next_sibling;
5536                 struct fib6_nh *nh = f6i->fib6_nh;
5537
5538                 nexthop_len = 0;
5539                 if (f6i->fib6_nsiblings) {
5540                         rt6_nh_nlmsg_size(nh, &nexthop_len);
5541
5542                         list_for_each_entry_safe(sibling, next_sibling,
5543                                                  &f6i->fib6_siblings, fib6_siblings) {
5544                                 rt6_nh_nlmsg_size(sibling->fib6_nh, &nexthop_len);
5545                         }
5546                 }
5547                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5548         }
5549
5550         return NLMSG_ALIGN(sizeof(struct rtmsg))
5551                + nla_total_size(16) /* RTA_SRC */
5552                + nla_total_size(16) /* RTA_DST */
5553                + nla_total_size(16) /* RTA_GATEWAY */
5554                + nla_total_size(16) /* RTA_PREFSRC */
5555                + nla_total_size(4) /* RTA_TABLE */
5556                + nla_total_size(4) /* RTA_IIF */
5557                + nla_total_size(4) /* RTA_OIF */
5558                + nla_total_size(4) /* RTA_PRIORITY */
5559                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5560                + nla_total_size(sizeof(struct rta_cacheinfo))
5561                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5562                + nla_total_size(1) /* RTA_PREF */
5563                + nexthop_len;
5564 }
5565
5566 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5567                                  unsigned char *flags)
5568 {
5569         if (nexthop_is_multipath(nh)) {
5570                 struct nlattr *mp;
5571
5572                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5573                 if (!mp)
5574                         goto nla_put_failure;
5575
5576                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5577                         goto nla_put_failure;
5578
5579                 nla_nest_end(skb, mp);
5580         } else {
5581                 struct fib6_nh *fib6_nh;
5582
5583                 fib6_nh = nexthop_fib6_nh(nh);
5584                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5585                                      flags, false) < 0)
5586                         goto nla_put_failure;
5587         }
5588
5589         return 0;
5590
5591 nla_put_failure:
5592         return -EMSGSIZE;
5593 }
5594
5595 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5596                          struct fib6_info *rt, struct dst_entry *dst,
5597                          struct in6_addr *dest, struct in6_addr *src,
5598                          int iif, int type, u32 portid, u32 seq,
5599                          unsigned int flags)
5600 {
5601         struct rt6_info *rt6 = (struct rt6_info *)dst;
5602         struct rt6key *rt6_dst, *rt6_src;
5603         u32 *pmetrics, table, rt6_flags;
5604         unsigned char nh_flags = 0;
5605         struct nlmsghdr *nlh;
5606         struct rtmsg *rtm;
5607         long expires = 0;
5608
5609         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5610         if (!nlh)
5611                 return -EMSGSIZE;
5612
5613         if (rt6) {
5614                 rt6_dst = &rt6->rt6i_dst;
5615                 rt6_src = &rt6->rt6i_src;
5616                 rt6_flags = rt6->rt6i_flags;
5617         } else {
5618                 rt6_dst = &rt->fib6_dst;
5619                 rt6_src = &rt->fib6_src;
5620                 rt6_flags = rt->fib6_flags;
5621         }
5622
5623         rtm = nlmsg_data(nlh);
5624         rtm->rtm_family = AF_INET6;
5625         rtm->rtm_dst_len = rt6_dst->plen;
5626         rtm->rtm_src_len = rt6_src->plen;
5627         rtm->rtm_tos = 0;
5628         if (rt->fib6_table)
5629                 table = rt->fib6_table->tb6_id;
5630         else
5631                 table = RT6_TABLE_UNSPEC;
5632         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5633         if (nla_put_u32(skb, RTA_TABLE, table))
5634                 goto nla_put_failure;
5635
5636         rtm->rtm_type = rt->fib6_type;
5637         rtm->rtm_flags = 0;
5638         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5639         rtm->rtm_protocol = rt->fib6_protocol;
5640
5641         if (rt6_flags & RTF_CACHE)
5642                 rtm->rtm_flags |= RTM_F_CLONED;
5643
5644         if (dest) {
5645                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5646                         goto nla_put_failure;
5647                 rtm->rtm_dst_len = 128;
5648         } else if (rtm->rtm_dst_len)
5649                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5650                         goto nla_put_failure;
5651 #ifdef CONFIG_IPV6_SUBTREES
5652         if (src) {
5653                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5654                         goto nla_put_failure;
5655                 rtm->rtm_src_len = 128;
5656         } else if (rtm->rtm_src_len &&
5657                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5658                 goto nla_put_failure;
5659 #endif
5660         if (iif) {
5661 #ifdef CONFIG_IPV6_MROUTE
5662                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5663                         int err = ip6mr_get_route(net, skb, rtm, portid);
5664
5665                         if (err == 0)
5666                                 return 0;
5667                         if (err < 0)
5668                                 goto nla_put_failure;
5669                 } else
5670 #endif
5671                         if (nla_put_u32(skb, RTA_IIF, iif))
5672                                 goto nla_put_failure;
5673         } else if (dest) {
5674                 struct in6_addr saddr_buf;
5675                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5676                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5677                         goto nla_put_failure;
5678         }
5679
5680         if (rt->fib6_prefsrc.plen) {
5681                 struct in6_addr saddr_buf;
5682                 saddr_buf = rt->fib6_prefsrc.addr;
5683                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5684                         goto nla_put_failure;
5685         }
5686
5687         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5688         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5689                 goto nla_put_failure;
5690
5691         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5692                 goto nla_put_failure;
5693
5694         /* For multipath routes, walk the siblings list and add
5695          * each as a nexthop within RTA_MULTIPATH.
5696          */
5697         if (rt6) {
5698                 if (rt6_flags & RTF_GATEWAY &&
5699                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5700                         goto nla_put_failure;
5701
5702                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5703                         goto nla_put_failure;
5704
5705                 if (dst->lwtstate &&
5706                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5707                         goto nla_put_failure;
5708         } else if (rt->fib6_nsiblings) {
5709                 struct fib6_info *sibling, *next_sibling;
5710                 struct nlattr *mp;
5711
5712                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5713                 if (!mp)
5714                         goto nla_put_failure;
5715
5716                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5717                                     rt->fib6_nh->fib_nh_weight, AF_INET6,
5718                                     0) < 0)
5719                         goto nla_put_failure;
5720
5721                 list_for_each_entry_safe(sibling, next_sibling,
5722                                          &rt->fib6_siblings, fib6_siblings) {
5723                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5724                                             sibling->fib6_nh->fib_nh_weight,
5725                                             AF_INET6, 0) < 0)
5726                                 goto nla_put_failure;
5727                 }
5728
5729                 nla_nest_end(skb, mp);
5730         } else if (rt->nh) {
5731                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5732                         goto nla_put_failure;
5733
5734                 if (nexthop_is_blackhole(rt->nh))
5735                         rtm->rtm_type = RTN_BLACKHOLE;
5736
5737                 if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5738                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5739                         goto nla_put_failure;
5740
5741                 rtm->rtm_flags |= nh_flags;
5742         } else {
5743                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5744                                      &nh_flags, false) < 0)
5745                         goto nla_put_failure;
5746
5747                 rtm->rtm_flags |= nh_flags;
5748         }
5749
5750         if (rt6_flags & RTF_EXPIRES) {
5751                 expires = dst ? dst->expires : rt->expires;
5752                 expires -= jiffies;
5753         }
5754
5755         if (!dst) {
5756                 if (READ_ONCE(rt->offload))
5757                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5758                 if (READ_ONCE(rt->trap))
5759                         rtm->rtm_flags |= RTM_F_TRAP;
5760                 if (READ_ONCE(rt->offload_failed))
5761                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5762         }
5763
5764         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5765                 goto nla_put_failure;
5766
5767         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5768                 goto nla_put_failure;
5769
5770
5771         nlmsg_end(skb, nlh);
5772         return 0;
5773
5774 nla_put_failure:
5775         nlmsg_cancel(skb, nlh);
5776         return -EMSGSIZE;
5777 }
5778
5779 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5780 {
5781         const struct net_device *dev = arg;
5782
5783         if (nh->fib_nh_dev == dev)
5784                 return 1;
5785
5786         return 0;
5787 }
5788
5789 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5790                                const struct net_device *dev)
5791 {
5792         if (f6i->nh) {
5793                 struct net_device *_dev = (struct net_device *)dev;
5794
5795                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5796                                                   fib6_info_nh_uses_dev,
5797                                                   _dev);
5798         }
5799
5800         if (f6i->fib6_nh->fib_nh_dev == dev)
5801                 return true;
5802
5803         if (f6i->fib6_nsiblings) {
5804                 struct fib6_info *sibling, *next_sibling;
5805
5806                 list_for_each_entry_safe(sibling, next_sibling,
5807                                          &f6i->fib6_siblings, fib6_siblings) {
5808                         if (sibling->fib6_nh->fib_nh_dev == dev)
5809                                 return true;
5810                 }
5811         }
5812
5813         return false;
5814 }
5815
5816 struct fib6_nh_exception_dump_walker {
5817         struct rt6_rtnl_dump_arg *dump;
5818         struct fib6_info *rt;
5819         unsigned int flags;
5820         unsigned int skip;
5821         unsigned int count;
5822 };
5823
5824 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5825 {
5826         struct fib6_nh_exception_dump_walker *w = arg;
5827         struct rt6_rtnl_dump_arg *dump = w->dump;
5828         struct rt6_exception_bucket *bucket;
5829         struct rt6_exception *rt6_ex;
5830         int i, err;
5831
5832         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5833         if (!bucket)
5834                 return 0;
5835
5836         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5837                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5838                         if (w->skip) {
5839                                 w->skip--;
5840                                 continue;
5841                         }
5842
5843                         /* Expiration of entries doesn't bump sernum, insertion
5844                          * does. Removal is triggered by insertion, so we can
5845                          * rely on the fact that if entries change between two
5846                          * partial dumps, this node is scanned again completely,
5847                          * see rt6_insert_exception() and fib6_dump_table().
5848                          *
5849                          * Count expired entries we go through as handled
5850                          * entries that we'll skip next time, in case of partial
5851                          * node dump. Otherwise, if entries expire meanwhile,
5852                          * we'll skip the wrong amount.
5853                          */
5854                         if (rt6_check_expired(rt6_ex->rt6i)) {
5855                                 w->count++;
5856                                 continue;
5857                         }
5858
5859                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5860                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5861                                             RTM_NEWROUTE,
5862                                             NETLINK_CB(dump->cb->skb).portid,
5863                                             dump->cb->nlh->nlmsg_seq, w->flags);
5864                         if (err)
5865                                 return err;
5866
5867                         w->count++;
5868                 }
5869                 bucket++;
5870         }
5871
5872         return 0;
5873 }
5874
5875 /* Return -1 if done with node, number of handled routes on partial dump */
5876 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5877 {
5878         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5879         struct fib_dump_filter *filter = &arg->filter;
5880         unsigned int flags = NLM_F_MULTI;
5881         struct net *net = arg->net;
5882         int count = 0;
5883
5884         if (rt == net->ipv6.fib6_null_entry)
5885                 return -1;
5886
5887         if ((filter->flags & RTM_F_PREFIX) &&
5888             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5889                 /* success since this is not a prefix route */
5890                 return -1;
5891         }
5892         if (filter->filter_set &&
5893             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5894              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5895              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5896                 return -1;
5897         }
5898
5899         if (filter->filter_set ||
5900             !filter->dump_routes || !filter->dump_exceptions) {
5901                 flags |= NLM_F_DUMP_FILTERED;
5902         }
5903
5904         if (filter->dump_routes) {
5905                 if (skip) {
5906                         skip--;
5907                 } else {
5908                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5909                                           0, RTM_NEWROUTE,
5910                                           NETLINK_CB(arg->cb->skb).portid,
5911                                           arg->cb->nlh->nlmsg_seq, flags)) {
5912                                 return 0;
5913                         }
5914                         count++;
5915                 }
5916         }
5917
5918         if (filter->dump_exceptions) {
5919                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5920                                                            .rt = rt,
5921                                                            .flags = flags,
5922                                                            .skip = skip,
5923                                                            .count = 0 };
5924                 int err;
5925
5926                 rcu_read_lock();
5927                 if (rt->nh) {
5928                         err = nexthop_for_each_fib6_nh(rt->nh,
5929                                                        rt6_nh_dump_exceptions,
5930                                                        &w);
5931                 } else {
5932                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5933                 }
5934                 rcu_read_unlock();
5935
5936                 if (err)
5937                         return count + w.count;
5938         }
5939
5940         return -1;
5941 }
5942
5943 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5944                                         const struct nlmsghdr *nlh,
5945                                         struct nlattr **tb,
5946                                         struct netlink_ext_ack *extack)
5947 {
5948         struct rtmsg *rtm;
5949         int i, err;
5950
5951         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5952                 NL_SET_ERR_MSG_MOD(extack,
5953                                    "Invalid header for get route request");
5954                 return -EINVAL;
5955         }
5956
5957         if (!netlink_strict_get_check(skb))
5958                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5959                                               rtm_ipv6_policy, extack);
5960
5961         rtm = nlmsg_data(nlh);
5962         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5963             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5964             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5965             rtm->rtm_type) {
5966                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5967                 return -EINVAL;
5968         }
5969         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5970                 NL_SET_ERR_MSG_MOD(extack,
5971                                    "Invalid flags for get route request");
5972                 return -EINVAL;
5973         }
5974
5975         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5976                                             rtm_ipv6_policy, extack);
5977         if (err)
5978                 return err;
5979
5980         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5981             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5982                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5983                 return -EINVAL;
5984         }
5985
5986         for (i = 0; i <= RTA_MAX; i++) {
5987                 if (!tb[i])
5988                         continue;
5989
5990                 switch (i) {
5991                 case RTA_SRC:
5992                 case RTA_DST:
5993                 case RTA_IIF:
5994                 case RTA_OIF:
5995                 case RTA_MARK:
5996                 case RTA_UID:
5997                 case RTA_SPORT:
5998                 case RTA_DPORT:
5999                 case RTA_IP_PROTO:
6000                         break;
6001                 default:
6002                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6003                         return -EINVAL;
6004                 }
6005         }
6006
6007         return 0;
6008 }
6009
6010 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6011                               struct netlink_ext_ack *extack)
6012 {
6013         struct net *net = sock_net(in_skb->sk);
6014         struct nlattr *tb[RTA_MAX+1];
6015         int err, iif = 0, oif = 0;
6016         struct fib6_info *from;
6017         struct dst_entry *dst;
6018         struct rt6_info *rt;
6019         struct sk_buff *skb;
6020         struct rtmsg *rtm;
6021         struct flowi6 fl6 = {};
6022         bool fibmatch;
6023
6024         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6025         if (err < 0)
6026                 goto errout;
6027
6028         err = -EINVAL;
6029         rtm = nlmsg_data(nlh);
6030         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6031         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6032
6033         if (tb[RTA_SRC]) {
6034                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6035                         goto errout;
6036
6037                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6038         }
6039
6040         if (tb[RTA_DST]) {
6041                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6042                         goto errout;
6043
6044                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6045         }
6046
6047         if (tb[RTA_IIF])
6048                 iif = nla_get_u32(tb[RTA_IIF]);
6049
6050         if (tb[RTA_OIF])
6051                 oif = nla_get_u32(tb[RTA_OIF]);
6052
6053         if (tb[RTA_MARK])
6054                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6055
6056         if (tb[RTA_UID])
6057                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6058                                            nla_get_u32(tb[RTA_UID]));
6059         else
6060                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6061
6062         if (tb[RTA_SPORT])
6063                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6064
6065         if (tb[RTA_DPORT])
6066                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6067
6068         if (tb[RTA_IP_PROTO]) {
6069                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6070                                                   &fl6.flowi6_proto, AF_INET6,
6071                                                   extack);
6072                 if (err)
6073                         goto errout;
6074         }
6075
6076         if (iif) {
6077                 struct net_device *dev;
6078                 int flags = 0;
6079
6080                 rcu_read_lock();
6081
6082                 dev = dev_get_by_index_rcu(net, iif);
6083                 if (!dev) {
6084                         rcu_read_unlock();
6085                         err = -ENODEV;
6086                         goto errout;
6087                 }
6088
6089                 fl6.flowi6_iif = iif;
6090
6091                 if (!ipv6_addr_any(&fl6.saddr))
6092                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6093
6094                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6095
6096                 rcu_read_unlock();
6097         } else {
6098                 fl6.flowi6_oif = oif;
6099
6100                 dst = ip6_route_output(net, NULL, &fl6);
6101         }
6102
6103
6104         rt = container_of(dst, struct rt6_info, dst);
6105         if (rt->dst.error) {
6106                 err = rt->dst.error;
6107                 ip6_rt_put(rt);
6108                 goto errout;
6109         }
6110
6111         if (rt == net->ipv6.ip6_null_entry) {
6112                 err = rt->dst.error;
6113                 ip6_rt_put(rt);
6114                 goto errout;
6115         }
6116
6117         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6118         if (!skb) {
6119                 ip6_rt_put(rt);
6120                 err = -ENOBUFS;
6121                 goto errout;
6122         }
6123
6124         skb_dst_set(skb, &rt->dst);
6125
6126         rcu_read_lock();
6127         from = rcu_dereference(rt->from);
6128         if (from) {
6129                 if (fibmatch)
6130                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6131                                             iif, RTM_NEWROUTE,
6132                                             NETLINK_CB(in_skb).portid,
6133                                             nlh->nlmsg_seq, 0);
6134                 else
6135                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6136                                             &fl6.saddr, iif, RTM_NEWROUTE,
6137                                             NETLINK_CB(in_skb).portid,
6138                                             nlh->nlmsg_seq, 0);
6139         } else {
6140                 err = -ENETUNREACH;
6141         }
6142         rcu_read_unlock();
6143
6144         if (err < 0) {
6145                 kfree_skb(skb);
6146                 goto errout;
6147         }
6148
6149         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6150 errout:
6151         return err;
6152 }
6153
6154 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6155                      unsigned int nlm_flags)
6156 {
6157         struct sk_buff *skb;
6158         struct net *net = info->nl_net;
6159         u32 seq;
6160         int err;
6161
6162         err = -ENOBUFS;
6163         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6164
6165         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6166         if (!skb)
6167                 goto errout;
6168
6169         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6170                             event, info->portid, seq, nlm_flags);
6171         if (err < 0) {
6172                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6173                 WARN_ON(err == -EMSGSIZE);
6174                 kfree_skb(skb);
6175                 goto errout;
6176         }
6177         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6178                     info->nlh, gfp_any());
6179         return;
6180 errout:
6181         if (err < 0)
6182                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6183 }
6184
6185 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6186                     struct nl_info *info)
6187 {
6188         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6189         struct sk_buff *skb;
6190         int err = -ENOBUFS;
6191
6192         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6193         if (!skb)
6194                 goto errout;
6195
6196         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6197                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6198         if (err < 0) {
6199                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6200                 WARN_ON(err == -EMSGSIZE);
6201                 kfree_skb(skb);
6202                 goto errout;
6203         }
6204         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6205                     info->nlh, gfp_any());
6206         return;
6207 errout:
6208         if (err < 0)
6209                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6210 }
6211
6212 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6213                             bool offload, bool trap, bool offload_failed)
6214 {
6215         struct sk_buff *skb;
6216         int err;
6217
6218         if (READ_ONCE(f6i->offload) == offload &&
6219             READ_ONCE(f6i->trap) == trap &&
6220             READ_ONCE(f6i->offload_failed) == offload_failed)
6221                 return;
6222
6223         WRITE_ONCE(f6i->offload, offload);
6224         WRITE_ONCE(f6i->trap, trap);
6225
6226         /* 2 means send notifications only if offload_failed was changed. */
6227         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6228             READ_ONCE(f6i->offload_failed) == offload_failed)
6229                 return;
6230
6231         WRITE_ONCE(f6i->offload_failed, offload_failed);
6232
6233         if (!rcu_access_pointer(f6i->fib6_node))
6234                 /* The route was removed from the tree, do not send
6235                  * notification.
6236                  */
6237                 return;
6238
6239         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6240                 return;
6241
6242         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6243         if (!skb) {
6244                 err = -ENOBUFS;
6245                 goto errout;
6246         }
6247
6248         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6249                             0, 0);
6250         if (err < 0) {
6251                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6252                 WARN_ON(err == -EMSGSIZE);
6253                 kfree_skb(skb);
6254                 goto errout;
6255         }
6256
6257         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6258         return;
6259
6260 errout:
6261         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6262 }
6263 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6264
6265 static int ip6_route_dev_notify(struct notifier_block *this,
6266                                 unsigned long event, void *ptr)
6267 {
6268         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6269         struct net *net = dev_net(dev);
6270
6271         if (!(dev->flags & IFF_LOOPBACK))
6272                 return NOTIFY_OK;
6273
6274         if (event == NETDEV_REGISTER) {
6275                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6276                 net->ipv6.ip6_null_entry->dst.dev = dev;
6277                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6278 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6279                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6280                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6281                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6282                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6283 #endif
6284          } else if (event == NETDEV_UNREGISTER &&
6285                     dev->reg_state != NETREG_UNREGISTERED) {
6286                 /* NETDEV_UNREGISTER could be fired for multiple times by
6287                  * netdev_wait_allrefs(). Make sure we only call this once.
6288                  */
6289                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6290 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6291                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6292                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6293 #endif
6294         }
6295
6296         return NOTIFY_OK;
6297 }
6298
6299 /*
6300  *      /proc
6301  */
6302
6303 #ifdef CONFIG_PROC_FS
6304 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6305 {
6306         struct net *net = (struct net *)seq->private;
6307         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6308                    net->ipv6.rt6_stats->fib_nodes,
6309                    net->ipv6.rt6_stats->fib_route_nodes,
6310                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6311                    net->ipv6.rt6_stats->fib_rt_entries,
6312                    net->ipv6.rt6_stats->fib_rt_cache,
6313                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6314                    net->ipv6.rt6_stats->fib_discarded_routes);
6315
6316         return 0;
6317 }
6318 #endif  /* CONFIG_PROC_FS */
6319
6320 #ifdef CONFIG_SYSCTL
6321
6322 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6323                               void *buffer, size_t *lenp, loff_t *ppos)
6324 {
6325         struct net *net;
6326         int delay;
6327         int ret;
6328         if (!write)
6329                 return -EINVAL;
6330
6331         net = (struct net *)ctl->extra1;
6332         delay = net->ipv6.sysctl.flush_delay;
6333         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6334         if (ret)
6335                 return ret;
6336
6337         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6338         return 0;
6339 }
6340
6341 static struct ctl_table ipv6_route_table_template[] = {
6342         {
6343                 .procname       =       "max_size",
6344                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6345                 .maxlen         =       sizeof(int),
6346                 .mode           =       0644,
6347                 .proc_handler   =       proc_dointvec,
6348         },
6349         {
6350                 .procname       =       "gc_thresh",
6351                 .data           =       &ip6_dst_ops_template.gc_thresh,
6352                 .maxlen         =       sizeof(int),
6353                 .mode           =       0644,
6354                 .proc_handler   =       proc_dointvec,
6355         },
6356         {
6357                 .procname       =       "flush",
6358                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6359                 .maxlen         =       sizeof(int),
6360                 .mode           =       0200,
6361                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6362         },
6363         {
6364                 .procname       =       "gc_min_interval",
6365                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6366                 .maxlen         =       sizeof(int),
6367                 .mode           =       0644,
6368                 .proc_handler   =       proc_dointvec_jiffies,
6369         },
6370         {
6371                 .procname       =       "gc_timeout",
6372                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6373                 .maxlen         =       sizeof(int),
6374                 .mode           =       0644,
6375                 .proc_handler   =       proc_dointvec_jiffies,
6376         },
6377         {
6378                 .procname       =       "gc_interval",
6379                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6380                 .maxlen         =       sizeof(int),
6381                 .mode           =       0644,
6382                 .proc_handler   =       proc_dointvec_jiffies,
6383         },
6384         {
6385                 .procname       =       "gc_elasticity",
6386                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6387                 .maxlen         =       sizeof(int),
6388                 .mode           =       0644,
6389                 .proc_handler   =       proc_dointvec,
6390         },
6391         {
6392                 .procname       =       "mtu_expires",
6393                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6394                 .maxlen         =       sizeof(int),
6395                 .mode           =       0644,
6396                 .proc_handler   =       proc_dointvec_jiffies,
6397         },
6398         {
6399                 .procname       =       "min_adv_mss",
6400                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6401                 .maxlen         =       sizeof(int),
6402                 .mode           =       0644,
6403                 .proc_handler   =       proc_dointvec,
6404         },
6405         {
6406                 .procname       =       "gc_min_interval_ms",
6407                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6408                 .maxlen         =       sizeof(int),
6409                 .mode           =       0644,
6410                 .proc_handler   =       proc_dointvec_ms_jiffies,
6411         },
6412         {
6413                 .procname       =       "skip_notify_on_dev_down",
6414                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6415                 .maxlen         =       sizeof(int),
6416                 .mode           =       0644,
6417                 .proc_handler   =       proc_dointvec_minmax,
6418                 .extra1         =       SYSCTL_ZERO,
6419                 .extra2         =       SYSCTL_ONE,
6420         },
6421         { }
6422 };
6423
6424 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6425 {
6426         struct ctl_table *table;
6427
6428         table = kmemdup(ipv6_route_table_template,
6429                         sizeof(ipv6_route_table_template),
6430                         GFP_KERNEL);
6431
6432         if (table) {
6433                 table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6434                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6435                 table[2].data = &net->ipv6.sysctl.flush_delay;
6436                 table[2].extra1 = net;
6437                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6438                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6439                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6440                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6441                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6442                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6443                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6444                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6445
6446                 /* Don't export sysctls to unprivileged users */
6447                 if (net->user_ns != &init_user_ns)
6448                         table[1].procname = NULL;
6449         }
6450
6451         return table;
6452 }
6453 #endif
6454
6455 static int __net_init ip6_route_net_init(struct net *net)
6456 {
6457         int ret = -ENOMEM;
6458
6459         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6460                sizeof(net->ipv6.ip6_dst_ops));
6461
6462         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6463                 goto out_ip6_dst_ops;
6464
6465         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6466         if (!net->ipv6.fib6_null_entry)
6467                 goto out_ip6_dst_entries;
6468         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6469                sizeof(*net->ipv6.fib6_null_entry));
6470
6471         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6472                                            sizeof(*net->ipv6.ip6_null_entry),
6473                                            GFP_KERNEL);
6474         if (!net->ipv6.ip6_null_entry)
6475                 goto out_fib6_null_entry;
6476         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6477         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6478                          ip6_template_metrics, true);
6479         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->dst.rt_uncached);
6480
6481 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6482         net->ipv6.fib6_has_custom_rules = false;
6483         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6484                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6485                                                GFP_KERNEL);
6486         if (!net->ipv6.ip6_prohibit_entry)
6487                 goto out_ip6_null_entry;
6488         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6489         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6490                          ip6_template_metrics, true);
6491         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->dst.rt_uncached);
6492
6493         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6494                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6495                                                GFP_KERNEL);
6496         if (!net->ipv6.ip6_blk_hole_entry)
6497                 goto out_ip6_prohibit_entry;
6498         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6499         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6500                          ip6_template_metrics, true);
6501         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->dst.rt_uncached);
6502 #ifdef CONFIG_IPV6_SUBTREES
6503         net->ipv6.fib6_routes_require_src = 0;
6504 #endif
6505 #endif
6506
6507         net->ipv6.sysctl.flush_delay = 0;
6508         net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
6509         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6510         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6511         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6512         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6513         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6514         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6515         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6516
6517         atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6518
6519         ret = 0;
6520 out:
6521         return ret;
6522
6523 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6524 out_ip6_prohibit_entry:
6525         kfree(net->ipv6.ip6_prohibit_entry);
6526 out_ip6_null_entry:
6527         kfree(net->ipv6.ip6_null_entry);
6528 #endif
6529 out_fib6_null_entry:
6530         kfree(net->ipv6.fib6_null_entry);
6531 out_ip6_dst_entries:
6532         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6533 out_ip6_dst_ops:
6534         goto out;
6535 }
6536
6537 static void __net_exit ip6_route_net_exit(struct net *net)
6538 {
6539         kfree(net->ipv6.fib6_null_entry);
6540         kfree(net->ipv6.ip6_null_entry);
6541 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6542         kfree(net->ipv6.ip6_prohibit_entry);
6543         kfree(net->ipv6.ip6_blk_hole_entry);
6544 #endif
6545         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6546 }
6547
6548 static int __net_init ip6_route_net_init_late(struct net *net)
6549 {
6550 #ifdef CONFIG_PROC_FS
6551         if (!proc_create_net("ipv6_route", 0, net->proc_net,
6552                              &ipv6_route_seq_ops,
6553                              sizeof(struct ipv6_route_iter)))
6554                 return -ENOMEM;
6555
6556         if (!proc_create_net_single("rt6_stats", 0444, net->proc_net,
6557                                     rt6_stats_seq_show, NULL)) {
6558                 remove_proc_entry("ipv6_route", net->proc_net);
6559                 return -ENOMEM;
6560         }
6561 #endif
6562         return 0;
6563 }
6564
6565 static void __net_exit ip6_route_net_exit_late(struct net *net)
6566 {
6567 #ifdef CONFIG_PROC_FS
6568         remove_proc_entry("ipv6_route", net->proc_net);
6569         remove_proc_entry("rt6_stats", net->proc_net);
6570 #endif
6571 }
6572
6573 static struct pernet_operations ip6_route_net_ops = {
6574         .init = ip6_route_net_init,
6575         .exit = ip6_route_net_exit,
6576 };
6577
6578 static int __net_init ipv6_inetpeer_init(struct net *net)
6579 {
6580         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6581
6582         if (!bp)
6583                 return -ENOMEM;
6584         inet_peer_base_init(bp);
6585         net->ipv6.peers = bp;
6586         return 0;
6587 }
6588
6589 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6590 {
6591         struct inet_peer_base *bp = net->ipv6.peers;
6592
6593         net->ipv6.peers = NULL;
6594         inetpeer_invalidate_tree(bp);
6595         kfree(bp);
6596 }
6597
6598 static struct pernet_operations ipv6_inetpeer_ops = {
6599         .init   =       ipv6_inetpeer_init,
6600         .exit   =       ipv6_inetpeer_exit,
6601 };
6602
6603 static struct pernet_operations ip6_route_net_late_ops = {
6604         .init = ip6_route_net_init_late,
6605         .exit = ip6_route_net_exit_late,
6606 };
6607
6608 static struct notifier_block ip6_route_dev_notifier = {
6609         .notifier_call = ip6_route_dev_notify,
6610         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6611 };
6612
6613 void __init ip6_route_init_special_entries(void)
6614 {
6615         /* Registering of the loopback is done before this portion of code,
6616          * the loopback reference in rt6_info will not be taken, do it
6617          * manually for init_net */
6618         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6619         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6620         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6621   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6622         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6623         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6624         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6625         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6626   #endif
6627 }
6628
6629 #if IS_BUILTIN(CONFIG_IPV6)
6630 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6631 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6632
6633 BTF_ID_LIST(btf_fib6_info_id)
6634 BTF_ID(struct, fib6_info)
6635
6636 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6637         .seq_ops                = &ipv6_route_seq_ops,
6638         .init_seq_private       = bpf_iter_init_seq_net,
6639         .fini_seq_private       = bpf_iter_fini_seq_net,
6640         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6641 };
6642
6643 static struct bpf_iter_reg ipv6_route_reg_info = {
6644         .target                 = "ipv6_route",
6645         .ctx_arg_info_size      = 1,
6646         .ctx_arg_info           = {
6647                 { offsetof(struct bpf_iter__ipv6_route, rt),
6648                   PTR_TO_BTF_ID_OR_NULL },
6649         },
6650         .seq_info               = &ipv6_route_seq_info,
6651 };
6652
6653 static int __init bpf_iter_register(void)
6654 {
6655         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6656         return bpf_iter_reg_target(&ipv6_route_reg_info);
6657 }
6658
6659 static void bpf_iter_unregister(void)
6660 {
6661         bpf_iter_unreg_target(&ipv6_route_reg_info);
6662 }
6663 #endif
6664 #endif
6665
6666 int __init ip6_route_init(void)
6667 {
6668         int ret;
6669         int cpu;
6670
6671         ret = -ENOMEM;
6672         ip6_dst_ops_template.kmem_cachep =
6673                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6674                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6675         if (!ip6_dst_ops_template.kmem_cachep)
6676                 goto out;
6677
6678         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6679         if (ret)
6680                 goto out_kmem_cache;
6681
6682         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6683         if (ret)
6684                 goto out_dst_entries;
6685
6686         ret = register_pernet_subsys(&ip6_route_net_ops);
6687         if (ret)
6688                 goto out_register_inetpeer;
6689
6690         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6691
6692         ret = fib6_init();
6693         if (ret)
6694                 goto out_register_subsys;
6695
6696         ret = xfrm6_init();
6697         if (ret)
6698                 goto out_fib6_init;
6699
6700         ret = fib6_rules_init();
6701         if (ret)
6702                 goto xfrm6_init;
6703
6704         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6705         if (ret)
6706                 goto fib6_rules_init;
6707
6708         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6709                                    inet6_rtm_newroute, NULL, 0);
6710         if (ret < 0)
6711                 goto out_register_late_subsys;
6712
6713         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6714                                    inet6_rtm_delroute, NULL, 0);
6715         if (ret < 0)
6716                 goto out_register_late_subsys;
6717
6718         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6719                                    inet6_rtm_getroute, NULL,
6720                                    RTNL_FLAG_DOIT_UNLOCKED);
6721         if (ret < 0)
6722                 goto out_register_late_subsys;
6723
6724         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6725         if (ret)
6726                 goto out_register_late_subsys;
6727
6728 #if IS_BUILTIN(CONFIG_IPV6)
6729 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6730         ret = bpf_iter_register();
6731         if (ret)
6732                 goto out_register_late_subsys;
6733 #endif
6734 #endif
6735
6736         for_each_possible_cpu(cpu) {
6737                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6738
6739                 INIT_LIST_HEAD(&ul->head);
6740                 INIT_LIST_HEAD(&ul->quarantine);
6741                 spin_lock_init(&ul->lock);
6742         }
6743
6744 out:
6745         return ret;
6746
6747 out_register_late_subsys:
6748         rtnl_unregister_all(PF_INET6);
6749         unregister_pernet_subsys(&ip6_route_net_late_ops);
6750 fib6_rules_init:
6751         fib6_rules_cleanup();
6752 xfrm6_init:
6753         xfrm6_fini();
6754 out_fib6_init:
6755         fib6_gc_cleanup();
6756 out_register_subsys:
6757         unregister_pernet_subsys(&ip6_route_net_ops);
6758 out_register_inetpeer:
6759         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6760 out_dst_entries:
6761         dst_entries_destroy(&ip6_dst_blackhole_ops);
6762 out_kmem_cache:
6763         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6764         goto out;
6765 }
6766
6767 void ip6_route_cleanup(void)
6768 {
6769 #if IS_BUILTIN(CONFIG_IPV6)
6770 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6771         bpf_iter_unregister();
6772 #endif
6773 #endif
6774         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6775         unregister_pernet_subsys(&ip6_route_net_late_ops);
6776         fib6_rules_cleanup();
6777         xfrm6_fini();
6778         fib6_gc_cleanup();
6779         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6780         unregister_pernet_subsys(&ip6_route_net_ops);
6781         dst_entries_destroy(&ip6_dst_blackhole_ops);
6782         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6783 }