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);
  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                                         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 {
 376         struct rt6_info *rt = (struct rt6_info *)dst;
 377         struct inet6_dev *idev = rt->rt6i_idev;
 378
 379         if (idev && idev->dev != blackhole_netdev) {
 380                 struct inet6_dev *blackhole_idev = in6_dev_get(blackhole_netdev);
 381
 382                 if (blackhole_idev) {
 383                         rt->rt6i_idev = blackhole_idev;
 384                         in6_dev_put(idev);
 385                 }
 386         }
 387 }
 388
 389 static bool __rt6_check_expired(const struct rt6_info *rt)
 390 {
 391         if (rt->rt6i_flags & RTF_EXPIRES)
 392                 return time_after(jiffies, rt->dst.expires);
 393         else
 394                 return false;
 395 }
 396
 397 static bool rt6_check_expired(const struct rt6_info *rt)
 398 {
 399         struct fib6_info *from;
 400
 401         from = rcu_dereference(rt->from);
 402
 403         if (rt->rt6i_flags & RTF_EXPIRES) {
 404                 if (time_after(jiffies, rt->dst.expires))
 405                         return true;
 406         } else if (from) {
 407                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 408                         fib6_check_expired(from);
 409         }
 410         return false;
 411 }
 412
 413 void fib6_select_path(const struct net *net, struct fib6_result *res,
 414                       struct flowi6 *fl6, int oif, bool have_oif_match,
 415                       const struct sk_buff *skb, int strict)
 416 {
 417         struct fib6_info *sibling, *next_sibling;
 418         struct fib6_info *match = res->f6i;
 419
 420         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 421                 goto out;
 422
 423         if (match->nh && have_oif_match && res->nh)
 424                 return;
 425
 426         if (skb)
 427                 IP6CB(skb)->flags |= IP6SKB_MULTIPATH;
 428
 429         /* We might have already computed the hash for ICMPv6 errors. In such
 430          * case it will always be non-zero. Otherwise now is the time to do it.
 431          */
 432         if (!fl6->mp_hash &&
 433             (!match->nh || nexthop_is_multipath(match->nh)))
 434                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 435
 436         if (unlikely(match->nh)) {
 437                 nexthop_path_fib6_result(res, fl6->mp_hash);
 438                 return;
 439         }
 440
 441         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 442                 goto out;
 443
 444         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 445                                  fib6_siblings) {
 446                 const struct fib6_nh *nh = sibling->fib6_nh;
 447                 int nh_upper_bound;
 448
 449                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 450                 if (fl6->mp_hash > nh_upper_bound)
 451                         continue;
 452                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 453                         break;
 454                 match = sibling;
 455                 break;
 456         }
 457
 458 out:
 459         res->f6i = match;
 460         res->nh = match->fib6_nh;
 461 }
 462
 463 /*
 464  *      Route lookup. rcu_read_lock() should be held.
 465  */
 466
 467 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 468                                const struct in6_addr *saddr, int oif, int flags)
 469 {
 470         const struct net_device *dev;
 471
 472         if (nh->fib_nh_flags & RTNH_F_DEAD)
 473                 return false;
 474
 475         dev = nh->fib_nh_dev;
 476         if (oif) {
 477                 if (dev->ifindex == oif)
 478                         return true;
 479         } else {
 480                 if (ipv6_chk_addr(net, saddr, dev,
 481                                   flags & RT6_LOOKUP_F_IFACE))
 482                         return true;
 483         }
 484
 485         return false;
 486 }
 487
 488 struct fib6_nh_dm_arg {
 489         struct net              *net;
 490         const struct in6_addr   *saddr;
 491         int                     oif;
 492         int                     flags;
 493         struct fib6_nh          *nh;
 494 };
 495
 496 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 497 {
 498         struct fib6_nh_dm_arg *arg = _arg;
 499
 500         arg->nh = nh;
 501         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 502                                   arg->flags);
 503 }
 504
 505 /* returns fib6_nh from nexthop or NULL */
 506 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 507                                         struct fib6_result *res,
 508                                         const struct in6_addr *saddr,
 509                                         int oif, int flags)
 510 {
 511         struct fib6_nh_dm_arg arg = {
 512                 .net   = net,
 513                 .saddr = saddr,
 514                 .oif   = oif,
 515                 .flags = flags,
 516         };
 517
 518         if (nexthop_is_blackhole(nh))
 519                 return NULL;
 520
 521         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 522                 return arg.nh;
 523
 524         return NULL;
 525 }
 526
 527 static void rt6_device_match(struct net *net, struct fib6_result *res,
 528                              const struct in6_addr *saddr, int oif, int flags)
 529 {
 530         struct fib6_info *f6i = res->f6i;
 531         struct fib6_info *spf6i;
 532         struct fib6_nh *nh;
 533
 534         if (!oif && ipv6_addr_any(saddr)) {
 535                 if (unlikely(f6i->nh)) {
 536                         nh = nexthop_fib6_nh(f6i->nh);
 537                         if (nexthop_is_blackhole(f6i->nh))
 538                                 goto out_blackhole;
 539                 } else {
 540                         nh = f6i->fib6_nh;
 541                 }
 542                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 543                         goto out;
 544         }
 545
 546         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 547                 bool matched = false;
 548
 549                 if (unlikely(spf6i->nh)) {
 550                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 551                                               oif, flags);
 552                         if (nh)
 553                                 matched = true;
 554                 } else {
 555                         nh = spf6i->fib6_nh;
 556                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 557                                 matched = true;
 558                 }
 559                 if (matched) {
 560                         res->f6i = spf6i;
 561                         goto out;
 562                 }
 563         }
 564
 565         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 566                 res->f6i = net->ipv6.fib6_null_entry;
 567                 nh = res->f6i->fib6_nh;
 568                 goto out;
 569         }
 570
 571         if (unlikely(f6i->nh)) {
 572                 nh = nexthop_fib6_nh(f6i->nh);
 573                 if (nexthop_is_blackhole(f6i->nh))
 574                         goto out_blackhole;
 575         } else {
 576                 nh = f6i->fib6_nh;
 577         }
 578
 579         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 580                 res->f6i = net->ipv6.fib6_null_entry;
 581                 nh = res->f6i->fib6_nh;
 582         }
 583 out:
 584         res->nh = nh;
 585         res->fib6_type = res->f6i->fib6_type;
 586         res->fib6_flags = res->f6i->fib6_flags;
 587         return;
 588
 589 out_blackhole:
 590         res->fib6_flags |= RTF_REJECT;
 591         res->fib6_type = RTN_BLACKHOLE;
 592         res->nh = nh;
 593 }
 594
 595 #ifdef CONFIG_IPV6_ROUTER_PREF
 596 struct __rt6_probe_work {
 597         struct work_struct work;
 598         struct in6_addr target;
 599         struct net_device *dev;
 600         netdevice_tracker dev_tracker;
 601 };
 602
 603 static void rt6_probe_deferred(struct work_struct *w)
 604 {
 605         struct in6_addr mcaddr;
 606         struct __rt6_probe_work *work =
 607                 container_of(w, struct __rt6_probe_work, work);
 608
 609         addrconf_addr_solict_mult(&work->target, &mcaddr);
 610         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 611         netdev_put(work->dev, &work->dev_tracker);
 612         kfree(work);
 613 }
 614
 615 static void rt6_probe(struct fib6_nh *fib6_nh)
 616 {
 617         struct __rt6_probe_work *work = NULL;
 618         const struct in6_addr *nh_gw;
 619         unsigned long last_probe;
 620         struct neighbour *neigh;
 621         struct net_device *dev;
 622         struct inet6_dev *idev;
 623
 624         /*
 625          * Okay, this does not seem to be appropriate
 626          * for now, however, we need to check if it
 627          * is really so; aka Router Reachability Probing.
 628          *
 629          * Router Reachability Probe MUST be rate-limited
 630          * to no more than one per minute.
 631          */
 632         if (!fib6_nh->fib_nh_gw_family)
 633                 return;
 634
 635         nh_gw = &fib6_nh->fib_nh_gw6;
 636         dev = fib6_nh->fib_nh_dev;
 637         rcu_read_lock();
 638         last_probe = READ_ONCE(fib6_nh->last_probe);
 639         idev = __in6_dev_get(dev);
 640         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 641         if (neigh) {
 642                 if (READ_ONCE(neigh->nud_state) & NUD_VALID)
 643                         goto out;
 644
 645                 write_lock_bh(&neigh->lock);
 646                 if (!(neigh->nud_state & NUD_VALID) &&
 647                     time_after(jiffies,
 648                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 649                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 650                         if (work)
 651                                 __neigh_set_probe_once(neigh);
 652                 }
 653                 write_unlock_bh(&neigh->lock);
 654         } else if (time_after(jiffies, last_probe +
 655                                        idev->cnf.rtr_probe_interval)) {
 656                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 657         }
 658
 659         if (!work || cmpxchg(&fib6_nh->last_probe,
 660                              last_probe, jiffies) != last_probe) {
 661                 kfree(work);
 662         } else {
 663                 INIT_WORK(&work->work, rt6_probe_deferred);
 664                 work->target = *nh_gw;
 665                 netdev_hold(dev, &work->dev_tracker, GFP_ATOMIC);
 666                 work->dev = dev;
 667                 schedule_work(&work->work);
 668         }
 669
 670 out:
 671         rcu_read_unlock();
 672 }
 673 #else
 674 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 675 {
 676 }
 677 #endif
 678
 679 /*
 680  * Default Router Selection (RFC 2461 6.3.6)
 681  */
 682 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 683 {
 684         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 685         struct neighbour *neigh;
 686
 687         rcu_read_lock();
 688         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 689                                           &fib6_nh->fib_nh_gw6);
 690         if (neigh) {
 691                 u8 nud_state = READ_ONCE(neigh->nud_state);
 692
 693                 if (nud_state & NUD_VALID)
 694                         ret = RT6_NUD_SUCCEED;
 695 #ifdef CONFIG_IPV6_ROUTER_PREF
 696                 else if (!(nud_state & NUD_FAILED))
 697                         ret = RT6_NUD_SUCCEED;
 698                 else
 699                         ret = RT6_NUD_FAIL_PROBE;
 700 #endif
 701         } else {
 702                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 703                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 704         }
 705         rcu_read_unlock();
 706
 707         return ret;
 708 }
 709
 710 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 711                            int strict)
 712 {
 713         int m = 0;
 714
 715         if (!oif || nh->fib_nh_dev->ifindex == oif)
 716                 m = 2;
 717
 718         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 719                 return RT6_NUD_FAIL_HARD;
 720 #ifdef CONFIG_IPV6_ROUTER_PREF
 721         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 722 #endif
 723         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 724             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 725                 int n = rt6_check_neigh(nh);
 726                 if (n < 0)
 727                         return n;
 728         }
 729         return m;
 730 }
 731
 732 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 733                        int oif, int strict, int *mpri, bool *do_rr)
 734 {
 735         bool match_do_rr = false;
 736         bool rc = false;
 737         int m;
 738
 739         if (nh->fib_nh_flags & RTNH_F_DEAD)
 740                 goto out;
 741
 742         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 743             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 744             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 745                 goto out;
 746
 747         m = rt6_score_route(nh, fib6_flags, oif, strict);
 748         if (m == RT6_NUD_FAIL_DO_RR) {
 749                 match_do_rr = true;
 750                 m = 0; /* lowest valid score */
 751         } else if (m == RT6_NUD_FAIL_HARD) {
 752                 goto out;
 753         }
 754
 755         if (strict & RT6_LOOKUP_F_REACHABLE)
 756                 rt6_probe(nh);
 757
 758         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 759         if (m > *mpri) {
 760                 *do_rr = match_do_rr;
 761                 *mpri = m;
 762                 rc = true;
 763         }
 764 out:
 765         return rc;
 766 }
 767
 768 struct fib6_nh_frl_arg {
 769         u32             flags;
 770         int             oif;
 771         int             strict;
 772         int             *mpri;
 773         bool            *do_rr;
 774         struct fib6_nh  *nh;
 775 };
 776
 777 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 778 {
 779         struct fib6_nh_frl_arg *arg = _arg;
 780
 781         arg->nh = nh;
 782         return find_match(nh, arg->flags, arg->oif, arg->strict,
 783                           arg->mpri, arg->do_rr);
 784 }
 785
 786 static void __find_rr_leaf(struct fib6_info *f6i_start,
 787                            struct fib6_info *nomatch, u32 metric,
 788                            struct fib6_result *res, struct fib6_info **cont,
 789                            int oif, int strict, bool *do_rr, int *mpri)
 790 {
 791         struct fib6_info *f6i;
 792
 793         for (f6i = f6i_start;
 794              f6i && f6i != nomatch;
 795              f6i = rcu_dereference(f6i->fib6_next)) {
 796                 bool matched = false;
 797                 struct fib6_nh *nh;
 798
 799                 if (cont && f6i->fib6_metric != metric) {
 800                         *cont = f6i;
 801                         return;
 802                 }
 803
 804                 if (fib6_check_expired(f6i))
 805                         continue;
 806
 807                 if (unlikely(f6i->nh)) {
 808                         struct fib6_nh_frl_arg arg = {
 809                                 .flags  = f6i->fib6_flags,
 810                                 .oif    = oif,
 811                                 .strict = strict,
 812                                 .mpri   = mpri,
 813                                 .do_rr  = do_rr
 814                         };
 815
 816                         if (nexthop_is_blackhole(f6i->nh)) {
 817                                 res->fib6_flags = RTF_REJECT;
 818                                 res->fib6_type = RTN_BLACKHOLE;
 819                                 res->f6i = f6i;
 820                                 res->nh = nexthop_fib6_nh(f6i->nh);
 821                                 return;
 822                         }
 823                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 824                                                      &arg)) {
 825                                 matched = true;
 826                                 nh = arg.nh;
 827                         }
 828                 } else {
 829                         nh = f6i->fib6_nh;
 830                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 831                                        mpri, do_rr))
 832                                 matched = true;
 833                 }
 834                 if (matched) {
 835                         res->f6i = f6i;
 836                         res->nh = nh;
 837                         res->fib6_flags = f6i->fib6_flags;
 838                         res->fib6_type = f6i->fib6_type;
 839                 }
 840         }
 841 }
 842
 843 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 844                          struct fib6_info *rr_head, int oif, int strict,
 845                          bool *do_rr, struct fib6_result *res)
 846 {
 847         u32 metric = rr_head->fib6_metric;
 848         struct fib6_info *cont = NULL;
 849         int mpri = -1;
 850
 851         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 852                        oif, strict, do_rr, &mpri);
 853
 854         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 855                        oif, strict, do_rr, &mpri);
 856
 857         if (res->f6i || !cont)
 858                 return;
 859
 860         __find_rr_leaf(cont, NULL, metric, res, NULL,
 861                        oif, strict, do_rr, &mpri);
 862 }
 863
 864 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 865                        struct fib6_result *res, int strict)
 866 {
 867         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 868         struct fib6_info *rt0;
 869         bool do_rr = false;
 870         int key_plen;
 871
 872         /* make sure this function or its helpers sets f6i */
 873         res->f6i = NULL;
 874
 875         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 876                 goto out;
 877
 878         rt0 = rcu_dereference(fn->rr_ptr);
 879         if (!rt0)
 880                 rt0 = leaf;
 881
 882         /* Double check to make sure fn is not an intermediate node
 883          * and fn->leaf does not points to its child's leaf
 884          * (This might happen if all routes under fn are deleted from
 885          * the tree and fib6_repair_tree() is called on the node.)
 886          */
 887         key_plen = rt0->fib6_dst.plen;
 888 #ifdef CONFIG_IPV6_SUBTREES
 889         if (rt0->fib6_src.plen)
 890                 key_plen = rt0->fib6_src.plen;
 891 #endif
 892         if (fn->fn_bit != key_plen)
 893                 goto out;
 894
 895         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 896         if (do_rr) {
 897                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 898
 899                 /* no entries matched; do round-robin */
 900                 if (!next || next->fib6_metric != rt0->fib6_metric)
 901                         next = leaf;
 902
 903                 if (next != rt0) {
 904                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 905                         /* make sure next is not being deleted from the tree */
 906                         if (next->fib6_node)
 907                                 rcu_assign_pointer(fn->rr_ptr, next);
 908                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 909                 }
 910         }
 911
 912 out:
 913         if (!res->f6i) {
 914                 res->f6i = net->ipv6.fib6_null_entry;
 915                 res->nh = res->f6i->fib6_nh;
 916                 res->fib6_flags = res->f6i->fib6_flags;
 917                 res->fib6_type = res->f6i->fib6_type;
 918         }
 919 }
 920
 921 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 922 {
 923         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 924                res->nh->fib_nh_gw_family;
 925 }
 926
 927 #ifdef CONFIG_IPV6_ROUTE_INFO
 928 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 929                   const struct in6_addr *gwaddr)
 930 {
 931         struct net *net = dev_net(dev);
 932         struct route_info *rinfo = (struct route_info *) opt;
 933         struct in6_addr prefix_buf, *prefix;
 934         struct fib6_table *table;
 935         unsigned int pref;
 936         unsigned long lifetime;
 937         struct fib6_info *rt;
 938
 939         if (len < sizeof(struct route_info)) {
 940                 return -EINVAL;
 941         }
 942
 943         /* Sanity check for prefix_len and length */
 944         if (rinfo->length > 3) {
 945                 return -EINVAL;
 946         } else if (rinfo->prefix_len > 128) {
 947                 return -EINVAL;
 948         } else if (rinfo->prefix_len > 64) {
 949                 if (rinfo->length < 2) {
 950                         return -EINVAL;
 951                 }
 952         } else if (rinfo->prefix_len > 0) {
 953                 if (rinfo->length < 1) {
 954                         return -EINVAL;
 955                 }
 956         }
 957
 958         pref = rinfo->route_pref;
 959         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 960                 return -EINVAL;
 961
 962         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 963
 964         if (rinfo->length == 3)
 965                 prefix = (struct in6_addr *)rinfo->prefix;
 966         else {
 967                 /* this function is safe */
 968                 ipv6_addr_prefix(&prefix_buf,
 969                                  (struct in6_addr *)rinfo->prefix,
 970                                  rinfo->prefix_len);
 971                 prefix = &prefix_buf;
 972         }
 973
 974         if (rinfo->prefix_len == 0)
 975                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 976         else
 977                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 978                                         gwaddr, dev);
 979
 980         if (rt && !lifetime) {
 981                 ip6_del_rt(net, rt, false);
 982                 rt = NULL;
 983         }
 984
 985         if (!rt && lifetime)
 986                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 987                                         dev, pref);
 988         else if (rt)
 989                 rt->fib6_flags = RTF_ROUTEINFO |
 990                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 991
 992         if (rt) {
 993                 table = rt->fib6_table;
 994                 spin_lock_bh(&table->tb6_lock);
 995
 996                 if (!addrconf_finite_timeout(lifetime)) {
 997                         fib6_clean_expires(rt);
 998                         fib6_remove_gc_list(rt);
 999                 } else {
1000                         fib6_set_expires(rt, jiffies + HZ * lifetime);
1001                         fib6_add_gc_list(rt);
1002                 }
1003
1004                 spin_unlock_bh(&table->tb6_lock);
1005
1006                 fib6_info_release(rt);
1007         }
1008         return 0;
1009 }
1010 #endif
1011
1012 /*
1013  *      Misc support functions
1014  */
1015
1016 /* called with rcu_lock held */
1017 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1018 {
1019         struct net_device *dev = res->nh->fib_nh_dev;
1020
1021         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1022                 /* for copies of local routes, dst->dev needs to be the
1023                  * device if it is a master device, the master device if
1024                  * device is enslaved, and the loopback as the default
1025                  */
1026                 if (netif_is_l3_slave(dev) &&
1027                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1028                         dev = l3mdev_master_dev_rcu(dev);
1029                 else if (!netif_is_l3_master(dev))
1030                         dev = dev_net(dev)->loopback_dev;
1031                 /* last case is netif_is_l3_master(dev) is true in which
1032                  * case we want dev returned to be dev
1033                  */
1034         }
1035
1036         return dev;
1037 }
1038
1039 static const int fib6_prop[RTN_MAX + 1] = {
1040         [RTN_UNSPEC]    = 0,
1041         [RTN_UNICAST]   = 0,
1042         [RTN_LOCAL]     = 0,
1043         [RTN_BROADCAST] = 0,
1044         [RTN_ANYCAST]   = 0,
1045         [RTN_MULTICAST] = 0,
1046         [RTN_BLACKHOLE] = -EINVAL,
1047         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1048         [RTN_PROHIBIT]  = -EACCES,
1049         [RTN_THROW]     = -EAGAIN,
1050         [RTN_NAT]       = -EINVAL,
1051         [RTN_XRESOLVE]  = -EINVAL,
1052 };
1053
1054 static int ip6_rt_type_to_error(u8 fib6_type)
1055 {
1056         return fib6_prop[fib6_type];
1057 }
1058
1059 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1060 {
1061         unsigned short flags = 0;
1062
1063         if (rt->dst_nocount)
1064                 flags |= DST_NOCOUNT;
1065         if (rt->dst_nopolicy)
1066                 flags |= DST_NOPOLICY;
1067
1068         return flags;
1069 }
1070
1071 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1072 {
1073         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1074
1075         switch (fib6_type) {
1076         case RTN_BLACKHOLE:
1077                 rt->dst.output = dst_discard_out;
1078                 rt->dst.input = dst_discard;
1079                 break;
1080         case RTN_PROHIBIT:
1081                 rt->dst.output = ip6_pkt_prohibit_out;
1082                 rt->dst.input = ip6_pkt_prohibit;
1083                 break;
1084         case RTN_THROW:
1085         case RTN_UNREACHABLE:
1086         default:
1087                 rt->dst.output = ip6_pkt_discard_out;
1088                 rt->dst.input = ip6_pkt_discard;
1089                 break;
1090         }
1091 }
1092
1093 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1094 {
1095         struct fib6_info *f6i = res->f6i;
1096
1097         if (res->fib6_flags & RTF_REJECT) {
1098                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1099                 return;
1100         }
1101
1102         rt->dst.error = 0;
1103         rt->dst.output = ip6_output;
1104
1105         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1106                 rt->dst.input = ip6_input;
1107         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1108                 rt->dst.input = ip6_mc_input;
1109         } else {
1110                 rt->dst.input = ip6_forward;
1111         }
1112
1113         if (res->nh->fib_nh_lws) {
1114                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1115                 lwtunnel_set_redirect(&rt->dst);
1116         }
1117
1118         rt->dst.lastuse = jiffies;
1119 }
1120
1121 /* Caller must already hold reference to @from */
1122 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1123 {
1124         rt->rt6i_flags &= ~RTF_EXPIRES;
1125         rcu_assign_pointer(rt->from, from);
1126         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1127 }
1128
1129 /* Caller must already hold reference to f6i in result */
1130 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1131 {
1132         const struct fib6_nh *nh = res->nh;
1133         const struct net_device *dev = nh->fib_nh_dev;
1134         struct fib6_info *f6i = res->f6i;
1135
1136         ip6_rt_init_dst(rt, res);
1137
1138         rt->rt6i_dst = f6i->fib6_dst;
1139         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1140         rt->rt6i_flags = res->fib6_flags;
1141         if (nh->fib_nh_gw_family) {
1142                 rt->rt6i_gateway = nh->fib_nh_gw6;
1143                 rt->rt6i_flags |= RTF_GATEWAY;
1144         }
1145         rt6_set_from(rt, f6i);
1146 #ifdef CONFIG_IPV6_SUBTREES
1147         rt->rt6i_src = f6i->fib6_src;
1148 #endif
1149 }
1150
1151 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1152                                         struct in6_addr *saddr)
1153 {
1154         struct fib6_node *pn, *sn;
1155         while (1) {
1156                 if (fn->fn_flags & RTN_TL_ROOT)
1157                         return NULL;
1158                 pn = rcu_dereference(fn->parent);
1159                 sn = FIB6_SUBTREE(pn);
1160                 if (sn && sn != fn)
1161                         fn = fib6_node_lookup(sn, NULL, saddr);
1162                 else
1163                         fn = pn;
1164                 if (fn->fn_flags & RTN_RTINFO)
1165                         return fn;
1166         }
1167 }
1168
1169 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1170 {
1171         struct rt6_info *rt = *prt;
1172
1173         if (dst_hold_safe(&rt->dst))
1174                 return true;
1175         if (net) {
1176                 rt = net->ipv6.ip6_null_entry;
1177                 dst_hold(&rt->dst);
1178         } else {
1179                 rt = NULL;
1180         }
1181         *prt = rt;
1182         return false;
1183 }
1184
1185 /* called with rcu_lock held */
1186 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1187 {
1188         struct net_device *dev = res->nh->fib_nh_dev;
1189         struct fib6_info *f6i = res->f6i;
1190         unsigned short flags;
1191         struct rt6_info *nrt;
1192
1193         if (!fib6_info_hold_safe(f6i))
1194                 goto fallback;
1195
1196         flags = fib6_info_dst_flags(f6i);
1197         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1198         if (!nrt) {
1199                 fib6_info_release(f6i);
1200                 goto fallback;
1201         }
1202
1203         ip6_rt_copy_init(nrt, res);
1204         return nrt;
1205
1206 fallback:
1207         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1208         dst_hold(&nrt->dst);
1209         return nrt;
1210 }
1211
1212 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1213                                              struct fib6_table *table,
1214                                              struct flowi6 *fl6,
1215                                              const struct sk_buff *skb,
1216                                              int flags)
1217 {
1218         struct fib6_result res = {};
1219         struct fib6_node *fn;
1220         struct rt6_info *rt;
1221
1222         rcu_read_lock();
1223         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1224 restart:
1225         res.f6i = rcu_dereference(fn->leaf);
1226         if (!res.f6i)
1227                 res.f6i = net->ipv6.fib6_null_entry;
1228         else
1229                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1230                                  flags);
1231
1232         if (res.f6i == net->ipv6.fib6_null_entry) {
1233                 fn = fib6_backtrack(fn, &fl6->saddr);
1234                 if (fn)
1235                         goto restart;
1236
1237                 rt = net->ipv6.ip6_null_entry;
1238                 dst_hold(&rt->dst);
1239                 goto out;
1240         } else if (res.fib6_flags & RTF_REJECT) {
1241                 goto do_create;
1242         }
1243
1244         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1245                          fl6->flowi6_oif != 0, skb, flags);
1246
1247         /* Search through exception table */
1248         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1249         if (rt) {
1250                 if (ip6_hold_safe(net, &rt))
1251                         dst_use_noref(&rt->dst, jiffies);
1252         } else {
1253 do_create:
1254                 rt = ip6_create_rt_rcu(&res);
1255         }
1256
1257 out:
1258         trace_fib6_table_lookup(net, &res, table, fl6);
1259
1260         rcu_read_unlock();
1261
1262         return rt;
1263 }
1264
1265 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1266                                    const struct sk_buff *skb, int flags)
1267 {
1268         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1269 }
1270 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1271
1272 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1273                             const struct in6_addr *saddr, int oif,
1274                             const struct sk_buff *skb, int strict)
1275 {
1276         struct flowi6 fl6 = {
1277                 .flowi6_oif = oif,
1278                 .daddr = *daddr,
1279         };
1280         struct dst_entry *dst;
1281         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1282
1283         if (saddr) {
1284                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1285                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1286         }
1287
1288         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1289         if (dst->error == 0)
1290                 return (struct rt6_info *) dst;
1291
1292         dst_release(dst);
1293
1294         return NULL;
1295 }
1296 EXPORT_SYMBOL(rt6_lookup);
1297
1298 /* ip6_ins_rt is called with FREE table->tb6_lock.
1299  * It takes new route entry, the addition fails by any reason the
1300  * route is released.
1301  * Caller must hold dst before calling it.
1302  */
1303
1304 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1305                         struct netlink_ext_ack *extack)
1306 {
1307         int err;
1308         struct fib6_table *table;
1309
1310         table = rt->fib6_table;
1311         spin_lock_bh(&table->tb6_lock);
1312         err = fib6_add(&table->tb6_root, rt, info, extack);
1313         spin_unlock_bh(&table->tb6_lock);
1314
1315         return err;
1316 }
1317
1318 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1319 {
1320         struct nl_info info = { .nl_net = net, };
1321
1322         return __ip6_ins_rt(rt, &info, NULL);
1323 }
1324
1325 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1326                                            const struct in6_addr *daddr,
1327                                            const struct in6_addr *saddr)
1328 {
1329         struct fib6_info *f6i = res->f6i;
1330         struct net_device *dev;
1331         struct rt6_info *rt;
1332
1333         /*
1334          *      Clone the route.
1335          */
1336
1337         if (!fib6_info_hold_safe(f6i))
1338                 return NULL;
1339
1340         dev = ip6_rt_get_dev_rcu(res);
1341         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1342         if (!rt) {
1343                 fib6_info_release(f6i);
1344                 return NULL;
1345         }
1346
1347         ip6_rt_copy_init(rt, res);
1348         rt->rt6i_flags |= RTF_CACHE;
1349         rt->rt6i_dst.addr = *daddr;
1350         rt->rt6i_dst.plen = 128;
1351
1352         if (!rt6_is_gw_or_nonexthop(res)) {
1353                 if (f6i->fib6_dst.plen != 128 &&
1354                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1355                         rt->rt6i_flags |= RTF_ANYCAST;
1356 #ifdef CONFIG_IPV6_SUBTREES
1357                 if (rt->rt6i_src.plen && saddr) {
1358                         rt->rt6i_src.addr = *saddr;
1359                         rt->rt6i_src.plen = 128;
1360                 }
1361 #endif
1362         }
1363
1364         return rt;
1365 }
1366
1367 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1368 {
1369         struct fib6_info *f6i = res->f6i;
1370         unsigned short flags = fib6_info_dst_flags(f6i);
1371         struct net_device *dev;
1372         struct rt6_info *pcpu_rt;
1373
1374         if (!fib6_info_hold_safe(f6i))
1375                 return NULL;
1376
1377         rcu_read_lock();
1378         dev = ip6_rt_get_dev_rcu(res);
1379         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1380         rcu_read_unlock();
1381         if (!pcpu_rt) {
1382                 fib6_info_release(f6i);
1383                 return NULL;
1384         }
1385         ip6_rt_copy_init(pcpu_rt, res);
1386         pcpu_rt->rt6i_flags |= RTF_PCPU;
1387
1388         if (f6i->nh)
1389                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1390
1391         return pcpu_rt;
1392 }
1393
1394 static bool rt6_is_valid(const struct rt6_info *rt6)
1395 {
1396         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1397 }
1398
1399 /* It should be called with rcu_read_lock() acquired */
1400 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1401 {
1402         struct rt6_info *pcpu_rt;
1403
1404         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1405
1406         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1407                 struct rt6_info *prev, **p;
1408
1409                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1410                 prev = xchg(p, NULL);
1411                 if (prev) {
1412                         dst_dev_put(&prev->dst);
1413                         dst_release(&prev->dst);
1414                 }
1415
1416                 pcpu_rt = NULL;
1417         }
1418
1419         return pcpu_rt;
1420 }
1421
1422 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1423                                             const struct fib6_result *res)
1424 {
1425         struct rt6_info *pcpu_rt, *prev, **p;
1426
1427         pcpu_rt = ip6_rt_pcpu_alloc(res);
1428         if (!pcpu_rt)
1429                 return NULL;
1430
1431         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1432         prev = cmpxchg(p, NULL, pcpu_rt);
1433         BUG_ON(prev);
1434
1435         if (res->f6i->fib6_destroying) {
1436                 struct fib6_info *from;
1437
1438                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1439                 fib6_info_release(from);
1440         }
1441
1442         return pcpu_rt;
1443 }
1444
1445 /* exception hash table implementation
1446  */
1447 static DEFINE_SPINLOCK(rt6_exception_lock);
1448
1449 /* Remove rt6_ex from hash table and free the memory
1450  * Caller must hold rt6_exception_lock
1451  */
1452 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1453                                  struct rt6_exception *rt6_ex)
1454 {
1455         struct fib6_info *from;
1456         struct net *net;
1457
1458         if (!bucket || !rt6_ex)
1459                 return;
1460
1461         net = dev_net(rt6_ex->rt6i->dst.dev);
1462         net->ipv6.rt6_stats->fib_rt_cache--;
1463
1464         /* purge completely the exception to allow releasing the held resources:
1465          * some [sk] cache may keep the dst around for unlimited time
1466          */
1467         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1468         fib6_info_release(from);
1469         dst_dev_put(&rt6_ex->rt6i->dst);
1470
1471         hlist_del_rcu(&rt6_ex->hlist);
1472         dst_release(&rt6_ex->rt6i->dst);
1473         kfree_rcu(rt6_ex, rcu);
1474         WARN_ON_ONCE(!bucket->depth);
1475         bucket->depth--;
1476 }
1477
1478 /* Remove oldest rt6_ex in bucket and free the memory
1479  * Caller must hold rt6_exception_lock
1480  */
1481 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1482 {
1483         struct rt6_exception *rt6_ex, *oldest = NULL;
1484
1485         if (!bucket)
1486                 return;
1487
1488         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1489                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1490                         oldest = rt6_ex;
1491         }
1492         rt6_remove_exception(bucket, oldest);
1493 }
1494
1495 static u32 rt6_exception_hash(const struct in6_addr *dst,
1496                               const struct in6_addr *src)
1497 {
1498         static siphash_aligned_key_t rt6_exception_key;
1499         struct {
1500                 struct in6_addr dst;
1501                 struct in6_addr src;
1502         } __aligned(SIPHASH_ALIGNMENT) combined = {
1503                 .dst = *dst,
1504         };
1505         u64 val;
1506
1507         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1508
1509 #ifdef CONFIG_IPV6_SUBTREES
1510         if (src)
1511                 combined.src = *src;
1512 #endif
1513         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1514
1515         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1516 }
1517
1518 /* Helper function to find the cached rt in the hash table
1519  * and update bucket pointer to point to the bucket for this
1520  * (daddr, saddr) pair
1521  * Caller must hold rt6_exception_lock
1522  */
1523 static struct rt6_exception *
1524 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1525                               const struct in6_addr *daddr,
1526                               const struct in6_addr *saddr)
1527 {
1528         struct rt6_exception *rt6_ex;
1529         u32 hval;
1530
1531         if (!(*bucket) || !daddr)
1532                 return NULL;
1533
1534         hval = rt6_exception_hash(daddr, saddr);
1535         *bucket += hval;
1536
1537         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1538                 struct rt6_info *rt6 = rt6_ex->rt6i;
1539                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1540
1541 #ifdef CONFIG_IPV6_SUBTREES
1542                 if (matched && saddr)
1543                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1544 #endif
1545                 if (matched)
1546                         return rt6_ex;
1547         }
1548         return NULL;
1549 }
1550
1551 /* Helper function to find the cached rt in the hash table
1552  * and update bucket pointer to point to the bucket for this
1553  * (daddr, saddr) pair
1554  * Caller must hold rcu_read_lock()
1555  */
1556 static struct rt6_exception *
1557 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1558                          const struct in6_addr *daddr,
1559                          const struct in6_addr *saddr)
1560 {
1561         struct rt6_exception *rt6_ex;
1562         u32 hval;
1563
1564         WARN_ON_ONCE(!rcu_read_lock_held());
1565
1566         if (!(*bucket) || !daddr)
1567                 return NULL;
1568
1569         hval = rt6_exception_hash(daddr, saddr);
1570         *bucket += hval;
1571
1572         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1573                 struct rt6_info *rt6 = rt6_ex->rt6i;
1574                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1575
1576 #ifdef CONFIG_IPV6_SUBTREES
1577                 if (matched && saddr)
1578                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1579 #endif
1580                 if (matched)
1581                         return rt6_ex;
1582         }
1583         return NULL;
1584 }
1585
1586 static unsigned int fib6_mtu(const struct fib6_result *res)
1587 {
1588         const struct fib6_nh *nh = res->nh;
1589         unsigned int mtu;
1590
1591         if (res->f6i->fib6_pmtu) {
1592                 mtu = res->f6i->fib6_pmtu;
1593         } else {
1594                 struct net_device *dev = nh->fib_nh_dev;
1595                 struct inet6_dev *idev;
1596
1597                 rcu_read_lock();
1598                 idev = __in6_dev_get(dev);
1599                 mtu = idev->cnf.mtu6;
1600                 rcu_read_unlock();
1601         }
1602
1603         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1604
1605         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1606 }
1607
1608 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1609
1610 /* used when the flushed bit is not relevant, only access to the bucket
1611  * (ie., all bucket users except rt6_insert_exception);
1612  *
1613  * called under rcu lock; sometimes called with rt6_exception_lock held
1614  */
1615 static
1616 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1617                                                        spinlock_t *lock)
1618 {
1619         struct rt6_exception_bucket *bucket;
1620
1621         if (lock)
1622                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1623                                                    lockdep_is_held(lock));
1624         else
1625                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1626
1627         /* remove bucket flushed bit if set */
1628         if (bucket) {
1629                 unsigned long p = (unsigned long)bucket;
1630
1631                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1632                 bucket = (struct rt6_exception_bucket *)p;
1633         }
1634
1635         return bucket;
1636 }
1637
1638 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1639 {
1640         unsigned long p = (unsigned long)bucket;
1641
1642         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1643 }
1644
1645 /* called with rt6_exception_lock held */
1646 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1647                                               spinlock_t *lock)
1648 {
1649         struct rt6_exception_bucket *bucket;
1650         unsigned long p;
1651
1652         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1653                                            lockdep_is_held(lock));
1654
1655         p = (unsigned long)bucket;
1656         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1657         bucket = (struct rt6_exception_bucket *)p;
1658         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1659 }
1660
1661 static int rt6_insert_exception(struct rt6_info *nrt,
1662                                 const struct fib6_result *res)
1663 {
1664         struct net *net = dev_net(nrt->dst.dev);
1665         struct rt6_exception_bucket *bucket;
1666         struct fib6_info *f6i = res->f6i;
1667         struct in6_addr *src_key = NULL;
1668         struct rt6_exception *rt6_ex;
1669         struct fib6_nh *nh = res->nh;
1670         int max_depth;
1671         int err = 0;
1672
1673         spin_lock_bh(&rt6_exception_lock);
1674
1675         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1676                                           lockdep_is_held(&rt6_exception_lock));
1677         if (!bucket) {
1678                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1679                                  GFP_ATOMIC);
1680                 if (!bucket) {
1681                         err = -ENOMEM;
1682                         goto out;
1683                 }
1684                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1685         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1686                 err = -EINVAL;
1687                 goto out;
1688         }
1689
1690 #ifdef CONFIG_IPV6_SUBTREES
1691         /* fib6_src.plen != 0 indicates f6i is in subtree
1692          * and exception table is indexed by a hash of
1693          * both fib6_dst and fib6_src.
1694          * Otherwise, the exception table is indexed by
1695          * a hash of only fib6_dst.
1696          */
1697         if (f6i->fib6_src.plen)
1698                 src_key = &nrt->rt6i_src.addr;
1699 #endif
1700         /* rt6_mtu_change() might lower mtu on f6i.
1701          * Only insert this exception route if its mtu
1702          * is less than f6i's mtu value.
1703          */
1704         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1705                 err = -EINVAL;
1706                 goto out;
1707         }
1708
1709         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1710                                                src_key);
1711         if (rt6_ex)
1712                 rt6_remove_exception(bucket, rt6_ex);
1713
1714         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1715         if (!rt6_ex) {
1716                 err = -ENOMEM;
1717                 goto out;
1718         }
1719         rt6_ex->rt6i = nrt;
1720         rt6_ex->stamp = jiffies;
1721         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1722         bucket->depth++;
1723         net->ipv6.rt6_stats->fib_rt_cache++;
1724
1725         /* Randomize max depth to avoid some side channels attacks. */
1726         max_depth = FIB6_MAX_DEPTH + get_random_u32_below(FIB6_MAX_DEPTH);
1727         while (bucket->depth > max_depth)
1728                 rt6_exception_remove_oldest(bucket);
1729
1730 out:
1731         spin_unlock_bh(&rt6_exception_lock);
1732
1733         /* Update fn->fn_sernum to invalidate all cached dst */
1734         if (!err) {
1735                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1736                 fib6_update_sernum(net, f6i);
1737                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1738                 fib6_force_start_gc(net);
1739         }
1740
1741         return err;
1742 }
1743
1744 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1745 {
1746         struct rt6_exception_bucket *bucket;
1747         struct rt6_exception *rt6_ex;
1748         struct hlist_node *tmp;
1749         int i;
1750
1751         spin_lock_bh(&rt6_exception_lock);
1752
1753         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1754         if (!bucket)
1755                 goto out;
1756
1757         /* Prevent rt6_insert_exception() to recreate the bucket list */
1758         if (!from)
1759                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1760
1761         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1762                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1763                         if (!from ||
1764                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1765                                 rt6_remove_exception(bucket, rt6_ex);
1766                 }
1767                 WARN_ON_ONCE(!from && bucket->depth);
1768                 bucket++;
1769         }
1770 out:
1771         spin_unlock_bh(&rt6_exception_lock);
1772 }
1773
1774 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1775 {
1776         struct fib6_info *f6i = arg;
1777
1778         fib6_nh_flush_exceptions(nh, f6i);
1779
1780         return 0;
1781 }
1782
1783 void rt6_flush_exceptions(struct fib6_info *f6i)
1784 {
1785         if (f6i->nh)
1786                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1787                                          f6i);
1788         else
1789                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1790 }
1791
1792 /* Find cached rt in the hash table inside passed in rt
1793  * Caller has to hold rcu_read_lock()
1794  */
1795 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1796                                            const struct in6_addr *daddr,
1797                                            const struct in6_addr *saddr)
1798 {
1799         const struct in6_addr *src_key = NULL;
1800         struct rt6_exception_bucket *bucket;
1801         struct rt6_exception *rt6_ex;
1802         struct rt6_info *ret = NULL;
1803
1804 #ifdef CONFIG_IPV6_SUBTREES
1805         /* fib6i_src.plen != 0 indicates f6i is in subtree
1806          * and exception table is indexed by a hash of
1807          * both fib6_dst and fib6_src.
1808          * However, the src addr used to create the hash
1809          * might not be exactly the passed in saddr which
1810          * is a /128 addr from the flow.
1811          * So we need to use f6i->fib6_src to redo lookup
1812          * if the passed in saddr does not find anything.
1813          * (See the logic in ip6_rt_cache_alloc() on how
1814          * rt->rt6i_src is updated.)
1815          */
1816         if (res->f6i->fib6_src.plen)
1817                 src_key = saddr;
1818 find_ex:
1819 #endif
1820         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1821         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1822
1823         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1824                 ret = rt6_ex->rt6i;
1825
1826 #ifdef CONFIG_IPV6_SUBTREES
1827         /* Use fib6_src as src_key and redo lookup */
1828         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1829                 src_key = &res->f6i->fib6_src.addr;
1830                 goto find_ex;
1831         }
1832 #endif
1833
1834         return ret;
1835 }
1836
1837 /* Remove the passed in cached rt from the hash table that contains it */
1838 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1839                                     const struct rt6_info *rt)
1840 {
1841         const struct in6_addr *src_key = NULL;
1842         struct rt6_exception_bucket *bucket;
1843         struct rt6_exception *rt6_ex;
1844         int err;
1845
1846         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1847                 return -ENOENT;
1848
1849         spin_lock_bh(&rt6_exception_lock);
1850         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1851
1852 #ifdef CONFIG_IPV6_SUBTREES
1853         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1854          * and exception table is indexed by a hash of
1855          * both rt6i_dst and rt6i_src.
1856          * Otherwise, the exception table is indexed by
1857          * a hash of only rt6i_dst.
1858          */
1859         if (plen)
1860                 src_key = &rt->rt6i_src.addr;
1861 #endif
1862         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1863                                                &rt->rt6i_dst.addr,
1864                                                src_key);
1865         if (rt6_ex) {
1866                 rt6_remove_exception(bucket, rt6_ex);
1867                 err = 0;
1868         } else {
1869                 err = -ENOENT;
1870         }
1871
1872         spin_unlock_bh(&rt6_exception_lock);
1873         return err;
1874 }
1875
1876 struct fib6_nh_excptn_arg {
1877         struct rt6_info *rt;
1878         int             plen;
1879 };
1880
1881 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1882 {
1883         struct fib6_nh_excptn_arg *arg = _arg;
1884         int err;
1885
1886         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1887         if (err == 0)
1888                 return 1;
1889
1890         return 0;
1891 }
1892
1893 static int rt6_remove_exception_rt(struct rt6_info *rt)
1894 {
1895         struct fib6_info *from;
1896
1897         from = rcu_dereference(rt->from);
1898         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1899                 return -EINVAL;
1900
1901         if (from->nh) {
1902                 struct fib6_nh_excptn_arg arg = {
1903                         .rt = rt,
1904                         .plen = from->fib6_src.plen
1905                 };
1906                 int rc;
1907
1908                 /* rc = 1 means an entry was found */
1909                 rc = nexthop_for_each_fib6_nh(from->nh,
1910                                               rt6_nh_remove_exception_rt,
1911                                               &arg);
1912                 return rc ? 0 : -ENOENT;
1913         }
1914
1915         return fib6_nh_remove_exception(from->fib6_nh,
1916                                         from->fib6_src.plen, rt);
1917 }
1918
1919 /* Find rt6_ex which contains the passed in rt cache and
1920  * refresh its stamp
1921  */
1922 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1923                                      const struct rt6_info *rt)
1924 {
1925         const struct in6_addr *src_key = NULL;
1926         struct rt6_exception_bucket *bucket;
1927         struct rt6_exception *rt6_ex;
1928
1929         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1930 #ifdef CONFIG_IPV6_SUBTREES
1931         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1932          * and exception table is indexed by a hash of
1933          * both rt6i_dst and rt6i_src.
1934          * Otherwise, the exception table is indexed by
1935          * a hash of only rt6i_dst.
1936          */
1937         if (plen)
1938                 src_key = &rt->rt6i_src.addr;
1939 #endif
1940         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1941         if (rt6_ex)
1942                 rt6_ex->stamp = jiffies;
1943 }
1944
1945 struct fib6_nh_match_arg {
1946         const struct net_device *dev;
1947         const struct in6_addr   *gw;
1948         struct fib6_nh          *match;
1949 };
1950
1951 /* determine if fib6_nh has given device and gateway */
1952 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1953 {
1954         struct fib6_nh_match_arg *arg = _arg;
1955
1956         if (arg->dev != nh->fib_nh_dev ||
1957             (arg->gw && !nh->fib_nh_gw_family) ||
1958             (!arg->gw && nh->fib_nh_gw_family) ||
1959             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1960                 return 0;
1961
1962         arg->match = nh;
1963
1964         /* found a match, break the loop */
1965         return 1;
1966 }
1967
1968 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1969 {
1970         struct fib6_info *from;
1971         struct fib6_nh *fib6_nh;
1972
1973         rcu_read_lock();
1974
1975         from = rcu_dereference(rt->from);
1976         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1977                 goto unlock;
1978
1979         if (from->nh) {
1980                 struct fib6_nh_match_arg arg = {
1981                         .dev = rt->dst.dev,
1982                         .gw = &rt->rt6i_gateway,
1983                 };
1984
1985                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1986
1987                 if (!arg.match)
1988                         goto unlock;
1989                 fib6_nh = arg.match;
1990         } else {
1991                 fib6_nh = from->fib6_nh;
1992         }
1993         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1994 unlock:
1995         rcu_read_unlock();
1996 }
1997
1998 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1999                                          struct rt6_info *rt, int mtu)
2000 {
2001         /* If the new MTU is lower than the route PMTU, this new MTU will be the
2002          * lowest MTU in the path: always allow updating the route PMTU to
2003          * reflect PMTU decreases.
2004          *
2005          * If the new MTU is higher, and the route PMTU is equal to the local
2006          * MTU, this means the old MTU is the lowest in the path, so allow
2007          * updating it: if other nodes now have lower MTUs, PMTU discovery will
2008          * handle this.
2009          */
2010
2011         if (dst_mtu(&rt->dst) >= mtu)
2012                 return true;
2013
2014         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2015                 return true;
2016
2017         return false;
2018 }
2019
2020 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2021                                        const struct fib6_nh *nh, int mtu)
2022 {
2023         struct rt6_exception_bucket *bucket;
2024         struct rt6_exception *rt6_ex;
2025         int i;
2026
2027         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2028         if (!bucket)
2029                 return;
2030
2031         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2032                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2033                         struct rt6_info *entry = rt6_ex->rt6i;
2034
2035                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2036                          * route), the metrics of its rt->from have already
2037                          * been updated.
2038                          */
2039                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2040                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2041                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2042                 }
2043                 bucket++;
2044         }
2045 }
2046
2047 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2048
2049 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2050                                             const struct in6_addr *gateway)
2051 {
2052         struct rt6_exception_bucket *bucket;
2053         struct rt6_exception *rt6_ex;
2054         struct hlist_node *tmp;
2055         int i;
2056
2057         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2058                 return;
2059
2060         spin_lock_bh(&rt6_exception_lock);
2061         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2062         if (bucket) {
2063                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2064                         hlist_for_each_entry_safe(rt6_ex, tmp,
2065                                                   &bucket->chain, hlist) {
2066                                 struct rt6_info *entry = rt6_ex->rt6i;
2067
2068                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2069                                     RTF_CACHE_GATEWAY &&
2070                                     ipv6_addr_equal(gateway,
2071                                                     &entry->rt6i_gateway)) {
2072                                         rt6_remove_exception(bucket, rt6_ex);
2073                                 }
2074                         }
2075                         bucket++;
2076                 }
2077         }
2078
2079         spin_unlock_bh(&rt6_exception_lock);
2080 }
2081
2082 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2083                                       struct rt6_exception *rt6_ex,
2084                                       struct fib6_gc_args *gc_args,
2085                                       unsigned long now)
2086 {
2087         struct rt6_info *rt = rt6_ex->rt6i;
2088
2089         /* we are pruning and obsoleting aged-out and non gateway exceptions
2090          * even if others have still references to them, so that on next
2091          * dst_check() such references can be dropped.
2092          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2093          * expired, independently from their aging, as per RFC 8201 section 4
2094          */
2095         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2096                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2097                         pr_debug("aging clone %p\n", rt);
2098                         rt6_remove_exception(bucket, rt6_ex);
2099                         return;
2100                 }
2101         } else if (time_after(jiffies, rt->dst.expires)) {
2102                 pr_debug("purging expired route %p\n", rt);
2103                 rt6_remove_exception(bucket, rt6_ex);
2104                 return;
2105         }
2106
2107         if (rt->rt6i_flags & RTF_GATEWAY) {
2108                 struct neighbour *neigh;
2109
2110                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2111
2112                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2113                         pr_debug("purging route %p via non-router but gateway\n",
2114                                  rt);
2115                         rt6_remove_exception(bucket, rt6_ex);
2116                         return;
2117                 }
2118         }
2119
2120         gc_args->more++;
2121 }
2122
2123 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2124                                    struct fib6_gc_args *gc_args,
2125                                    unsigned long now)
2126 {
2127         struct rt6_exception_bucket *bucket;
2128         struct rt6_exception *rt6_ex;
2129         struct hlist_node *tmp;
2130         int i;
2131
2132         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2133                 return;
2134
2135         rcu_read_lock_bh();
2136         spin_lock(&rt6_exception_lock);
2137         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2138         if (bucket) {
2139                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2140                         hlist_for_each_entry_safe(rt6_ex, tmp,
2141                                                   &bucket->chain, hlist) {
2142                                 rt6_age_examine_exception(bucket, rt6_ex,
2143                                                           gc_args, now);
2144                         }
2145                         bucket++;
2146                 }
2147         }
2148         spin_unlock(&rt6_exception_lock);
2149         rcu_read_unlock_bh();
2150 }
2151
2152 struct fib6_nh_age_excptn_arg {
2153         struct fib6_gc_args     *gc_args;
2154         unsigned long           now;
2155 };
2156
2157 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2158 {
2159         struct fib6_nh_age_excptn_arg *arg = _arg;
2160
2161         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2162         return 0;
2163 }
2164
2165 void rt6_age_exceptions(struct fib6_info *f6i,
2166                         struct fib6_gc_args *gc_args,
2167                         unsigned long now)
2168 {
2169         if (f6i->nh) {
2170                 struct fib6_nh_age_excptn_arg arg = {
2171                         .gc_args = gc_args,
2172                         .now = now
2173                 };
2174
2175                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2176                                          &arg);
2177         } else {
2178                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2179         }
2180 }
2181
2182 /* must be called with rcu lock held */
2183 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2184                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2185 {
2186         struct fib6_node *fn, *saved_fn;
2187
2188         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2189         saved_fn = fn;
2190
2191 redo_rt6_select:
2192         rt6_select(net, fn, oif, res, strict);
2193         if (res->f6i == net->ipv6.fib6_null_entry) {
2194                 fn = fib6_backtrack(fn, &fl6->saddr);
2195                 if (fn)
2196                         goto redo_rt6_select;
2197                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2198                         /* also consider unreachable route */
2199                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2200                         fn = saved_fn;
2201                         goto redo_rt6_select;
2202                 }
2203         }
2204
2205         trace_fib6_table_lookup(net, res, table, fl6);
2206
2207         return 0;
2208 }
2209
2210 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2211                                int oif, struct flowi6 *fl6,
2212                                const struct sk_buff *skb, int flags)
2213 {
2214         struct fib6_result res = {};
2215         struct rt6_info *rt = NULL;
2216         int strict = 0;
2217
2218         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2219                      !rcu_read_lock_held());
2220
2221         strict |= flags & RT6_LOOKUP_F_IFACE;
2222         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2223         if (net->ipv6.devconf_all->forwarding == 0)
2224                 strict |= RT6_LOOKUP_F_REACHABLE;
2225
2226         rcu_read_lock();
2227
2228         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2229         if (res.f6i == net->ipv6.fib6_null_entry)
2230                 goto out;
2231
2232         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2233
2234         /*Search through exception table */
2235         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2236         if (rt) {
2237                 goto out;
2238         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2239                             !res.nh->fib_nh_gw_family)) {
2240                 /* Create a RTF_CACHE clone which will not be
2241                  * owned by the fib6 tree.  It is for the special case where
2242                  * the daddr in the skb during the neighbor look-up is different
2243                  * from the fl6->daddr used to look-up route here.
2244                  */
2245                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2246
2247                 if (rt) {
2248                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2249                          * As rt6_uncached_list_add() does not consume refcnt,
2250                          * this refcnt is always returned to the caller even
2251                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2252                          */
2253                         rt6_uncached_list_add(rt);
2254                         rcu_read_unlock();
2255
2256                         return rt;
2257                 }
2258         } else {
2259                 /* Get a percpu copy */
2260                 local_bh_disable();
2261                 rt = rt6_get_pcpu_route(&res);
2262
2263                 if (!rt)
2264                         rt = rt6_make_pcpu_route(net, &res);
2265
2266                 local_bh_enable();
2267         }
2268 out:
2269         if (!rt)
2270                 rt = net->ipv6.ip6_null_entry;
2271         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2272                 ip6_hold_safe(net, &rt);
2273         rcu_read_unlock();
2274
2275         return rt;
2276 }
2277 EXPORT_SYMBOL_GPL(ip6_pol_route);
2278
2279 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2280                                             struct fib6_table *table,
2281                                             struct flowi6 *fl6,
2282                                             const struct sk_buff *skb,
2283                                             int flags)
2284 {
2285         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2286 }
2287
2288 struct dst_entry *ip6_route_input_lookup(struct net *net,
2289                                          struct net_device *dev,
2290                                          struct flowi6 *fl6,
2291                                          const struct sk_buff *skb,
2292                                          int flags)
2293 {
2294         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2295                 flags |= RT6_LOOKUP_F_IFACE;
2296
2297         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2298 }
2299 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2300
2301 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2302                                   struct flow_keys *keys,
2303                                   struct flow_keys *flkeys)
2304 {
2305         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2306         const struct ipv6hdr *key_iph = outer_iph;
2307         struct flow_keys *_flkeys = flkeys;
2308         const struct ipv6hdr *inner_iph;
2309         const struct icmp6hdr *icmph;
2310         struct ipv6hdr _inner_iph;
2311         struct icmp6hdr _icmph;
2312
2313         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2314                 goto out;
2315
2316         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2317                                    sizeof(_icmph), &_icmph);
2318         if (!icmph)
2319                 goto out;
2320
2321         if (!icmpv6_is_err(icmph->icmp6_type))
2322                 goto out;
2323
2324         inner_iph = skb_header_pointer(skb,
2325                                        skb_transport_offset(skb) + sizeof(*icmph),
2326                                        sizeof(_inner_iph), &_inner_iph);
2327         if (!inner_iph)
2328                 goto out;
2329
2330         key_iph = inner_iph;
2331         _flkeys = NULL;
2332 out:
2333         if (_flkeys) {
2334                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2335                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2336                 keys->tags.flow_label = _flkeys->tags.flow_label;
2337                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2338         } else {
2339                 keys->addrs.v6addrs.src = key_iph->saddr;
2340                 keys->addrs.v6addrs.dst = key_iph->daddr;
2341                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2342                 keys->basic.ip_proto = key_iph->nexthdr;
2343         }
2344 }
2345
2346 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2347                                            const struct sk_buff *skb,
2348                                            bool *p_has_inner)
2349 {
2350         u32 hash_fields = ip6_multipath_hash_fields(net);
2351         struct flow_keys keys, hash_keys;
2352
2353         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2354                 return 0;
2355
2356         memset(&hash_keys, 0, sizeof(hash_keys));
2357         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2358
2359         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2360         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2361                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2362         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2363                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2364         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2365                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2366         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2367                 hash_keys.tags.flow_label = keys.tags.flow_label;
2368         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2369                 hash_keys.ports.src = keys.ports.src;
2370         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2371                 hash_keys.ports.dst = keys.ports.dst;
2372
2373         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2374         return flow_hash_from_keys(&hash_keys);
2375 }
2376
2377 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2378                                            const struct sk_buff *skb,
2379                                            bool has_inner)
2380 {
2381         u32 hash_fields = ip6_multipath_hash_fields(net);
2382         struct flow_keys keys, hash_keys;
2383
2384         /* We assume the packet carries an encapsulation, but if none was
2385          * encountered during dissection of the outer flow, then there is no
2386          * point in calling the flow dissector again.
2387          */
2388         if (!has_inner)
2389                 return 0;
2390
2391         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2392                 return 0;
2393
2394         memset(&hash_keys, 0, sizeof(hash_keys));
2395         skb_flow_dissect_flow_keys(skb, &keys, 0);
2396
2397         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2398                 return 0;
2399
2400         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2401                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2402                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2403                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2404                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2405                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2406         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2407                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2408                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2409                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2410                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2411                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2412                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2413                         hash_keys.tags.flow_label = keys.tags.flow_label;
2414         }
2415
2416         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2417                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2418         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2419                 hash_keys.ports.src = keys.ports.src;
2420         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2421                 hash_keys.ports.dst = keys.ports.dst;
2422
2423         return flow_hash_from_keys(&hash_keys);
2424 }
2425
2426 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2427                                          const struct sk_buff *skb)
2428 {
2429         u32 mhash, mhash_inner;
2430         bool has_inner = true;
2431
2432         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2433         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2434
2435         return jhash_2words(mhash, mhash_inner, 0);
2436 }
2437
2438 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2439                                          const struct flowi6 *fl6)
2440 {
2441         u32 hash_fields = ip6_multipath_hash_fields(net);
2442         struct flow_keys hash_keys;
2443
2444         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2445                 return 0;
2446
2447         memset(&hash_keys, 0, sizeof(hash_keys));
2448         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2449         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2450                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2451         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2452                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2453         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2454                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2455         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2456                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2457         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2458                 hash_keys.ports.src = fl6->fl6_sport;
2459         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2460                 hash_keys.ports.dst = fl6->fl6_dport;
2461
2462         return flow_hash_from_keys(&hash_keys);
2463 }
2464
2465 /* if skb is set it will be used and fl6 can be NULL */
2466 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2467                        const struct sk_buff *skb, struct flow_keys *flkeys)
2468 {
2469         struct flow_keys hash_keys;
2470         u32 mhash = 0;
2471
2472         switch (ip6_multipath_hash_policy(net)) {
2473         case 0:
2474                 memset(&hash_keys, 0, sizeof(hash_keys));
2475                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2476                 if (skb) {
2477                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2478                 } else {
2479                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2480                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2481                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2482                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2483                 }
2484                 mhash = flow_hash_from_keys(&hash_keys);
2485                 break;
2486         case 1:
2487                 if (skb) {
2488                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2489                         struct flow_keys keys;
2490
2491                         /* short-circuit if we already have L4 hash present */
2492                         if (skb->l4_hash)
2493                                 return skb_get_hash_raw(skb) >> 1;
2494
2495                         memset(&hash_keys, 0, sizeof(hash_keys));
2496
2497                         if (!flkeys) {
2498                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2499                                 flkeys = &keys;
2500                         }
2501                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2502                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2503                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2504                         hash_keys.ports.src = flkeys->ports.src;
2505                         hash_keys.ports.dst = flkeys->ports.dst;
2506                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2507                 } else {
2508                         memset(&hash_keys, 0, sizeof(hash_keys));
2509                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2510                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2511                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2512                         hash_keys.ports.src = fl6->fl6_sport;
2513                         hash_keys.ports.dst = fl6->fl6_dport;
2514                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2515                 }
2516                 mhash = flow_hash_from_keys(&hash_keys);
2517                 break;
2518         case 2:
2519                 memset(&hash_keys, 0, sizeof(hash_keys));
2520                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2521                 if (skb) {
2522                         struct flow_keys keys;
2523
2524                         if (!flkeys) {
2525                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2526                                 flkeys = &keys;
2527                         }
2528
2529                         /* Inner can be v4 or v6 */
2530                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2531                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2532                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2533                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2534                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2535                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2536                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2537                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2538                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2539                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2540                         } else {
2541                                 /* Same as case 0 */
2542                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2543                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2544                         }
2545                 } else {
2546                         /* Same as case 0 */
2547                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2548                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2549                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2550                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2551                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2552                 }
2553                 mhash = flow_hash_from_keys(&hash_keys);
2554                 break;
2555         case 3:
2556                 if (skb)
2557                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2558                 else
2559                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2560                 break;
2561         }
2562
2563         return mhash >> 1;
2564 }
2565
2566 /* Called with rcu held */
2567 void ip6_route_input(struct sk_buff *skb)
2568 {
2569         const struct ipv6hdr *iph = ipv6_hdr(skb);
2570         struct net *net = dev_net(skb->dev);
2571         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2572         struct ip_tunnel_info *tun_info;
2573         struct flowi6 fl6 = {
2574                 .flowi6_iif = skb->dev->ifindex,
2575                 .daddr = iph->daddr,
2576                 .saddr = iph->saddr,
2577                 .flowlabel = ip6_flowinfo(iph),
2578                 .flowi6_mark = skb->mark,
2579                 .flowi6_proto = iph->nexthdr,
2580         };
2581         struct flow_keys *flkeys = NULL, _flkeys;
2582
2583         tun_info = skb_tunnel_info(skb);
2584         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2585                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2586
2587         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2588                 flkeys = &_flkeys;
2589
2590         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2591                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2592         skb_dst_drop(skb);
2593         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2594                                                       &fl6, skb, flags));
2595 }
2596
2597 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2598                                              struct fib6_table *table,
2599                                              struct flowi6 *fl6,
2600                                              const struct sk_buff *skb,
2601                                              int flags)
2602 {
2603         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2604 }
2605
2606 static struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2607                                                       const struct sock *sk,
2608                                                       struct flowi6 *fl6,
2609                                                       int flags)
2610 {
2611         bool any_src;
2612
2613         if (ipv6_addr_type(&fl6->daddr) &
2614             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2615                 struct dst_entry *dst;
2616
2617                 /* This function does not take refcnt on the dst */
2618                 dst = l3mdev_link_scope_lookup(net, fl6);
2619                 if (dst)
2620                         return dst;
2621         }
2622
2623         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2624
2625         flags |= RT6_LOOKUP_F_DST_NOREF;
2626         any_src = ipv6_addr_any(&fl6->saddr);
2627         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2628             (fl6->flowi6_oif && any_src))
2629                 flags |= RT6_LOOKUP_F_IFACE;
2630
2631         if (!any_src)
2632                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2633         else if (sk)
2634                 flags |= rt6_srcprefs2flags(READ_ONCE(inet6_sk(sk)->srcprefs));
2635
2636         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2637 }
2638
2639 struct dst_entry *ip6_route_output_flags(struct net *net,
2640                                          const struct sock *sk,
2641                                          struct flowi6 *fl6,
2642                                          int flags)
2643 {
2644         struct dst_entry *dst;
2645         struct rt6_info *rt6;
2646
2647         rcu_read_lock();
2648         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2649         rt6 = (struct rt6_info *)dst;
2650         /* For dst cached in uncached_list, refcnt is already taken. */
2651         if (list_empty(&rt6->dst.rt_uncached) && !dst_hold_safe(dst)) {
2652                 dst = &net->ipv6.ip6_null_entry->dst;
2653                 dst_hold(dst);
2654         }
2655         rcu_read_unlock();
2656
2657         return dst;
2658 }
2659 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2660
2661 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2662 {
2663         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2664         struct net_device *loopback_dev = net->loopback_dev;
2665         struct dst_entry *new = NULL;
2666
2667         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev,
2668                        DST_OBSOLETE_DEAD, 0);
2669         if (rt) {
2670                 rt6_info_init(rt);
2671                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2672
2673                 new = &rt->dst;
2674                 new->__use = 1;
2675                 new->input = dst_discard;
2676                 new->output = dst_discard_out;
2677
2678                 dst_copy_metrics(new, &ort->dst);
2679
2680                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2681                 rt->rt6i_gateway = ort->rt6i_gateway;
2682                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2683
2684                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2685 #ifdef CONFIG_IPV6_SUBTREES
2686                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2687 #endif
2688         }
2689
2690         dst_release(dst_orig);
2691         return new ? new : ERR_PTR(-ENOMEM);
2692 }
2693
2694 /*
2695  *      Destination cache support functions
2696  */
2697
2698 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2699 {
2700         u32 rt_cookie = 0;
2701
2702         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2703                 return false;
2704
2705         if (fib6_check_expired(f6i))
2706                 return false;
2707
2708         return true;
2709 }
2710
2711 static struct dst_entry *rt6_check(struct rt6_info *rt,
2712                                    struct fib6_info *from,
2713                                    u32 cookie)
2714 {
2715         u32 rt_cookie = 0;
2716
2717         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2718             rt_cookie != cookie)
2719                 return NULL;
2720
2721         if (rt6_check_expired(rt))
2722                 return NULL;
2723
2724         return &rt->dst;
2725 }
2726
2727 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2728                                             struct fib6_info *from,
2729                                             u32 cookie)
2730 {
2731         if (!__rt6_check_expired(rt) &&
2732             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2733             fib6_check(from, cookie))
2734                 return &rt->dst;
2735         else
2736                 return NULL;
2737 }
2738
2739 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2740                                                         u32 cookie)
2741 {
2742         struct dst_entry *dst_ret;
2743         struct fib6_info *from;
2744         struct rt6_info *rt;
2745
2746         rt = container_of(dst, struct rt6_info, dst);
2747
2748         if (rt->sernum)
2749                 return rt6_is_valid(rt) ? dst : NULL;
2750
2751         rcu_read_lock();
2752
2753         /* All IPV6 dsts are created with ->obsolete set to the value
2754          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2755          * into this function always.
2756          */
2757
2758         from = rcu_dereference(rt->from);
2759
2760         if (from && (rt->rt6i_flags & RTF_PCPU ||
2761             unlikely(!list_empty(&rt->dst.rt_uncached))))
2762                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2763         else
2764                 dst_ret = rt6_check(rt, from, cookie);
2765
2766         rcu_read_unlock();
2767
2768         return dst_ret;
2769 }
2770 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2771
2772 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2773 {
2774         struct rt6_info *rt = (struct rt6_info *) dst;
2775
2776         if (rt) {
2777                 if (rt->rt6i_flags & RTF_CACHE) {
2778                         rcu_read_lock();
2779                         if (rt6_check_expired(rt)) {
2780                                 rt6_remove_exception_rt(rt);
2781                                 dst = NULL;
2782                         }
2783                         rcu_read_unlock();
2784                 } else {
2785                         dst_release(dst);
2786                         dst = NULL;
2787                 }
2788         }
2789         return dst;
2790 }
2791
2792 static void ip6_link_failure(struct sk_buff *skb)
2793 {
2794         struct rt6_info *rt;
2795
2796         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2797
2798         rt = (struct rt6_info *) skb_dst(skb);
2799         if (rt) {
2800                 rcu_read_lock();
2801                 if (rt->rt6i_flags & RTF_CACHE) {
2802                         rt6_remove_exception_rt(rt);
2803                 } else {
2804                         struct fib6_info *from;
2805                         struct fib6_node *fn;
2806
2807                         from = rcu_dereference(rt->from);
2808                         if (from) {
2809                                 fn = rcu_dereference(from->fib6_node);
2810                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2811                                         WRITE_ONCE(fn->fn_sernum, -1);
2812                         }
2813                 }
2814                 rcu_read_unlock();
2815         }
2816 }
2817
2818 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2819 {
2820         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2821                 struct fib6_info *from;
2822
2823                 rcu_read_lock();
2824                 from = rcu_dereference(rt0->from);
2825                 if (from)
2826                         rt0->dst.expires = from->expires;
2827                 rcu_read_unlock();
2828         }
2829
2830         dst_set_expires(&rt0->dst, timeout);
2831         rt0->rt6i_flags |= RTF_EXPIRES;
2832 }
2833
2834 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2835 {
2836         struct net *net = dev_net(rt->dst.dev);
2837
2838         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2839         rt->rt6i_flags |= RTF_MODIFIED;
2840         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2841 }
2842
2843 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2844 {
2845         return !(rt->rt6i_flags & RTF_CACHE) &&
2846                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2847 }
2848
2849 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2850                                  const struct ipv6hdr *iph, u32 mtu,
2851                                  bool confirm_neigh)
2852 {
2853         const struct in6_addr *daddr, *saddr;
2854         struct rt6_info *rt6 = (struct rt6_info *)dst;
2855
2856         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2857          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2858          * [see also comment in rt6_mtu_change_route()]
2859          */
2860
2861         if (iph) {
2862                 daddr = &iph->daddr;
2863                 saddr = &iph->saddr;
2864         } else if (sk) {
2865                 daddr = &sk->sk_v6_daddr;
2866                 saddr = &inet6_sk(sk)->saddr;
2867         } else {
2868                 daddr = NULL;
2869                 saddr = NULL;
2870         }
2871
2872         if (confirm_neigh)
2873                 dst_confirm_neigh(dst, daddr);
2874
2875         if (mtu < IPV6_MIN_MTU)
2876                 return;
2877         if (mtu >= dst_mtu(dst))
2878                 return;
2879
2880         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2881                 rt6_do_update_pmtu(rt6, mtu);
2882                 /* update rt6_ex->stamp for cache */
2883                 if (rt6->rt6i_flags & RTF_CACHE)
2884                         rt6_update_exception_stamp_rt(rt6);
2885         } else if (daddr) {
2886                 struct fib6_result res = {};
2887                 struct rt6_info *nrt6;
2888
2889                 rcu_read_lock();
2890                 res.f6i = rcu_dereference(rt6->from);
2891                 if (!res.f6i)
2892                         goto out_unlock;
2893
2894                 res.fib6_flags = res.f6i->fib6_flags;
2895                 res.fib6_type = res.f6i->fib6_type;
2896
2897                 if (res.f6i->nh) {
2898                         struct fib6_nh_match_arg arg = {
2899                                 .dev = dst->dev,
2900                                 .gw = &rt6->rt6i_gateway,
2901                         };
2902
2903                         nexthop_for_each_fib6_nh(res.f6i->nh,
2904                                                  fib6_nh_find_match, &arg);
2905
2906                         /* fib6_info uses a nexthop that does not have fib6_nh
2907                          * using the dst->dev + gw. Should be impossible.
2908                          */
2909                         if (!arg.match)
2910                                 goto out_unlock;
2911
2912                         res.nh = arg.match;
2913                 } else {
2914                         res.nh = res.f6i->fib6_nh;
2915                 }
2916
2917                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2918                 if (nrt6) {
2919                         rt6_do_update_pmtu(nrt6, mtu);
2920                         if (rt6_insert_exception(nrt6, &res))
2921                                 dst_release_immediate(&nrt6->dst);
2922                 }
2923 out_unlock:
2924                 rcu_read_unlock();
2925         }
2926 }
2927
2928 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2929                                struct sk_buff *skb, u32 mtu,
2930                                bool confirm_neigh)
2931 {
2932         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2933                              confirm_neigh);
2934 }
2935
2936 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2937                      int oif, u32 mark, kuid_t uid)
2938 {
2939         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2940         struct dst_entry *dst;
2941         struct flowi6 fl6 = {
2942                 .flowi6_oif = oif,
2943                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2944                 .daddr = iph->daddr,
2945                 .saddr = iph->saddr,
2946                 .flowlabel = ip6_flowinfo(iph),
2947                 .flowi6_uid = uid,
2948         };
2949
2950         dst = ip6_route_output(net, NULL, &fl6);
2951         if (!dst->error)
2952                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2953         dst_release(dst);
2954 }
2955 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2956
2957 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2958 {
2959         int oif = sk->sk_bound_dev_if;
2960         struct dst_entry *dst;
2961
2962         if (!oif && skb->dev)
2963                 oif = l3mdev_master_ifindex(skb->dev);
2964
2965         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, READ_ONCE(sk->sk_mark),
2966                         sk->sk_uid);
2967
2968         dst = __sk_dst_get(sk);
2969         if (!dst || !dst->obsolete ||
2970             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2971                 return;
2972
2973         bh_lock_sock(sk);
2974         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2975                 ip6_datagram_dst_update(sk, false);
2976         bh_unlock_sock(sk);
2977 }
2978 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2979
2980 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2981                            const struct flowi6 *fl6)
2982 {
2983 #ifdef CONFIG_IPV6_SUBTREES
2984         struct ipv6_pinfo *np = inet6_sk(sk);
2985 #endif
2986
2987         ip6_dst_store(sk, dst,
2988                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2989                       &sk->sk_v6_daddr : NULL,
2990 #ifdef CONFIG_IPV6_SUBTREES
2991                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2992                       &np->saddr :
2993 #endif
2994                       NULL);
2995 }
2996
2997 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2998                                   struct flowi6 *fl6,
2999                                   const struct in6_addr *gw,
3000                                   struct rt6_info **ret)
3001 {
3002         const struct fib6_nh *nh = res->nh;
3003
3004         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
3005             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
3006                 return false;
3007
3008         /* rt_cache's gateway might be different from its 'parent'
3009          * in the case of an ip redirect.
3010          * So we keep searching in the exception table if the gateway
3011          * is different.
3012          */
3013         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3014                 struct rt6_info *rt_cache;
3015
3016                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3017                 if (rt_cache &&
3018                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3019                         *ret = rt_cache;
3020                         return true;
3021                 }
3022                 return false;
3023         }
3024         return true;
3025 }
3026
3027 struct fib6_nh_rd_arg {
3028         struct fib6_result      *res;
3029         struct flowi6           *fl6;
3030         const struct in6_addr   *gw;
3031         struct rt6_info         **ret;
3032 };
3033
3034 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3035 {
3036         struct fib6_nh_rd_arg *arg = _arg;
3037
3038         arg->res->nh = nh;
3039         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3040 }
3041
3042 /* Handle redirects */
3043 struct ip6rd_flowi {
3044         struct flowi6 fl6;
3045         struct in6_addr gateway;
3046 };
3047
3048 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3049                                              struct fib6_table *table,
3050                                              struct flowi6 *fl6,
3051                                              const struct sk_buff *skb,
3052                                              int flags)
3053 {
3054         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3055         struct rt6_info *ret = NULL;
3056         struct fib6_result res = {};
3057         struct fib6_nh_rd_arg arg = {
3058                 .res = &res,
3059                 .fl6 = fl6,
3060                 .gw  = &rdfl->gateway,
3061                 .ret = &ret
3062         };
3063         struct fib6_info *rt;
3064         struct fib6_node *fn;
3065
3066         /* Get the "current" route for this destination and
3067          * check if the redirect has come from appropriate router.
3068          *
3069          * RFC 4861 specifies that redirects should only be
3070          * accepted if they come from the nexthop to the target.
3071          * Due to the way the routes are chosen, this notion
3072          * is a bit fuzzy and one might need to check all possible
3073          * routes.
3074          */
3075
3076         rcu_read_lock();
3077         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3078 restart:
3079         for_each_fib6_node_rt_rcu(fn) {
3080                 res.f6i = rt;
3081                 if (fib6_check_expired(rt))
3082                         continue;
3083                 if (rt->fib6_flags & RTF_REJECT)
3084                         break;
3085                 if (unlikely(rt->nh)) {
3086                         if (nexthop_is_blackhole(rt->nh))
3087                                 continue;
3088                         /* on match, res->nh is filled in and potentially ret */
3089                         if (nexthop_for_each_fib6_nh(rt->nh,
3090                                                      fib6_nh_redirect_match,
3091                                                      &arg))
3092                                 goto out;
3093                 } else {
3094                         res.nh = rt->fib6_nh;
3095                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3096                                                   &ret))
3097                                 goto out;
3098                 }
3099         }
3100
3101         if (!rt)
3102                 rt = net->ipv6.fib6_null_entry;
3103         else if (rt->fib6_flags & RTF_REJECT) {
3104                 ret = net->ipv6.ip6_null_entry;
3105                 goto out;
3106         }
3107
3108         if (rt == net->ipv6.fib6_null_entry) {
3109                 fn = fib6_backtrack(fn, &fl6->saddr);
3110                 if (fn)
3111                         goto restart;
3112         }
3113
3114         res.f6i = rt;
3115         res.nh = rt->fib6_nh;
3116 out:
3117         if (ret) {
3118                 ip6_hold_safe(net, &ret);
3119         } else {
3120                 res.fib6_flags = res.f6i->fib6_flags;
3121                 res.fib6_type = res.f6i->fib6_type;
3122                 ret = ip6_create_rt_rcu(&res);
3123         }
3124
3125         rcu_read_unlock();
3126
3127         trace_fib6_table_lookup(net, &res, table, fl6);
3128         return ret;
3129 };
3130
3131 static struct dst_entry *ip6_route_redirect(struct net *net,
3132                                             const struct flowi6 *fl6,
3133                                             const struct sk_buff *skb,
3134                                             const struct in6_addr *gateway)
3135 {
3136         int flags = RT6_LOOKUP_F_HAS_SADDR;
3137         struct ip6rd_flowi rdfl;
3138
3139         rdfl.fl6 = *fl6;
3140         rdfl.gateway = *gateway;
3141
3142         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3143                                 flags, __ip6_route_redirect);
3144 }
3145
3146 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3147                   kuid_t uid)
3148 {
3149         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3150         struct dst_entry *dst;
3151         struct flowi6 fl6 = {
3152                 .flowi6_iif = LOOPBACK_IFINDEX,
3153                 .flowi6_oif = oif,
3154                 .flowi6_mark = mark,
3155                 .daddr = iph->daddr,
3156                 .saddr = iph->saddr,
3157                 .flowlabel = ip6_flowinfo(iph),
3158                 .flowi6_uid = uid,
3159         };
3160
3161         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3162         rt6_do_redirect(dst, NULL, skb);
3163         dst_release(dst);
3164 }
3165 EXPORT_SYMBOL_GPL(ip6_redirect);
3166
3167 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3168 {
3169         const struct ipv6hdr *iph = ipv6_hdr(skb);
3170         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3171         struct dst_entry *dst;
3172         struct flowi6 fl6 = {
3173                 .flowi6_iif = LOOPBACK_IFINDEX,
3174                 .flowi6_oif = oif,
3175                 .daddr = msg->dest,
3176                 .saddr = iph->daddr,
3177                 .flowi6_uid = sock_net_uid(net, NULL),
3178         };
3179
3180         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3181         rt6_do_redirect(dst, NULL, skb);
3182         dst_release(dst);
3183 }
3184
3185 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3186 {
3187         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if,
3188                      READ_ONCE(sk->sk_mark), sk->sk_uid);
3189 }
3190 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3191
3192 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3193 {
3194         struct net_device *dev = dst->dev;
3195         unsigned int mtu = dst_mtu(dst);
3196         struct net *net = dev_net(dev);
3197
3198         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3199
3200         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3201                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3202
3203         /*
3204          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3205          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3206          * IPV6_MAXPLEN is also valid and means: "any MSS,
3207          * rely only on pmtu discovery"
3208          */
3209         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3210                 mtu = IPV6_MAXPLEN;
3211         return mtu;
3212 }
3213
3214 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3215 {
3216         return ip6_dst_mtu_maybe_forward(dst, false);
3217 }
3218 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3219
3220 /* MTU selection:
3221  * 1. mtu on route is locked - use it
3222  * 2. mtu from nexthop exception
3223  * 3. mtu from egress device
3224  *
3225  * based on ip6_dst_mtu_forward and exception logic of
3226  * rt6_find_cached_rt; called with rcu_read_lock
3227  */
3228 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3229                       const struct in6_addr *daddr,
3230                       const struct in6_addr *saddr)
3231 {
3232         const struct fib6_nh *nh = res->nh;
3233         struct fib6_info *f6i = res->f6i;
3234         struct inet6_dev *idev;
3235         struct rt6_info *rt;
3236         u32 mtu = 0;
3237
3238         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3239                 mtu = f6i->fib6_pmtu;
3240                 if (mtu)
3241                         goto out;
3242         }
3243
3244         rt = rt6_find_cached_rt(res, daddr, saddr);
3245         if (unlikely(rt)) {
3246                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3247         } else {
3248                 struct net_device *dev = nh->fib_nh_dev;
3249
3250                 mtu = IPV6_MIN_MTU;
3251                 idev = __in6_dev_get(dev);
3252                 if (idev && idev->cnf.mtu6 > mtu)
3253                         mtu = idev->cnf.mtu6;
3254         }
3255
3256         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3257 out:
3258         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3259 }
3260
3261 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3262                                   struct flowi6 *fl6)
3263 {
3264         struct dst_entry *dst;
3265         struct rt6_info *rt;
3266         struct inet6_dev *idev = in6_dev_get(dev);
3267         struct net *net = dev_net(dev);
3268
3269         if (unlikely(!idev))
3270                 return ERR_PTR(-ENODEV);
3271
3272         rt = ip6_dst_alloc(net, dev, 0);
3273         if (unlikely(!rt)) {
3274                 in6_dev_put(idev);
3275                 dst = ERR_PTR(-ENOMEM);
3276                 goto out;
3277         }
3278
3279         rt->dst.input = ip6_input;
3280         rt->dst.output  = ip6_output;
3281         rt->rt6i_gateway  = fl6->daddr;
3282         rt->rt6i_dst.addr = fl6->daddr;
3283         rt->rt6i_dst.plen = 128;
3284         rt->rt6i_idev     = idev;
3285         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3286
3287         /* Add this dst into uncached_list so that rt6_disable_ip() can
3288          * do proper release of the net_device
3289          */
3290         rt6_uncached_list_add(rt);
3291
3292         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3293
3294 out:
3295         return dst;
3296 }
3297
3298 static void ip6_dst_gc(struct dst_ops *ops)
3299 {
3300         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3301         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3302         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3303         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3304         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3305         unsigned int val;
3306         int entries;
3307
3308         if (time_after(rt_last_gc + rt_min_interval, jiffies))
3309                 goto out;
3310
3311         fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3312         entries = dst_entries_get_slow(ops);
3313         if (entries < ops->gc_thresh)
3314                 atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3315 out:
3316         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3317         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3318 }
3319
3320 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3321                                const struct in6_addr *gw_addr, u32 tbid,
3322                                int flags, struct fib6_result *res)
3323 {
3324         struct flowi6 fl6 = {
3325                 .flowi6_oif = cfg->fc_ifindex,
3326                 .daddr = *gw_addr,
3327                 .saddr = cfg->fc_prefsrc,
3328         };
3329         struct fib6_table *table;
3330         int err;
3331
3332         table = fib6_get_table(net, tbid);
3333         if (!table)
3334                 return -EINVAL;
3335
3336         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3337                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3338
3339         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3340
3341         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3342         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3343                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3344                                  cfg->fc_ifindex != 0, NULL, flags);
3345
3346         return err;
3347 }
3348
3349 static int ip6_route_check_nh_onlink(struct net *net,
3350                                      struct fib6_config *cfg,
3351                                      const struct net_device *dev,
3352                                      struct netlink_ext_ack *extack)
3353 {
3354         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3355         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3356         struct fib6_result res = {};
3357         int err;
3358
3359         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3360         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3361             /* ignore match if it is the default route */
3362             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3363             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3364                 NL_SET_ERR_MSG(extack,
3365                                "Nexthop has invalid gateway or device mismatch");
3366                 err = -EINVAL;
3367         }
3368
3369         return err;
3370 }
3371
3372 static int ip6_route_check_nh(struct net *net,
3373                               struct fib6_config *cfg,
3374                               struct net_device **_dev,
3375                               netdevice_tracker *dev_tracker,
3376                               struct inet6_dev **idev)
3377 {
3378         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3379         struct net_device *dev = _dev ? *_dev : NULL;
3380         int flags = RT6_LOOKUP_F_IFACE;
3381         struct fib6_result res = {};
3382         int err = -EHOSTUNREACH;
3383
3384         if (cfg->fc_table) {
3385                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3386                                           cfg->fc_table, flags, &res);
3387                 /* gw_addr can not require a gateway or resolve to a reject
3388                  * route. If a device is given, it must match the result.
3389                  */
3390                 if (err || res.fib6_flags & RTF_REJECT ||
3391                     res.nh->fib_nh_gw_family ||
3392                     (dev && dev != res.nh->fib_nh_dev))
3393                         err = -EHOSTUNREACH;
3394         }
3395
3396         if (err < 0) {
3397                 struct flowi6 fl6 = {
3398                         .flowi6_oif = cfg->fc_ifindex,
3399                         .daddr = *gw_addr,
3400                 };
3401
3402                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3403                 if (err || res.fib6_flags & RTF_REJECT ||
3404                     res.nh->fib_nh_gw_family)
3405                         err = -EHOSTUNREACH;
3406
3407                 if (err)
3408                         return err;
3409
3410                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3411                                  cfg->fc_ifindex != 0, NULL, flags);
3412         }
3413
3414         err = 0;
3415         if (dev) {
3416                 if (dev != res.nh->fib_nh_dev)
3417                         err = -EHOSTUNREACH;
3418         } else {
3419                 *_dev = dev = res.nh->fib_nh_dev;
3420                 netdev_hold(dev, dev_tracker, GFP_ATOMIC);
3421                 *idev = in6_dev_get(dev);
3422         }
3423
3424         return err;
3425 }
3426
3427 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3428                            struct net_device **_dev,
3429                            netdevice_tracker *dev_tracker,
3430                            struct inet6_dev **idev,
3431                            struct netlink_ext_ack *extack)
3432 {
3433         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3434         int gwa_type = ipv6_addr_type(gw_addr);
3435         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3436         const struct net_device *dev = *_dev;
3437         bool need_addr_check = !dev;
3438         int err = -EINVAL;
3439
3440         /* if gw_addr is local we will fail to detect this in case
3441          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3442          * will return already-added prefix route via interface that
3443          * prefix route was assigned to, which might be non-loopback.
3444          */
3445         if (dev &&
3446             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3447                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3448                 goto out;
3449         }
3450
3451         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3452                 /* IPv6 strictly inhibits using not link-local
3453                  * addresses as nexthop address.
3454                  * Otherwise, router will not able to send redirects.
3455                  * It is very good, but in some (rare!) circumstances
3456                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3457                  * some exceptions. --ANK
3458                  * We allow IPv4-mapped nexthops to support RFC4798-type
3459                  * addressing
3460                  */
3461                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3462                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3463                         goto out;
3464                 }
3465
3466                 rcu_read_lock();
3467
3468                 if (cfg->fc_flags & RTNH_F_ONLINK)
3469                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3470                 else
3471                         err = ip6_route_check_nh(net, cfg, _dev, dev_tracker,
3472                                                  idev);
3473
3474                 rcu_read_unlock();
3475
3476                 if (err)
3477                         goto out;
3478         }
3479
3480         /* reload in case device was changed */
3481         dev = *_dev;
3482
3483         err = -EINVAL;
3484         if (!dev) {
3485                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3486                 goto out;
3487         } else if (dev->flags & IFF_LOOPBACK) {
3488                 NL_SET_ERR_MSG(extack,
3489                                "Egress device can not be loopback device for this route");
3490                 goto out;
3491         }
3492
3493         /* if we did not check gw_addr above, do so now that the
3494          * egress device has been resolved.
3495          */
3496         if (need_addr_check &&
3497             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3498                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3499                 goto out;
3500         }
3501
3502         err = 0;
3503 out:
3504         return err;
3505 }
3506
3507 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3508 {
3509         if ((flags & RTF_REJECT) ||
3510             (dev && (dev->flags & IFF_LOOPBACK) &&
3511              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3512              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3513                 return true;
3514
3515         return false;
3516 }
3517
3518 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3519                  struct fib6_config *cfg, gfp_t gfp_flags,
3520                  struct netlink_ext_ack *extack)
3521 {
3522         netdevice_tracker *dev_tracker = &fib6_nh->fib_nh_dev_tracker;
3523         struct net_device *dev = NULL;
3524         struct inet6_dev *idev = NULL;
3525         int addr_type;
3526         int err;
3527
3528         fib6_nh->fib_nh_family = AF_INET6;
3529 #ifdef CONFIG_IPV6_ROUTER_PREF
3530         fib6_nh->last_probe = jiffies;
3531 #endif
3532         if (cfg->fc_is_fdb) {
3533                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3534                 fib6_nh->fib_nh_gw_family = AF_INET6;
3535                 return 0;
3536         }
3537
3538         err = -ENODEV;
3539         if (cfg->fc_ifindex) {
3540                 dev = netdev_get_by_index(net, cfg->fc_ifindex,
3541                                           dev_tracker, gfp_flags);
3542                 if (!dev)
3543                         goto out;
3544                 idev = in6_dev_get(dev);
3545                 if (!idev)
3546                         goto out;
3547         }
3548
3549         if (cfg->fc_flags & RTNH_F_ONLINK) {
3550                 if (!dev) {
3551                         NL_SET_ERR_MSG(extack,
3552                                        "Nexthop device required for onlink");
3553                         goto out;
3554                 }
3555
3556                 if (!(dev->flags & IFF_UP)) {
3557                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3558                         err = -ENETDOWN;
3559                         goto out;
3560                 }
3561
3562                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3563         }
3564
3565         fib6_nh->fib_nh_weight = 1;
3566
3567         /* We cannot add true routes via loopback here,
3568          * they would result in kernel looping; promote them to reject routes
3569          */
3570         addr_type = ipv6_addr_type(&cfg->fc_dst);
3571         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3572                 /* hold loopback dev/idev if we haven't done so. */
3573                 if (dev != net->loopback_dev) {
3574                         if (dev) {
3575                                 netdev_put(dev, dev_tracker);
3576                                 in6_dev_put(idev);
3577                         }
3578                         dev = net->loopback_dev;
3579                         netdev_hold(dev, dev_tracker, gfp_flags);
3580                         idev = in6_dev_get(dev);
3581                         if (!idev) {
3582                                 err = -ENODEV;
3583                                 goto out;
3584                         }
3585                 }
3586                 goto pcpu_alloc;
3587         }
3588
3589         if (cfg->fc_flags & RTF_GATEWAY) {
3590                 err = ip6_validate_gw(net, cfg, &dev, dev_tracker,
3591                                       &idev, extack);
3592                 if (err)
3593                         goto out;
3594
3595                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3596                 fib6_nh->fib_nh_gw_family = AF_INET6;
3597         }
3598
3599         err = -ENODEV;
3600         if (!dev)
3601                 goto out;
3602
3603         if (idev->cnf.disable_ipv6) {
3604                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3605                 err = -EACCES;
3606                 goto out;
3607         }
3608
3609         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3610                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3611                 err = -ENETDOWN;
3612                 goto out;
3613         }
3614
3615         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3616             !netif_carrier_ok(dev))
3617                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3618
3619         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3620                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3621         if (err)
3622                 goto out;
3623
3624 pcpu_alloc:
3625         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3626         if (!fib6_nh->rt6i_pcpu) {
3627                 err = -ENOMEM;
3628                 goto out;
3629         }
3630
3631         fib6_nh->fib_nh_dev = dev;
3632         fib6_nh->fib_nh_oif = dev->ifindex;
3633         err = 0;
3634 out:
3635         if (idev)
3636                 in6_dev_put(idev);
3637
3638         if (err) {
3639                 lwtstate_put(fib6_nh->fib_nh_lws);
3640                 fib6_nh->fib_nh_lws = NULL;
3641                 netdev_put(dev, dev_tracker);
3642         }
3643
3644         return err;
3645 }
3646
3647 void fib6_nh_release(struct fib6_nh *fib6_nh)
3648 {
3649         struct rt6_exception_bucket *bucket;
3650
3651         rcu_read_lock();
3652
3653         fib6_nh_flush_exceptions(fib6_nh, NULL);
3654         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3655         if (bucket) {
3656                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3657                 kfree(bucket);
3658         }
3659
3660         rcu_read_unlock();
3661
3662         fib6_nh_release_dsts(fib6_nh);
3663         free_percpu(fib6_nh->rt6i_pcpu);
3664
3665         fib_nh_common_release(&fib6_nh->nh_common);
3666 }
3667
3668 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3669 {
3670         int cpu;
3671
3672         if (!fib6_nh->rt6i_pcpu)
3673                 return;
3674
3675         for_each_possible_cpu(cpu) {
3676                 struct rt6_info *pcpu_rt, **ppcpu_rt;
3677
3678                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3679                 pcpu_rt = xchg(ppcpu_rt, NULL);
3680                 if (pcpu_rt) {
3681                         dst_dev_put(&pcpu_rt->dst);
3682                         dst_release(&pcpu_rt->dst);
3683                 }
3684         }
3685 }
3686
3687 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3688                                               gfp_t gfp_flags,
3689                                               struct netlink_ext_ack *extack)
3690 {
3691         struct net *net = cfg->fc_nlinfo.nl_net;
3692         struct fib6_info *rt = NULL;
3693         struct nexthop *nh = NULL;
3694         struct fib6_table *table;
3695         struct fib6_nh *fib6_nh;
3696         int err = -EINVAL;
3697         int addr_type;
3698
3699         /* RTF_PCPU is an internal flag; can not be set by userspace */
3700         if (cfg->fc_flags & RTF_PCPU) {
3701                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3702                 goto out;
3703         }
3704
3705         /* RTF_CACHE is an internal flag; can not be set by userspace */
3706         if (cfg->fc_flags & RTF_CACHE) {
3707                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3708                 goto out;
3709         }
3710
3711         if (cfg->fc_type > RTN_MAX) {
3712                 NL_SET_ERR_MSG(extack, "Invalid route type");
3713                 goto out;
3714         }
3715
3716         if (cfg->fc_dst_len > 128) {
3717                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3718                 goto out;
3719         }
3720         if (cfg->fc_src_len > 128) {
3721                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3722                 goto out;
3723         }
3724 #ifndef CONFIG_IPV6_SUBTREES
3725         if (cfg->fc_src_len) {
3726                 NL_SET_ERR_MSG(extack,
3727                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3728                 goto out;
3729         }
3730 #endif
3731         if (cfg->fc_nh_id) {
3732                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3733                 if (!nh) {
3734                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3735                         goto out;
3736                 }
3737                 err = fib6_check_nexthop(nh, cfg, extack);
3738                 if (err)
3739                         goto out;
3740         }
3741
3742         err = -ENOBUFS;
3743         if (cfg->fc_nlinfo.nlh &&
3744             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3745                 table = fib6_get_table(net, cfg->fc_table);
3746                 if (!table) {
3747                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3748                         table = fib6_new_table(net, cfg->fc_table);
3749                 }
3750         } else {
3751                 table = fib6_new_table(net, cfg->fc_table);
3752         }
3753
3754         if (!table)
3755                 goto out;
3756
3757         err = -ENOMEM;
3758         rt = fib6_info_alloc(gfp_flags, !nh);
3759         if (!rt)
3760                 goto out;
3761
3762         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3763                                                extack);
3764         if (IS_ERR(rt->fib6_metrics)) {
3765                 err = PTR_ERR(rt->fib6_metrics);
3766                 /* Do not leave garbage there. */
3767                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3768                 goto out_free;
3769         }
3770
3771         if (cfg->fc_flags & RTF_ADDRCONF)
3772                 rt->dst_nocount = true;
3773
3774         if (cfg->fc_flags & RTF_EXPIRES)
3775                 fib6_set_expires(rt, jiffies +
3776                                 clock_t_to_jiffies(cfg->fc_expires));
3777
3778         if (cfg->fc_protocol == RTPROT_UNSPEC)
3779                 cfg->fc_protocol = RTPROT_BOOT;
3780         rt->fib6_protocol = cfg->fc_protocol;
3781
3782         rt->fib6_table = table;
3783         rt->fib6_metric = cfg->fc_metric;
3784         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3785         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3786
3787         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3788         rt->fib6_dst.plen = cfg->fc_dst_len;
3789
3790 #ifdef CONFIG_IPV6_SUBTREES
3791         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3792         rt->fib6_src.plen = cfg->fc_src_len;
3793 #endif
3794         if (nh) {
3795                 if (rt->fib6_src.plen) {
3796                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3797                         goto out_free;
3798                 }
3799                 if (!nexthop_get(nh)) {
3800                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3801                         goto out_free;
3802                 }
3803                 rt->nh = nh;
3804                 fib6_nh = nexthop_fib6_nh(rt->nh);
3805         } else {
3806                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3807                 if (err)
3808                         goto out;
3809
3810                 fib6_nh = rt->fib6_nh;
3811
3812                 /* We cannot add true routes via loopback here, they would
3813                  * result in kernel looping; promote them to reject routes
3814                  */
3815                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3816                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3817                                    addr_type))
3818                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3819         }
3820
3821         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3822                 struct net_device *dev = fib6_nh->fib_nh_dev;
3823
3824                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3825                         NL_SET_ERR_MSG(extack, "Invalid source address");
3826                         err = -EINVAL;
3827                         goto out;
3828                 }
3829                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3830                 rt->fib6_prefsrc.plen = 128;
3831         } else
3832                 rt->fib6_prefsrc.plen = 0;
3833
3834         return rt;
3835 out:
3836         fib6_info_release(rt);
3837         return ERR_PTR(err);
3838 out_free:
3839         ip_fib_metrics_put(rt->fib6_metrics);
3840         kfree(rt);
3841         return ERR_PTR(err);
3842 }
3843
3844 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3845                   struct netlink_ext_ack *extack)
3846 {
3847         struct fib6_info *rt;
3848         int err;
3849
3850         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3851         if (IS_ERR(rt))
3852                 return PTR_ERR(rt);
3853
3854         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3855         fib6_info_release(rt);
3856
3857         return err;
3858 }
3859
3860 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3861 {
3862         struct net *net = info->nl_net;
3863         struct fib6_table *table;
3864         int err;
3865
3866         if (rt == net->ipv6.fib6_null_entry) {
3867                 err = -ENOENT;
3868                 goto out;
3869         }
3870
3871         table = rt->fib6_table;
3872         spin_lock_bh(&table->tb6_lock);
3873         err = fib6_del(rt, info);
3874         spin_unlock_bh(&table->tb6_lock);
3875
3876 out:
3877         fib6_info_release(rt);
3878         return err;
3879 }
3880
3881 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3882 {
3883         struct nl_info info = {
3884                 .nl_net = net,
3885                 .skip_notify = skip_notify
3886         };
3887
3888         return __ip6_del_rt(rt, &info);
3889 }
3890
3891 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3892 {
3893         struct nl_info *info = &cfg->fc_nlinfo;
3894         struct net *net = info->nl_net;
3895         struct sk_buff *skb = NULL;
3896         struct fib6_table *table;
3897         int err = -ENOENT;
3898
3899         if (rt == net->ipv6.fib6_null_entry)
3900                 goto out_put;
3901         table = rt->fib6_table;
3902         spin_lock_bh(&table->tb6_lock);
3903
3904         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3905                 struct fib6_info *sibling, *next_sibling;
3906                 struct fib6_node *fn;
3907
3908                 /* prefer to send a single notification with all hops */
3909                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3910                 if (skb) {
3911                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3912
3913                         if (rt6_fill_node(net, skb, rt, NULL,
3914                                           NULL, NULL, 0, RTM_DELROUTE,
3915                                           info->portid, seq, 0) < 0) {
3916                                 kfree_skb(skb);
3917                                 skb = NULL;
3918                         } else
3919                                 info->skip_notify = 1;
3920                 }
3921
3922                 /* 'rt' points to the first sibling route. If it is not the
3923                  * leaf, then we do not need to send a notification. Otherwise,
3924                  * we need to check if the last sibling has a next route or not
3925                  * and emit a replace or delete notification, respectively.
3926                  */
3927                 info->skip_notify_kernel = 1;
3928                 fn = rcu_dereference_protected(rt->fib6_node,
3929                                             lockdep_is_held(&table->tb6_lock));
3930                 if (rcu_access_pointer(fn->leaf) == rt) {
3931                         struct fib6_info *last_sibling, *replace_rt;
3932
3933                         last_sibling = list_last_entry(&rt->fib6_siblings,
3934                                                        struct fib6_info,
3935                                                        fib6_siblings);
3936                         replace_rt = rcu_dereference_protected(
3937                                             last_sibling->fib6_next,
3938                                             lockdep_is_held(&table->tb6_lock));
3939                         if (replace_rt)
3940                                 call_fib6_entry_notifiers_replace(net,
3941                                                                   replace_rt);
3942                         else
3943                                 call_fib6_multipath_entry_notifiers(net,
3944                                                        FIB_EVENT_ENTRY_DEL,
3945                                                        rt, rt->fib6_nsiblings,
3946                                                        NULL);
3947                 }
3948                 list_for_each_entry_safe(sibling, next_sibling,
3949                                          &rt->fib6_siblings,
3950                                          fib6_siblings) {
3951                         err = fib6_del(sibling, info);
3952                         if (err)
3953                                 goto out_unlock;
3954                 }
3955         }
3956
3957         err = fib6_del(rt, info);
3958 out_unlock:
3959         spin_unlock_bh(&table->tb6_lock);
3960 out_put:
3961         fib6_info_release(rt);
3962
3963         if (skb) {
3964                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3965                             info->nlh, gfp_any());
3966         }
3967         return err;
3968 }
3969
3970 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3971 {
3972         int rc = -ESRCH;
3973
3974         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3975                 goto out;
3976
3977         if (cfg->fc_flags & RTF_GATEWAY &&
3978             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3979                 goto out;
3980
3981         rc = rt6_remove_exception_rt(rt);
3982 out:
3983         return rc;
3984 }
3985
3986 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3987                              struct fib6_nh *nh)
3988 {
3989         struct fib6_result res = {
3990                 .f6i = rt,
3991                 .nh = nh,
3992         };
3993         struct rt6_info *rt_cache;
3994
3995         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3996         if (rt_cache)
3997                 return __ip6_del_cached_rt(rt_cache, cfg);
3998
3999         return 0;
4000 }
4001
4002 struct fib6_nh_del_cached_rt_arg {
4003         struct fib6_config *cfg;
4004         struct fib6_info *f6i;
4005 };
4006
4007 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4008 {
4009         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4010         int rc;
4011
4012         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4013         return rc != -ESRCH ? rc : 0;
4014 }
4015
4016 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4017 {
4018         struct fib6_nh_del_cached_rt_arg arg = {
4019                 .cfg = cfg,
4020                 .f6i = f6i
4021         };
4022
4023         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4024 }
4025
4026 static int ip6_route_del(struct fib6_config *cfg,
4027                          struct netlink_ext_ack *extack)
4028 {
4029         struct fib6_table *table;
4030         struct fib6_info *rt;
4031         struct fib6_node *fn;
4032         int err = -ESRCH;
4033
4034         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4035         if (!table) {
4036                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4037                 return err;
4038         }
4039
4040         rcu_read_lock();
4041
4042         fn = fib6_locate(&table->tb6_root,
4043                          &cfg->fc_dst, cfg->fc_dst_len,
4044                          &cfg->fc_src, cfg->fc_src_len,
4045                          !(cfg->fc_flags & RTF_CACHE));
4046
4047         if (fn) {
4048                 for_each_fib6_node_rt_rcu(fn) {
4049                         struct fib6_nh *nh;
4050
4051                         if (rt->nh && cfg->fc_nh_id &&
4052                             rt->nh->id != cfg->fc_nh_id)
4053                                 continue;
4054
4055                         if (cfg->fc_flags & RTF_CACHE) {
4056                                 int rc = 0;
4057
4058                                 if (rt->nh) {
4059                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4060                                 } else if (cfg->fc_nh_id) {
4061                                         continue;
4062                                 } else {
4063                                         nh = rt->fib6_nh;
4064                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4065                                 }
4066                                 if (rc != -ESRCH) {
4067                                         rcu_read_unlock();
4068                                         return rc;
4069                                 }
4070                                 continue;
4071                         }
4072
4073                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4074                                 continue;
4075                         if (cfg->fc_protocol &&
4076                             cfg->fc_protocol != rt->fib6_protocol)
4077                                 continue;
4078
4079                         if (rt->nh) {
4080                                 if (!fib6_info_hold_safe(rt))
4081                                         continue;
4082                                 rcu_read_unlock();
4083
4084                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4085                         }
4086                         if (cfg->fc_nh_id)
4087                                 continue;
4088
4089                         nh = rt->fib6_nh;
4090                         if (cfg->fc_ifindex &&
4091                             (!nh->fib_nh_dev ||
4092                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4093                                 continue;
4094                         if (cfg->fc_flags & RTF_GATEWAY &&
4095                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4096                                 continue;
4097                         if (!fib6_info_hold_safe(rt))
4098                                 continue;
4099                         rcu_read_unlock();
4100
4101                         /* if gateway was specified only delete the one hop */
4102                         if (cfg->fc_flags & RTF_GATEWAY)
4103                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4104
4105                         return __ip6_del_rt_siblings(rt, cfg);
4106                 }
4107         }
4108         rcu_read_unlock();
4109
4110         return err;
4111 }
4112
4113 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4114 {
4115         struct netevent_redirect netevent;
4116         struct rt6_info *rt, *nrt = NULL;
4117         struct fib6_result res = {};
4118         struct ndisc_options ndopts;
4119         struct inet6_dev *in6_dev;
4120         struct neighbour *neigh;
4121         struct rd_msg *msg;
4122         int optlen, on_link;
4123         u8 *lladdr;
4124
4125         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4126         optlen -= sizeof(*msg);
4127
4128         if (optlen < 0) {
4129                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4130                 return;
4131         }
4132
4133         msg = (struct rd_msg *)icmp6_hdr(skb);
4134
4135         if (ipv6_addr_is_multicast(&msg->dest)) {
4136                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4137                 return;
4138         }
4139
4140         on_link = 0;
4141         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4142                 on_link = 1;
4143         } else if (ipv6_addr_type(&msg->target) !=
4144                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4145                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4146                 return;
4147         }
4148
4149         in6_dev = __in6_dev_get(skb->dev);
4150         if (!in6_dev)
4151                 return;
4152         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4153                 return;
4154
4155         /* RFC2461 8.1:
4156          *      The IP source address of the Redirect MUST be the same as the current
4157          *      first-hop router for the specified ICMP Destination Address.
4158          */
4159
4160         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4161                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4162                 return;
4163         }
4164
4165         lladdr = NULL;
4166         if (ndopts.nd_opts_tgt_lladdr) {
4167                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4168                                              skb->dev);
4169                 if (!lladdr) {
4170                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4171                         return;
4172                 }
4173         }
4174
4175         rt = (struct rt6_info *) dst;
4176         if (rt->rt6i_flags & RTF_REJECT) {
4177                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4178                 return;
4179         }
4180
4181         /* Redirect received -> path was valid.
4182          * Look, redirects are sent only in response to data packets,
4183          * so that this nexthop apparently is reachable. --ANK
4184          */
4185         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4186
4187         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4188         if (!neigh)
4189                 return;
4190
4191         /*
4192          *      We have finally decided to accept it.
4193          */
4194
4195         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4196                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4197                      NEIGH_UPDATE_F_OVERRIDE|
4198                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4199                                      NEIGH_UPDATE_F_ISROUTER)),
4200                      NDISC_REDIRECT, &ndopts);
4201
4202         rcu_read_lock();
4203         res.f6i = rcu_dereference(rt->from);
4204         if (!res.f6i)
4205                 goto out;
4206
4207         if (res.f6i->nh) {
4208                 struct fib6_nh_match_arg arg = {
4209                         .dev = dst->dev,
4210                         .gw = &rt->rt6i_gateway,
4211                 };
4212
4213                 nexthop_for_each_fib6_nh(res.f6i->nh,
4214                                          fib6_nh_find_match, &arg);
4215
4216                 /* fib6_info uses a nexthop that does not have fib6_nh
4217                  * using the dst->dev. Should be impossible
4218                  */
4219                 if (!arg.match)
4220                         goto out;
4221                 res.nh = arg.match;
4222         } else {
4223                 res.nh = res.f6i->fib6_nh;
4224         }
4225
4226         res.fib6_flags = res.f6i->fib6_flags;
4227         res.fib6_type = res.f6i->fib6_type;
4228         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4229         if (!nrt)
4230                 goto out;
4231
4232         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4233         if (on_link)
4234                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4235
4236         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4237
4238         /* rt6_insert_exception() will take care of duplicated exceptions */
4239         if (rt6_insert_exception(nrt, &res)) {
4240                 dst_release_immediate(&nrt->dst);
4241                 goto out;
4242         }
4243
4244         netevent.old = &rt->dst;
4245         netevent.new = &nrt->dst;
4246         netevent.daddr = &msg->dest;
4247         netevent.neigh = neigh;
4248         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4249
4250 out:
4251         rcu_read_unlock();
4252         neigh_release(neigh);
4253 }
4254
4255 #ifdef CONFIG_IPV6_ROUTE_INFO
4256 static struct fib6_info *rt6_get_route_info(struct net *net,
4257                                            const struct in6_addr *prefix, int prefixlen,
4258                                            const struct in6_addr *gwaddr,
4259                                            struct net_device *dev)
4260 {
4261         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4262         int ifindex = dev->ifindex;
4263         struct fib6_node *fn;
4264         struct fib6_info *rt = NULL;
4265         struct fib6_table *table;
4266
4267         table = fib6_get_table(net, tb_id);
4268         if (!table)
4269                 return NULL;
4270
4271         rcu_read_lock();
4272         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4273         if (!fn)
4274                 goto out;
4275
4276         for_each_fib6_node_rt_rcu(fn) {
4277                 /* these routes do not use nexthops */
4278                 if (rt->nh)
4279                         continue;
4280                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4281                         continue;
4282                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4283                     !rt->fib6_nh->fib_nh_gw_family)
4284                         continue;
4285                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4286                         continue;
4287                 if (!fib6_info_hold_safe(rt))
4288                         continue;
4289                 break;
4290         }
4291 out:
4292         rcu_read_unlock();
4293         return rt;
4294 }
4295
4296 static struct fib6_info *rt6_add_route_info(struct net *net,
4297                                            const struct in6_addr *prefix, int prefixlen,
4298                                            const struct in6_addr *gwaddr,
4299                                            struct net_device *dev,
4300                                            unsigned int pref)
4301 {
4302         struct fib6_config cfg = {
4303                 .fc_metric      = IP6_RT_PRIO_USER,
4304                 .fc_ifindex     = dev->ifindex,
4305                 .fc_dst_len     = prefixlen,
4306                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4307                                   RTF_UP | RTF_PREF(pref),
4308                 .fc_protocol = RTPROT_RA,
4309                 .fc_type = RTN_UNICAST,
4310                 .fc_nlinfo.portid = 0,
4311                 .fc_nlinfo.nlh = NULL,
4312                 .fc_nlinfo.nl_net = net,
4313         };
4314
4315         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4316         cfg.fc_dst = *prefix;
4317         cfg.fc_gateway = *gwaddr;
4318
4319         /* We should treat it as a default route if prefix length is 0. */
4320         if (!prefixlen)
4321                 cfg.fc_flags |= RTF_DEFAULT;
4322
4323         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4324
4325         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4326 }
4327 #endif
4328
4329 struct fib6_info *rt6_get_dflt_router(struct net *net,
4330                                      const struct in6_addr *addr,
4331                                      struct net_device *dev)
4332 {
4333         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4334         struct fib6_info *rt;
4335         struct fib6_table *table;
4336
4337         table = fib6_get_table(net, tb_id);
4338         if (!table)
4339                 return NULL;
4340
4341         rcu_read_lock();
4342         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4343                 struct fib6_nh *nh;
4344
4345                 /* RA routes do not use nexthops */
4346                 if (rt->nh)
4347                         continue;
4348
4349                 nh = rt->fib6_nh;
4350                 if (dev == nh->fib_nh_dev &&
4351                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4352                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4353                         break;
4354         }
4355         if (rt && !fib6_info_hold_safe(rt))
4356                 rt = NULL;
4357         rcu_read_unlock();
4358         return rt;
4359 }
4360
4361 struct fib6_info *rt6_add_dflt_router(struct net *net,
4362                                      const struct in6_addr *gwaddr,
4363                                      struct net_device *dev,
4364                                      unsigned int pref,
4365                                      u32 defrtr_usr_metric,
4366                                      int lifetime)
4367 {
4368         struct fib6_config cfg = {
4369                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4370                 .fc_metric      = defrtr_usr_metric,
4371                 .fc_ifindex     = dev->ifindex,
4372                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4373                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4374                 .fc_protocol = RTPROT_RA,
4375                 .fc_type = RTN_UNICAST,
4376                 .fc_nlinfo.portid = 0,
4377                 .fc_nlinfo.nlh = NULL,
4378                 .fc_nlinfo.nl_net = net,
4379                 .fc_expires = jiffies_to_clock_t(lifetime * HZ),
4380         };
4381
4382         cfg.fc_gateway = *gwaddr;
4383
4384         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4385                 struct fib6_table *table;
4386
4387                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4388                 if (table)
4389                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4390         }
4391
4392         return rt6_get_dflt_router(net, gwaddr, dev);
4393 }
4394
4395 static void __rt6_purge_dflt_routers(struct net *net,
4396                                      struct fib6_table *table)
4397 {
4398         struct fib6_info *rt;
4399
4400 restart:
4401         rcu_read_lock();
4402         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4403                 struct net_device *dev = fib6_info_nh_dev(rt);
4404                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4405
4406                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4407                     (!idev || idev->cnf.accept_ra != 2) &&
4408                     fib6_info_hold_safe(rt)) {
4409                         rcu_read_unlock();
4410                         ip6_del_rt(net, rt, false);
4411                         goto restart;
4412                 }
4413         }
4414         rcu_read_unlock();
4415
4416         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4417 }
4418
4419 void rt6_purge_dflt_routers(struct net *net)
4420 {
4421         struct fib6_table *table;
4422         struct hlist_head *head;
4423         unsigned int h;
4424
4425         rcu_read_lock();
4426
4427         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4428                 head = &net->ipv6.fib_table_hash[h];
4429                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4430                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4431                                 __rt6_purge_dflt_routers(net, table);
4432                 }
4433         }
4434
4435         rcu_read_unlock();
4436 }
4437
4438 static void rtmsg_to_fib6_config(struct net *net,
4439                                  struct in6_rtmsg *rtmsg,
4440                                  struct fib6_config *cfg)
4441 {
4442         *cfg = (struct fib6_config){
4443                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4444                          : RT6_TABLE_MAIN,
4445                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4446                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4447                 .fc_expires = rtmsg->rtmsg_info,
4448                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4449                 .fc_src_len = rtmsg->rtmsg_src_len,
4450                 .fc_flags = rtmsg->rtmsg_flags,
4451                 .fc_type = rtmsg->rtmsg_type,
4452
4453                 .fc_nlinfo.nl_net = net,
4454
4455                 .fc_dst = rtmsg->rtmsg_dst,
4456                 .fc_src = rtmsg->rtmsg_src,
4457                 .fc_gateway = rtmsg->rtmsg_gateway,
4458         };
4459 }
4460
4461 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4462 {
4463         struct fib6_config cfg;
4464         int err;
4465
4466         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4467                 return -EINVAL;
4468         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4469                 return -EPERM;
4470
4471         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4472
4473         rtnl_lock();
4474         switch (cmd) {
4475         case SIOCADDRT:
4476                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4477                 break;
4478         case SIOCDELRT:
4479                 err = ip6_route_del(&cfg, NULL);
4480                 break;
4481         }
4482         rtnl_unlock();
4483         return err;
4484 }
4485
4486 /*
4487  *      Drop the packet on the floor
4488  */
4489
4490 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4491 {
4492         struct dst_entry *dst = skb_dst(skb);
4493         struct net *net = dev_net(dst->dev);
4494         struct inet6_dev *idev;
4495         SKB_DR(reason);
4496         int type;
4497
4498         if (netif_is_l3_master(skb->dev) ||
4499             dst->dev == net->loopback_dev)
4500                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4501         else
4502                 idev = ip6_dst_idev(dst);
4503
4504         switch (ipstats_mib_noroutes) {
4505         case IPSTATS_MIB_INNOROUTES:
4506                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4507                 if (type == IPV6_ADDR_ANY) {
4508                         SKB_DR_SET(reason, IP_INADDRERRORS);
4509                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4510                         break;
4511                 }
4512                 SKB_DR_SET(reason, IP_INNOROUTES);
4513                 fallthrough;
4514         case IPSTATS_MIB_OUTNOROUTES:
4515                 SKB_DR_OR(reason, IP_OUTNOROUTES);
4516                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4517                 break;
4518         }
4519
4520         /* Start over by dropping the dst for l3mdev case */
4521         if (netif_is_l3_master(skb->dev))
4522                 skb_dst_drop(skb);
4523
4524         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4525         kfree_skb_reason(skb, reason);
4526         return 0;
4527 }
4528
4529 static int ip6_pkt_discard(struct sk_buff *skb)
4530 {
4531         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4532 }
4533
4534 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4535 {
4536         skb->dev = skb_dst(skb)->dev;
4537         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4538 }
4539
4540 static int ip6_pkt_prohibit(struct sk_buff *skb)
4541 {
4542         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4543 }
4544
4545 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4546 {
4547         skb->dev = skb_dst(skb)->dev;
4548         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4549 }
4550
4551 /*
4552  *      Allocate a dst for local (unicast / anycast) address.
4553  */
4554
4555 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4556                                      struct inet6_dev *idev,
4557                                      const struct in6_addr *addr,
4558                                      bool anycast, gfp_t gfp_flags,
4559                                      struct netlink_ext_ack *extack)
4560 {
4561         struct fib6_config cfg = {
4562                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4563                 .fc_ifindex = idev->dev->ifindex,
4564                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4565                 .fc_dst = *addr,
4566                 .fc_dst_len = 128,
4567                 .fc_protocol = RTPROT_KERNEL,
4568                 .fc_nlinfo.nl_net = net,
4569                 .fc_ignore_dev_down = true,
4570         };
4571         struct fib6_info *f6i;
4572
4573         if (anycast) {
4574                 cfg.fc_type = RTN_ANYCAST;
4575                 cfg.fc_flags |= RTF_ANYCAST;
4576         } else {
4577                 cfg.fc_type = RTN_LOCAL;
4578                 cfg.fc_flags |= RTF_LOCAL;
4579         }
4580
4581         f6i = ip6_route_info_create(&cfg, gfp_flags, extack);
4582         if (!IS_ERR(f6i)) {
4583                 f6i->dst_nocount = true;
4584
4585                 if (!anycast &&
4586                     (net->ipv6.devconf_all->disable_policy ||
4587                      idev->cnf.disable_policy))
4588                         f6i->dst_nopolicy = true;
4589         }
4590
4591         return f6i;
4592 }
4593
4594 /* remove deleted ip from prefsrc entries */
4595 struct arg_dev_net_ip {
4596         struct net *net;
4597         struct in6_addr *addr;
4598 };
4599
4600 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4601 {
4602         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4603         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4604
4605         if (!rt->nh &&
4606             rt != net->ipv6.fib6_null_entry &&
4607             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr) &&
4608             !ipv6_chk_addr(net, addr, rt->fib6_nh->fib_nh_dev, 0)) {
4609                 spin_lock_bh(&rt6_exception_lock);
4610                 /* remove prefsrc entry */
4611                 rt->fib6_prefsrc.plen = 0;
4612                 spin_unlock_bh(&rt6_exception_lock);
4613         }
4614         return 0;
4615 }
4616
4617 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4618 {
4619         struct net *net = dev_net(ifp->idev->dev);
4620         struct arg_dev_net_ip adni = {
4621                 .net = net,
4622                 .addr = &ifp->addr,
4623         };
4624         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4625 }
4626
4627 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4628
4629 /* Remove routers and update dst entries when gateway turn into host. */
4630 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4631 {
4632         struct in6_addr *gateway = (struct in6_addr *)arg;
4633         struct fib6_nh *nh;
4634
4635         /* RA routes do not use nexthops */
4636         if (rt->nh)
4637                 return 0;
4638
4639         nh = rt->fib6_nh;
4640         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4641             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4642                 return -1;
4643
4644         /* Further clean up cached routes in exception table.
4645          * This is needed because cached route may have a different
4646          * gateway than its 'parent' in the case of an ip redirect.
4647          */
4648         fib6_nh_exceptions_clean_tohost(nh, gateway);
4649
4650         return 0;
4651 }
4652
4653 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4654 {
4655         fib6_clean_all(net, fib6_clean_tohost, gateway);
4656 }
4657
4658 struct arg_netdev_event {
4659         const struct net_device *dev;
4660         union {
4661                 unsigned char nh_flags;
4662                 unsigned long event;
4663         };
4664 };
4665
4666 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4667 {
4668         struct fib6_info *iter;
4669         struct fib6_node *fn;
4670
4671         fn = rcu_dereference_protected(rt->fib6_node,
4672                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4673         iter = rcu_dereference_protected(fn->leaf,
4674                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4675         while (iter) {
4676                 if (iter->fib6_metric == rt->fib6_metric &&
4677                     rt6_qualify_for_ecmp(iter))
4678                         return iter;
4679                 iter = rcu_dereference_protected(iter->fib6_next,
4680                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4681         }
4682
4683         return NULL;
4684 }
4685
4686 /* only called for fib entries with builtin fib6_nh */
4687 static bool rt6_is_dead(const struct fib6_info *rt)
4688 {
4689         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4690             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4691              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4692                 return true;
4693
4694         return false;
4695 }
4696
4697 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4698 {
4699         struct fib6_info *iter;
4700         int total = 0;
4701
4702         if (!rt6_is_dead(rt))
4703                 total += rt->fib6_nh->fib_nh_weight;
4704
4705         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4706                 if (!rt6_is_dead(iter))
4707                         total += iter->fib6_nh->fib_nh_weight;
4708         }
4709
4710         return total;
4711 }
4712
4713 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4714 {
4715         int upper_bound = -1;
4716
4717         if (!rt6_is_dead(rt)) {
4718                 *weight += rt->fib6_nh->fib_nh_weight;
4719                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4720                                                     total) - 1;
4721         }
4722         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4723 }
4724
4725 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4726 {
4727         struct fib6_info *iter;
4728         int weight = 0;
4729
4730         rt6_upper_bound_set(rt, &weight, total);
4731
4732         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4733                 rt6_upper_bound_set(iter, &weight, total);
4734 }
4735
4736 void rt6_multipath_rebalance(struct fib6_info *rt)
4737 {
4738         struct fib6_info *first;
4739         int total;
4740
4741         /* In case the entire multipath route was marked for flushing,
4742          * then there is no need to rebalance upon the removal of every
4743          * sibling route.
4744          */
4745         if (!rt->fib6_nsiblings || rt->should_flush)
4746                 return;
4747
4748         /* During lookup routes are evaluated in order, so we need to
4749          * make sure upper bounds are assigned from the first sibling
4750          * onwards.
4751          */
4752         first = rt6_multipath_first_sibling(rt);
4753         if (WARN_ON_ONCE(!first))
4754                 return;
4755
4756         total = rt6_multipath_total_weight(first);
4757         rt6_multipath_upper_bound_set(first, total);
4758 }
4759
4760 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4761 {
4762         const struct arg_netdev_event *arg = p_arg;
4763         struct net *net = dev_net(arg->dev);
4764
4765         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4766             rt->fib6_nh->fib_nh_dev == arg->dev) {
4767                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4768                 fib6_update_sernum_upto_root(net, rt);
4769                 rt6_multipath_rebalance(rt);
4770         }
4771
4772         return 0;
4773 }
4774
4775 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4776 {
4777         struct arg_netdev_event arg = {
4778                 .dev = dev,
4779                 {
4780                         .nh_flags = nh_flags,
4781                 },
4782         };
4783
4784         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4785                 arg.nh_flags |= RTNH_F_LINKDOWN;
4786
4787         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4788 }
4789
4790 /* only called for fib entries with inline fib6_nh */
4791 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4792                                    const struct net_device *dev)
4793 {
4794         struct fib6_info *iter;
4795
4796         if (rt->fib6_nh->fib_nh_dev == dev)
4797                 return true;
4798         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4799                 if (iter->fib6_nh->fib_nh_dev == dev)
4800                         return true;
4801
4802         return false;
4803 }
4804
4805 static void rt6_multipath_flush(struct fib6_info *rt)
4806 {
4807         struct fib6_info *iter;
4808
4809         rt->should_flush = 1;
4810         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4811                 iter->should_flush = 1;
4812 }
4813
4814 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4815                                              const struct net_device *down_dev)
4816 {
4817         struct fib6_info *iter;
4818         unsigned int dead = 0;
4819
4820         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4821             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4822                 dead++;
4823         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4824                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4825                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4826                         dead++;
4827
4828         return dead;
4829 }
4830
4831 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4832                                        const struct net_device *dev,
4833                                        unsigned char nh_flags)
4834 {
4835         struct fib6_info *iter;
4836
4837         if (rt->fib6_nh->fib_nh_dev == dev)
4838                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4839         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4840                 if (iter->fib6_nh->fib_nh_dev == dev)
4841                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4842 }
4843
4844 /* called with write lock held for table with rt */
4845 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4846 {
4847         const struct arg_netdev_event *arg = p_arg;
4848         const struct net_device *dev = arg->dev;
4849         struct net *net = dev_net(dev);
4850
4851         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4852                 return 0;
4853
4854         switch (arg->event) {
4855         case NETDEV_UNREGISTER:
4856                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4857         case NETDEV_DOWN:
4858                 if (rt->should_flush)
4859                         return -1;
4860                 if (!rt->fib6_nsiblings)
4861                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4862                 if (rt6_multipath_uses_dev(rt, dev)) {
4863                         unsigned int count;
4864
4865                         count = rt6_multipath_dead_count(rt, dev);
4866                         if (rt->fib6_nsiblings + 1 == count) {
4867                                 rt6_multipath_flush(rt);
4868                                 return -1;
4869                         }
4870                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4871                                                    RTNH_F_LINKDOWN);
4872                         fib6_update_sernum(net, rt);
4873                         rt6_multipath_rebalance(rt);
4874                 }
4875                 return -2;
4876         case NETDEV_CHANGE:
4877                 if (rt->fib6_nh->fib_nh_dev != dev ||
4878                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4879                         break;
4880                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4881                 rt6_multipath_rebalance(rt);
4882                 break;
4883         }
4884
4885         return 0;
4886 }
4887
4888 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4889 {
4890         struct arg_netdev_event arg = {
4891                 .dev = dev,
4892                 {
4893                         .event = event,
4894                 },
4895         };
4896         struct net *net = dev_net(dev);
4897
4898         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4899                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4900         else
4901                 fib6_clean_all(net, fib6_ifdown, &arg);
4902 }
4903
4904 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4905 {
4906         rt6_sync_down_dev(dev, event);
4907         rt6_uncached_list_flush_dev(dev);
4908         neigh_ifdown(&nd_tbl, dev);
4909 }
4910
4911 struct rt6_mtu_change_arg {
4912         struct net_device *dev;
4913         unsigned int mtu;
4914         struct fib6_info *f6i;
4915 };
4916
4917 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4918 {
4919         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4920         struct fib6_info *f6i = arg->f6i;
4921
4922         /* For administrative MTU increase, there is no way to discover
4923          * IPv6 PMTU increase, so PMTU increase should be updated here.
4924          * Since RFC 1981 doesn't include administrative MTU increase
4925          * update PMTU increase is a MUST. (i.e. jumbo frame)
4926          */
4927         if (nh->fib_nh_dev == arg->dev) {
4928                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4929                 u32 mtu = f6i->fib6_pmtu;
4930
4931                 if (mtu >= arg->mtu ||
4932                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4933                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4934
4935                 spin_lock_bh(&rt6_exception_lock);
4936                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4937                 spin_unlock_bh(&rt6_exception_lock);
4938         }
4939
4940         return 0;
4941 }
4942
4943 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4944 {
4945         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4946         struct inet6_dev *idev;
4947
4948         /* In IPv6 pmtu discovery is not optional,
4949            so that RTAX_MTU lock cannot disable it.
4950            We still use this lock to block changes
4951            caused by addrconf/ndisc.
4952         */
4953
4954         idev = __in6_dev_get(arg->dev);
4955         if (!idev)
4956                 return 0;
4957
4958         if (fib6_metric_locked(f6i, RTAX_MTU))
4959                 return 0;
4960
4961         arg->f6i = f6i;
4962         if (f6i->nh) {
4963                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4964                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4965                                                 arg);
4966         }
4967
4968         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4969 }
4970
4971 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4972 {
4973         struct rt6_mtu_change_arg arg = {
4974                 .dev = dev,
4975                 .mtu = mtu,
4976         };
4977
4978         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4979 }
4980
4981 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4982         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4983         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4984         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4985         [RTA_OIF]               = { .type = NLA_U32 },
4986         [RTA_IIF]               = { .type = NLA_U32 },
4987         [RTA_PRIORITY]          = { .type = NLA_U32 },
4988         [RTA_METRICS]           = { .type = NLA_NESTED },
4989         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4990         [RTA_PREF]              = { .type = NLA_U8 },
4991         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4992         [RTA_ENCAP]             = { .type = NLA_NESTED },
4993         [RTA_EXPIRES]           = { .type = NLA_U32 },
4994         [RTA_UID]               = { .type = NLA_U32 },
4995         [RTA_MARK]              = { .type = NLA_U32 },
4996         [RTA_TABLE]             = { .type = NLA_U32 },
4997         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4998         [RTA_SPORT]             = { .type = NLA_U16 },
4999         [RTA_DPORT]             = { .type = NLA_U16 },
5000         [RTA_NH_ID]             = { .type = NLA_U32 },
5001 };
5002
5003 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
5004                               struct fib6_config *cfg,
5005                               struct netlink_ext_ack *extack)
5006 {
5007         struct rtmsg *rtm;
5008         struct nlattr *tb[RTA_MAX+1];
5009         unsigned int pref;
5010         int err;
5011
5012         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5013                                      rtm_ipv6_policy, extack);
5014         if (err < 0)
5015                 goto errout;
5016
5017         err = -EINVAL;
5018         rtm = nlmsg_data(nlh);
5019
5020         if (rtm->rtm_tos) {
5021                 NL_SET_ERR_MSG(extack,
5022                                "Invalid dsfield (tos): option not available for IPv6");
5023                 goto errout;
5024         }
5025
5026         *cfg = (struct fib6_config){
5027                 .fc_table = rtm->rtm_table,
5028                 .fc_dst_len = rtm->rtm_dst_len,
5029                 .fc_src_len = rtm->rtm_src_len,
5030                 .fc_flags = RTF_UP,
5031                 .fc_protocol = rtm->rtm_protocol,
5032                 .fc_type = rtm->rtm_type,
5033
5034                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5035                 .fc_nlinfo.nlh = nlh,
5036                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5037         };
5038
5039         if (rtm->rtm_type == RTN_UNREACHABLE ||
5040             rtm->rtm_type == RTN_BLACKHOLE ||
5041             rtm->rtm_type == RTN_PROHIBIT ||
5042             rtm->rtm_type == RTN_THROW)
5043                 cfg->fc_flags |= RTF_REJECT;
5044
5045         if (rtm->rtm_type == RTN_LOCAL)
5046                 cfg->fc_flags |= RTF_LOCAL;
5047
5048         if (rtm->rtm_flags & RTM_F_CLONED)
5049                 cfg->fc_flags |= RTF_CACHE;
5050
5051         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5052
5053         if (tb[RTA_NH_ID]) {
5054                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5055                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5056                         NL_SET_ERR_MSG(extack,
5057                                        "Nexthop specification and nexthop id are mutually exclusive");
5058                         goto errout;
5059                 }
5060                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5061         }
5062
5063         if (tb[RTA_GATEWAY]) {
5064                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5065                 cfg->fc_flags |= RTF_GATEWAY;
5066         }
5067         if (tb[RTA_VIA]) {
5068                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5069                 goto errout;
5070         }
5071
5072         if (tb[RTA_DST]) {
5073                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5074
5075                 if (nla_len(tb[RTA_DST]) < plen)
5076                         goto errout;
5077
5078                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5079         }
5080
5081         if (tb[RTA_SRC]) {
5082                 int plen = (rtm->rtm_src_len + 7) >> 3;
5083
5084                 if (nla_len(tb[RTA_SRC]) < plen)
5085                         goto errout;
5086
5087                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5088         }
5089
5090         if (tb[RTA_PREFSRC])
5091                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5092
5093         if (tb[RTA_OIF])
5094                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5095
5096         if (tb[RTA_PRIORITY])
5097                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5098
5099         if (tb[RTA_METRICS]) {
5100                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5101                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5102         }
5103
5104         if (tb[RTA_TABLE])
5105                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5106
5107         if (tb[RTA_MULTIPATH]) {
5108                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5109                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5110
5111                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5112                                                      cfg->fc_mp_len, extack);
5113                 if (err < 0)
5114                         goto errout;
5115         }
5116
5117         if (tb[RTA_PREF]) {
5118                 pref = nla_get_u8(tb[RTA_PREF]);
5119                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5120                     pref != ICMPV6_ROUTER_PREF_HIGH)
5121                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5122                 cfg->fc_flags |= RTF_PREF(pref);
5123         }
5124
5125         if (tb[RTA_ENCAP])
5126                 cfg->fc_encap = tb[RTA_ENCAP];
5127
5128         if (tb[RTA_ENCAP_TYPE]) {
5129                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5130
5131                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5132                 if (err < 0)
5133                         goto errout;
5134         }
5135
5136         if (tb[RTA_EXPIRES]) {
5137                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5138
5139                 if (addrconf_finite_timeout(timeout)) {
5140                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5141                         cfg->fc_flags |= RTF_EXPIRES;
5142                 }
5143         }
5144
5145         err = 0;
5146 errout:
5147         return err;
5148 }
5149
5150 struct rt6_nh {
5151         struct fib6_info *fib6_info;
5152         struct fib6_config r_cfg;
5153         struct list_head next;
5154 };
5155
5156 static int ip6_route_info_append(struct net *net,
5157                                  struct list_head *rt6_nh_list,
5158                                  struct fib6_info *rt,
5159                                  struct fib6_config *r_cfg)
5160 {
5161         struct rt6_nh *nh;
5162         int err = -EEXIST;
5163
5164         list_for_each_entry(nh, rt6_nh_list, next) {
5165                 /* check if fib6_info already exists */
5166                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5167                         return err;
5168         }
5169
5170         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5171         if (!nh)
5172                 return -ENOMEM;
5173         nh->fib6_info = rt;
5174         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5175         list_add_tail(&nh->next, rt6_nh_list);
5176
5177         return 0;
5178 }
5179
5180 static void ip6_route_mpath_notify(struct fib6_info *rt,
5181                                    struct fib6_info *rt_last,
5182                                    struct nl_info *info,
5183                                    __u16 nlflags)
5184 {
5185         /* if this is an APPEND route, then rt points to the first route
5186          * inserted and rt_last points to last route inserted. Userspace
5187          * wants a consistent dump of the route which starts at the first
5188          * nexthop. Since sibling routes are always added at the end of
5189          * the list, find the first sibling of the last route appended
5190          */
5191         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5192                 rt = list_first_entry(&rt_last->fib6_siblings,
5193                                       struct fib6_info,
5194                                       fib6_siblings);
5195         }
5196
5197         if (rt)
5198                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5199 }
5200
5201 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5202 {
5203         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5204         bool should_notify = false;
5205         struct fib6_info *leaf;
5206         struct fib6_node *fn;
5207
5208         rcu_read_lock();
5209         fn = rcu_dereference(rt->fib6_node);
5210         if (!fn)
5211                 goto out;
5212
5213         leaf = rcu_dereference(fn->leaf);
5214         if (!leaf)
5215                 goto out;
5216
5217         if (rt == leaf ||
5218             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5219              rt6_qualify_for_ecmp(leaf)))
5220                 should_notify = true;
5221 out:
5222         rcu_read_unlock();
5223
5224         return should_notify;
5225 }
5226
5227 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5228                              struct netlink_ext_ack *extack)
5229 {
5230         if (nla_len(nla) < sizeof(*gw)) {
5231                 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5232                 return -EINVAL;
5233         }
5234
5235         *gw = nla_get_in6_addr(nla);
5236
5237         return 0;
5238 }
5239
5240 static int ip6_route_multipath_add(struct fib6_config *cfg,
5241                                    struct netlink_ext_ack *extack)
5242 {
5243         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5244         struct nl_info *info = &cfg->fc_nlinfo;
5245         struct fib6_config r_cfg;
5246         struct rtnexthop *rtnh;
5247         struct fib6_info *rt;
5248         struct rt6_nh *err_nh;
5249         struct rt6_nh *nh, *nh_safe;
5250         __u16 nlflags;
5251         int remaining;
5252         int attrlen;
5253         int err = 1;
5254         int nhn = 0;
5255         int replace = (cfg->fc_nlinfo.nlh &&
5256                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5257         LIST_HEAD(rt6_nh_list);
5258
5259         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5260         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5261                 nlflags |= NLM_F_APPEND;
5262
5263         remaining = cfg->fc_mp_len;
5264         rtnh = (struct rtnexthop *)cfg->fc_mp;
5265
5266         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5267          * fib6_info structs per nexthop
5268          */
5269         while (rtnh_ok(rtnh, remaining)) {
5270                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5271                 if (rtnh->rtnh_ifindex)
5272                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5273
5274                 attrlen = rtnh_attrlen(rtnh);
5275                 if (attrlen > 0) {
5276                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5277
5278                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5279                         if (nla) {
5280                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5281                                                         extack);
5282                                 if (err)
5283                                         goto cleanup;
5284
5285                                 r_cfg.fc_flags |= RTF_GATEWAY;
5286                         }
5287                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5288
5289                         /* RTA_ENCAP_TYPE length checked in
5290                          * lwtunnel_valid_encap_type_attr
5291                          */
5292                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5293                         if (nla)
5294                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5295                 }
5296
5297                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5298                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5299                 if (IS_ERR(rt)) {
5300                         err = PTR_ERR(rt);
5301                         rt = NULL;
5302                         goto cleanup;
5303                 }
5304                 if (!rt6_qualify_for_ecmp(rt)) {
5305                         err = -EINVAL;
5306                         NL_SET_ERR_MSG(extack,
5307                                        "Device only routes can not be added for IPv6 using the multipath API.");
5308                         fib6_info_release(rt);
5309                         goto cleanup;
5310                 }
5311
5312                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5313
5314                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5315                                             rt, &r_cfg);
5316                 if (err) {
5317                         fib6_info_release(rt);
5318                         goto cleanup;
5319                 }
5320
5321                 rtnh = rtnh_next(rtnh, &remaining);
5322         }
5323
5324         if (list_empty(&rt6_nh_list)) {
5325                 NL_SET_ERR_MSG(extack,
5326                                "Invalid nexthop configuration - no valid nexthops");
5327                 return -EINVAL;
5328         }
5329
5330         /* for add and replace send one notification with all nexthops.
5331          * Skip the notification in fib6_add_rt2node and send one with
5332          * the full route when done
5333          */
5334         info->skip_notify = 1;
5335
5336         /* For add and replace, send one notification with all nexthops. For
5337          * append, send one notification with all appended nexthops.
5338          */
5339         info->skip_notify_kernel = 1;
5340
5341         err_nh = NULL;
5342         list_for_each_entry(nh, &rt6_nh_list, next) {
5343                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5344                 fib6_info_release(nh->fib6_info);
5345
5346                 if (!err) {
5347                         /* save reference to last route successfully inserted */
5348                         rt_last = nh->fib6_info;
5349
5350                         /* save reference to first route for notification */
5351                         if (!rt_notif)
5352                                 rt_notif = nh->fib6_info;
5353                 }
5354
5355                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5356                 nh->fib6_info = NULL;
5357                 if (err) {
5358                         if (replace && nhn)
5359                                 NL_SET_ERR_MSG_MOD(extack,
5360                                                    "multipath route replace failed (check consistency of installed routes)");
5361                         err_nh = nh;
5362                         goto add_errout;
5363                 }
5364
5365                 /* Because each route is added like a single route we remove
5366                  * these flags after the first nexthop: if there is a collision,
5367                  * we have already failed to add the first nexthop:
5368                  * fib6_add_rt2node() has rejected it; when replacing, old
5369                  * nexthops have been replaced by first new, the rest should
5370                  * be added to it.
5371                  */
5372                 if (cfg->fc_nlinfo.nlh) {
5373                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5374                                                              NLM_F_REPLACE);
5375                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5376                 }
5377                 nhn++;
5378         }
5379
5380         /* An in-kernel notification should only be sent in case the new
5381          * multipath route is added as the first route in the node, or if
5382          * it was appended to it. We pass 'rt_notif' since it is the first
5383          * sibling and might allow us to skip some checks in the replace case.
5384          */
5385         if (ip6_route_mpath_should_notify(rt_notif)) {
5386                 enum fib_event_type fib_event;
5387
5388                 if (rt_notif->fib6_nsiblings != nhn - 1)
5389                         fib_event = FIB_EVENT_ENTRY_APPEND;
5390                 else
5391                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5392
5393                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5394                                                           fib_event, rt_notif,
5395                                                           nhn - 1, extack);
5396                 if (err) {
5397                         /* Delete all the siblings that were just added */
5398                         err_nh = NULL;
5399                         goto add_errout;
5400                 }
5401         }
5402
5403         /* success ... tell user about new route */
5404         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5405         goto cleanup;
5406
5407 add_errout:
5408         /* send notification for routes that were added so that
5409          * the delete notifications sent by ip6_route_del are
5410          * coherent
5411          */
5412         if (rt_notif)
5413                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5414
5415         /* Delete routes that were already added */
5416         list_for_each_entry(nh, &rt6_nh_list, next) {
5417                 if (err_nh == nh)
5418                         break;
5419                 ip6_route_del(&nh->r_cfg, extack);
5420         }
5421
5422 cleanup:
5423         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5424                 if (nh->fib6_info)
5425                         fib6_info_release(nh->fib6_info);
5426                 list_del(&nh->next);
5427                 kfree(nh);
5428         }
5429
5430         return err;
5431 }
5432
5433 static int ip6_route_multipath_del(struct fib6_config *cfg,
5434                                    struct netlink_ext_ack *extack)
5435 {
5436         struct fib6_config r_cfg;
5437         struct rtnexthop *rtnh;
5438         int last_err = 0;
5439         int remaining;
5440         int attrlen;
5441         int err;
5442
5443         remaining = cfg->fc_mp_len;
5444         rtnh = (struct rtnexthop *)cfg->fc_mp;
5445
5446         /* Parse a Multipath Entry */
5447         while (rtnh_ok(rtnh, remaining)) {
5448                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5449                 if (rtnh->rtnh_ifindex)
5450                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5451
5452                 attrlen = rtnh_attrlen(rtnh);
5453                 if (attrlen > 0) {
5454                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5455
5456                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5457                         if (nla) {
5458                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5459                                                         extack);
5460                                 if (err) {
5461                                         last_err = err;
5462                                         goto next_rtnh;
5463                                 }
5464
5465                                 r_cfg.fc_flags |= RTF_GATEWAY;
5466                         }
5467                 }
5468                 err = ip6_route_del(&r_cfg, extack);
5469                 if (err)
5470                         last_err = err;
5471
5472 next_rtnh:
5473                 rtnh = rtnh_next(rtnh, &remaining);
5474         }
5475
5476         return last_err;
5477 }
5478
5479 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5480                               struct netlink_ext_ack *extack)
5481 {
5482         struct fib6_config cfg;
5483         int err;
5484
5485         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5486         if (err < 0)
5487                 return err;
5488
5489         if (cfg.fc_nh_id &&
5490             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5491                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5492                 return -EINVAL;
5493         }
5494
5495         if (cfg.fc_mp)
5496                 return ip6_route_multipath_del(&cfg, extack);
5497         else {
5498                 cfg.fc_delete_all_nh = 1;
5499                 return ip6_route_del(&cfg, extack);
5500         }
5501 }
5502
5503 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5504                               struct netlink_ext_ack *extack)
5505 {
5506         struct fib6_config cfg;
5507         int err;
5508
5509         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5510         if (err < 0)
5511                 return err;
5512
5513         if (cfg.fc_metric == 0)
5514                 cfg.fc_metric = IP6_RT_PRIO_USER;
5515
5516         if (cfg.fc_mp)
5517                 return ip6_route_multipath_add(&cfg, extack);
5518         else
5519                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5520 }
5521
5522 /* add the overhead of this fib6_nh to nexthop_len */
5523 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5524 {
5525         int *nexthop_len = arg;
5526
5527         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5528                      + NLA_ALIGN(sizeof(struct rtnexthop))
5529                      + nla_total_size(16); /* RTA_GATEWAY */
5530
5531         if (nh->fib_nh_lws) {
5532                 /* RTA_ENCAP_TYPE */
5533                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5534                 /* RTA_ENCAP */
5535                 *nexthop_len += nla_total_size(2);
5536         }
5537
5538         return 0;
5539 }
5540
5541 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5542 {
5543         int nexthop_len;
5544
5545         if (f6i->nh) {
5546                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5547                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5548                                          &nexthop_len);
5549         } else {
5550                 struct fib6_info *sibling, *next_sibling;
5551                 struct fib6_nh *nh = f6i->fib6_nh;
5552
5553                 nexthop_len = 0;
5554                 if (f6i->fib6_nsiblings) {
5555                         rt6_nh_nlmsg_size(nh, &nexthop_len);
5556
5557                         list_for_each_entry_safe(sibling, next_sibling,
5558                                                  &f6i->fib6_siblings, fib6_siblings) {
5559                                 rt6_nh_nlmsg_size(sibling->fib6_nh, &nexthop_len);
5560                         }
5561                 }
5562                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5563         }
5564
5565         return NLMSG_ALIGN(sizeof(struct rtmsg))
5566                + nla_total_size(16) /* RTA_SRC */
5567                + nla_total_size(16) /* RTA_DST */
5568                + nla_total_size(16) /* RTA_GATEWAY */
5569                + nla_total_size(16) /* RTA_PREFSRC */
5570                + nla_total_size(4) /* RTA_TABLE */
5571                + nla_total_size(4) /* RTA_IIF */
5572                + nla_total_size(4) /* RTA_OIF */
5573                + nla_total_size(4) /* RTA_PRIORITY */
5574                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5575                + nla_total_size(sizeof(struct rta_cacheinfo))
5576                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5577                + nla_total_size(1) /* RTA_PREF */
5578                + nexthop_len;
5579 }
5580
5581 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5582                                  unsigned char *flags)
5583 {
5584         if (nexthop_is_multipath(nh)) {
5585                 struct nlattr *mp;
5586
5587                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5588                 if (!mp)
5589                         goto nla_put_failure;
5590
5591                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5592                         goto nla_put_failure;
5593
5594                 nla_nest_end(skb, mp);
5595         } else {
5596                 struct fib6_nh *fib6_nh;
5597
5598                 fib6_nh = nexthop_fib6_nh(nh);
5599                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5600                                      flags, false) < 0)
5601                         goto nla_put_failure;
5602         }
5603
5604         return 0;
5605
5606 nla_put_failure:
5607         return -EMSGSIZE;
5608 }
5609
5610 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5611                          struct fib6_info *rt, struct dst_entry *dst,
5612                          struct in6_addr *dest, struct in6_addr *src,
5613                          int iif, int type, u32 portid, u32 seq,
5614                          unsigned int flags)
5615 {
5616         struct rt6_info *rt6 = (struct rt6_info *)dst;
5617         struct rt6key *rt6_dst, *rt6_src;
5618         u32 *pmetrics, table, rt6_flags;
5619         unsigned char nh_flags = 0;
5620         struct nlmsghdr *nlh;
5621         struct rtmsg *rtm;
5622         long expires = 0;
5623
5624         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5625         if (!nlh)
5626                 return -EMSGSIZE;
5627
5628         if (rt6) {
5629                 rt6_dst = &rt6->rt6i_dst;
5630                 rt6_src = &rt6->rt6i_src;
5631                 rt6_flags = rt6->rt6i_flags;
5632         } else {
5633                 rt6_dst = &rt->fib6_dst;
5634                 rt6_src = &rt->fib6_src;
5635                 rt6_flags = rt->fib6_flags;
5636         }
5637
5638         rtm = nlmsg_data(nlh);
5639         rtm->rtm_family = AF_INET6;
5640         rtm->rtm_dst_len = rt6_dst->plen;
5641         rtm->rtm_src_len = rt6_src->plen;
5642         rtm->rtm_tos = 0;
5643         if (rt->fib6_table)
5644                 table = rt->fib6_table->tb6_id;
5645         else
5646                 table = RT6_TABLE_UNSPEC;
5647         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5648         if (nla_put_u32(skb, RTA_TABLE, table))
5649                 goto nla_put_failure;
5650
5651         rtm->rtm_type = rt->fib6_type;
5652         rtm->rtm_flags = 0;
5653         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5654         rtm->rtm_protocol = rt->fib6_protocol;
5655
5656         if (rt6_flags & RTF_CACHE)
5657                 rtm->rtm_flags |= RTM_F_CLONED;
5658
5659         if (dest) {
5660                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5661                         goto nla_put_failure;
5662                 rtm->rtm_dst_len = 128;
5663         } else if (rtm->rtm_dst_len)
5664                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5665                         goto nla_put_failure;
5666 #ifdef CONFIG_IPV6_SUBTREES
5667         if (src) {
5668                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5669                         goto nla_put_failure;
5670                 rtm->rtm_src_len = 128;
5671         } else if (rtm->rtm_src_len &&
5672                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5673                 goto nla_put_failure;
5674 #endif
5675         if (iif) {
5676 #ifdef CONFIG_IPV6_MROUTE
5677                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5678                         int err = ip6mr_get_route(net, skb, rtm, portid);
5679
5680                         if (err == 0)
5681                                 return 0;
5682                         if (err < 0)
5683                                 goto nla_put_failure;
5684                 } else
5685 #endif
5686                         if (nla_put_u32(skb, RTA_IIF, iif))
5687                                 goto nla_put_failure;
5688         } else if (dest) {
5689                 struct in6_addr saddr_buf;
5690                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5691                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5692                         goto nla_put_failure;
5693         }
5694
5695         if (rt->fib6_prefsrc.plen) {
5696                 struct in6_addr saddr_buf;
5697                 saddr_buf = rt->fib6_prefsrc.addr;
5698                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5699                         goto nla_put_failure;
5700         }
5701
5702         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5703         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5704                 goto nla_put_failure;
5705
5706         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5707                 goto nla_put_failure;
5708
5709         /* For multipath routes, walk the siblings list and add
5710          * each as a nexthop within RTA_MULTIPATH.
5711          */
5712         if (rt6) {
5713                 if (rt6_flags & RTF_GATEWAY &&
5714                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5715                         goto nla_put_failure;
5716
5717                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5718                         goto nla_put_failure;
5719
5720                 if (dst->lwtstate &&
5721                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5722                         goto nla_put_failure;
5723         } else if (rt->fib6_nsiblings) {
5724                 struct fib6_info *sibling, *next_sibling;
5725                 struct nlattr *mp;
5726
5727                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5728                 if (!mp)
5729                         goto nla_put_failure;
5730
5731                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5732                                     rt->fib6_nh->fib_nh_weight, AF_INET6,
5733                                     0) < 0)
5734                         goto nla_put_failure;
5735
5736                 list_for_each_entry_safe(sibling, next_sibling,
5737                                          &rt->fib6_siblings, fib6_siblings) {
5738                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5739                                             sibling->fib6_nh->fib_nh_weight,
5740                                             AF_INET6, 0) < 0)
5741                                 goto nla_put_failure;
5742                 }
5743
5744                 nla_nest_end(skb, mp);
5745         } else if (rt->nh) {
5746                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5747                         goto nla_put_failure;
5748
5749                 if (nexthop_is_blackhole(rt->nh))
5750                         rtm->rtm_type = RTN_BLACKHOLE;
5751
5752                 if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5753                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5754                         goto nla_put_failure;
5755
5756                 rtm->rtm_flags |= nh_flags;
5757         } else {
5758                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5759                                      &nh_flags, false) < 0)
5760                         goto nla_put_failure;
5761
5762                 rtm->rtm_flags |= nh_flags;
5763         }
5764
5765         if (rt6_flags & RTF_EXPIRES) {
5766                 expires = dst ? dst->expires : rt->expires;
5767                 expires -= jiffies;
5768         }
5769
5770         if (!dst) {
5771                 if (READ_ONCE(rt->offload))
5772                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5773                 if (READ_ONCE(rt->trap))
5774                         rtm->rtm_flags |= RTM_F_TRAP;
5775                 if (READ_ONCE(rt->offload_failed))
5776                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5777         }
5778
5779         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5780                 goto nla_put_failure;
5781
5782         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5783                 goto nla_put_failure;
5784
5785
5786         nlmsg_end(skb, nlh);
5787         return 0;
5788
5789 nla_put_failure:
5790         nlmsg_cancel(skb, nlh);
5791         return -EMSGSIZE;
5792 }
5793
5794 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5795 {
5796         const struct net_device *dev = arg;
5797
5798         if (nh->fib_nh_dev == dev)
5799                 return 1;
5800
5801         return 0;
5802 }
5803
5804 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5805                                const struct net_device *dev)
5806 {
5807         if (f6i->nh) {
5808                 struct net_device *_dev = (struct net_device *)dev;
5809
5810                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5811                                                   fib6_info_nh_uses_dev,
5812                                                   _dev);
5813         }
5814
5815         if (f6i->fib6_nh->fib_nh_dev == dev)
5816                 return true;
5817
5818         if (f6i->fib6_nsiblings) {
5819                 struct fib6_info *sibling, *next_sibling;
5820
5821                 list_for_each_entry_safe(sibling, next_sibling,
5822                                          &f6i->fib6_siblings, fib6_siblings) {
5823                         if (sibling->fib6_nh->fib_nh_dev == dev)
5824                                 return true;
5825                 }
5826         }
5827
5828         return false;
5829 }
5830
5831 struct fib6_nh_exception_dump_walker {
5832         struct rt6_rtnl_dump_arg *dump;
5833         struct fib6_info *rt;
5834         unsigned int flags;
5835         unsigned int skip;
5836         unsigned int count;
5837 };
5838
5839 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5840 {
5841         struct fib6_nh_exception_dump_walker *w = arg;
5842         struct rt6_rtnl_dump_arg *dump = w->dump;
5843         struct rt6_exception_bucket *bucket;
5844         struct rt6_exception *rt6_ex;
5845         int i, err;
5846
5847         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5848         if (!bucket)
5849                 return 0;
5850
5851         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5852                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5853                         if (w->skip) {
5854                                 w->skip--;
5855                                 continue;
5856                         }
5857
5858                         /* Expiration of entries doesn't bump sernum, insertion
5859                          * does. Removal is triggered by insertion, so we can
5860                          * rely on the fact that if entries change between two
5861                          * partial dumps, this node is scanned again completely,
5862                          * see rt6_insert_exception() and fib6_dump_table().
5863                          *
5864                          * Count expired entries we go through as handled
5865                          * entries that we'll skip next time, in case of partial
5866                          * node dump. Otherwise, if entries expire meanwhile,
5867                          * we'll skip the wrong amount.
5868                          */
5869                         if (rt6_check_expired(rt6_ex->rt6i)) {
5870                                 w->count++;
5871                                 continue;
5872                         }
5873
5874                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5875                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5876                                             RTM_NEWROUTE,
5877                                             NETLINK_CB(dump->cb->skb).portid,
5878                                             dump->cb->nlh->nlmsg_seq, w->flags);
5879                         if (err)
5880                                 return err;
5881
5882                         w->count++;
5883                 }
5884                 bucket++;
5885         }
5886
5887         return 0;
5888 }
5889
5890 /* Return -1 if done with node, number of handled routes on partial dump */
5891 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5892 {
5893         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5894         struct fib_dump_filter *filter = &arg->filter;
5895         unsigned int flags = NLM_F_MULTI;
5896         struct net *net = arg->net;
5897         int count = 0;
5898
5899         if (rt == net->ipv6.fib6_null_entry)
5900                 return -1;
5901
5902         if ((filter->flags & RTM_F_PREFIX) &&
5903             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5904                 /* success since this is not a prefix route */
5905                 return -1;
5906         }
5907         if (filter->filter_set &&
5908             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5909              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5910              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5911                 return -1;
5912         }
5913
5914         if (filter->filter_set ||
5915             !filter->dump_routes || !filter->dump_exceptions) {
5916                 flags |= NLM_F_DUMP_FILTERED;
5917         }
5918
5919         if (filter->dump_routes) {
5920                 if (skip) {
5921                         skip--;
5922                 } else {
5923                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5924                                           0, RTM_NEWROUTE,
5925                                           NETLINK_CB(arg->cb->skb).portid,
5926                                           arg->cb->nlh->nlmsg_seq, flags)) {
5927                                 return 0;
5928                         }
5929                         count++;
5930                 }
5931         }
5932
5933         if (filter->dump_exceptions) {
5934                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5935                                                            .rt = rt,
5936                                                            .flags = flags,
5937                                                            .skip = skip,
5938                                                            .count = 0 };
5939                 int err;
5940
5941                 rcu_read_lock();
5942                 if (rt->nh) {
5943                         err = nexthop_for_each_fib6_nh(rt->nh,
5944                                                        rt6_nh_dump_exceptions,
5945                                                        &w);
5946                 } else {
5947                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5948                 }
5949                 rcu_read_unlock();
5950
5951                 if (err)
5952                         return count + w.count;
5953         }
5954
5955         return -1;
5956 }
5957
5958 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5959                                         const struct nlmsghdr *nlh,
5960                                         struct nlattr **tb,
5961                                         struct netlink_ext_ack *extack)
5962 {
5963         struct rtmsg *rtm;
5964         int i, err;
5965
5966         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5967                 NL_SET_ERR_MSG_MOD(extack,
5968                                    "Invalid header for get route request");
5969                 return -EINVAL;
5970         }
5971
5972         if (!netlink_strict_get_check(skb))
5973                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5974                                               rtm_ipv6_policy, extack);
5975
5976         rtm = nlmsg_data(nlh);
5977         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5978             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5979             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5980             rtm->rtm_type) {
5981                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5982                 return -EINVAL;
5983         }
5984         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5985                 NL_SET_ERR_MSG_MOD(extack,
5986                                    "Invalid flags for get route request");
5987                 return -EINVAL;
5988         }
5989
5990         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5991                                             rtm_ipv6_policy, extack);
5992         if (err)
5993                 return err;
5994
5995         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5996             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5997                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5998                 return -EINVAL;
5999         }
6000
6001         for (i = 0; i <= RTA_MAX; i++) {
6002                 if (!tb[i])
6003                         continue;
6004
6005                 switch (i) {
6006                 case RTA_SRC:
6007                 case RTA_DST:
6008                 case RTA_IIF:
6009                 case RTA_OIF:
6010                 case RTA_MARK:
6011                 case RTA_UID:
6012                 case RTA_SPORT:
6013                 case RTA_DPORT:
6014                 case RTA_IP_PROTO:
6015                         break;
6016                 default:
6017                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6018                         return -EINVAL;
6019                 }
6020         }
6021
6022         return 0;
6023 }
6024
6025 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6026                               struct netlink_ext_ack *extack)
6027 {
6028         struct net *net = sock_net(in_skb->sk);
6029         struct nlattr *tb[RTA_MAX+1];
6030         int err, iif = 0, oif = 0;
6031         struct fib6_info *from;
6032         struct dst_entry *dst;
6033         struct rt6_info *rt;
6034         struct sk_buff *skb;
6035         struct rtmsg *rtm;
6036         struct flowi6 fl6 = {};
6037         bool fibmatch;
6038
6039         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6040         if (err < 0)
6041                 goto errout;
6042
6043         err = -EINVAL;
6044         rtm = nlmsg_data(nlh);
6045         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6046         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6047
6048         if (tb[RTA_SRC]) {
6049                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6050                         goto errout;
6051
6052                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6053         }
6054
6055         if (tb[RTA_DST]) {
6056                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6057                         goto errout;
6058
6059                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6060         }
6061
6062         if (tb[RTA_IIF])
6063                 iif = nla_get_u32(tb[RTA_IIF]);
6064
6065         if (tb[RTA_OIF])
6066                 oif = nla_get_u32(tb[RTA_OIF]);
6067
6068         if (tb[RTA_MARK])
6069                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6070
6071         if (tb[RTA_UID])
6072                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6073                                            nla_get_u32(tb[RTA_UID]));
6074         else
6075                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6076
6077         if (tb[RTA_SPORT])
6078                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6079
6080         if (tb[RTA_DPORT])
6081                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6082
6083         if (tb[RTA_IP_PROTO]) {
6084                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6085                                                   &fl6.flowi6_proto, AF_INET6,
6086                                                   extack);
6087                 if (err)
6088                         goto errout;
6089         }
6090
6091         if (iif) {
6092                 struct net_device *dev;
6093                 int flags = 0;
6094
6095                 rcu_read_lock();
6096
6097                 dev = dev_get_by_index_rcu(net, iif);
6098                 if (!dev) {
6099                         rcu_read_unlock();
6100                         err = -ENODEV;
6101                         goto errout;
6102                 }
6103
6104                 fl6.flowi6_iif = iif;
6105
6106                 if (!ipv6_addr_any(&fl6.saddr))
6107                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6108
6109                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6110
6111                 rcu_read_unlock();
6112         } else {
6113                 fl6.flowi6_oif = oif;
6114
6115                 dst = ip6_route_output(net, NULL, &fl6);
6116         }
6117
6118
6119         rt = container_of(dst, struct rt6_info, dst);
6120         if (rt->dst.error) {
6121                 err = rt->dst.error;
6122                 ip6_rt_put(rt);
6123                 goto errout;
6124         }
6125
6126         if (rt == net->ipv6.ip6_null_entry) {
6127                 err = rt->dst.error;
6128                 ip6_rt_put(rt);
6129                 goto errout;
6130         }
6131
6132         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6133         if (!skb) {
6134                 ip6_rt_put(rt);
6135                 err = -ENOBUFS;
6136                 goto errout;
6137         }
6138
6139         skb_dst_set(skb, &rt->dst);
6140
6141         rcu_read_lock();
6142         from = rcu_dereference(rt->from);
6143         if (from) {
6144                 if (fibmatch)
6145                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6146                                             iif, RTM_NEWROUTE,
6147                                             NETLINK_CB(in_skb).portid,
6148                                             nlh->nlmsg_seq, 0);
6149                 else
6150                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6151                                             &fl6.saddr, iif, RTM_NEWROUTE,
6152                                             NETLINK_CB(in_skb).portid,
6153                                             nlh->nlmsg_seq, 0);
6154         } else {
6155                 err = -ENETUNREACH;
6156         }
6157         rcu_read_unlock();
6158
6159         if (err < 0) {
6160                 kfree_skb(skb);
6161                 goto errout;
6162         }
6163
6164         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6165 errout:
6166         return err;
6167 }
6168
6169 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6170                      unsigned int nlm_flags)
6171 {
6172         struct sk_buff *skb;
6173         struct net *net = info->nl_net;
6174         u32 seq;
6175         int err;
6176
6177         err = -ENOBUFS;
6178         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6179
6180         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6181         if (!skb)
6182                 goto errout;
6183
6184         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6185                             event, info->portid, seq, nlm_flags);
6186         if (err < 0) {
6187                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6188                 WARN_ON(err == -EMSGSIZE);
6189                 kfree_skb(skb);
6190                 goto errout;
6191         }
6192         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6193                     info->nlh, gfp_any());
6194         return;
6195 errout:
6196         if (err < 0)
6197                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6198 }
6199
6200 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6201                     struct nl_info *info)
6202 {
6203         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6204         struct sk_buff *skb;
6205         int err = -ENOBUFS;
6206
6207         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6208         if (!skb)
6209                 goto errout;
6210
6211         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6212                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6213         if (err < 0) {
6214                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6215                 WARN_ON(err == -EMSGSIZE);
6216                 kfree_skb(skb);
6217                 goto errout;
6218         }
6219         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6220                     info->nlh, gfp_any());
6221         return;
6222 errout:
6223         if (err < 0)
6224                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6225 }
6226
6227 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6228                             bool offload, bool trap, bool offload_failed)
6229 {
6230         struct sk_buff *skb;
6231         int err;
6232
6233         if (READ_ONCE(f6i->offload) == offload &&
6234             READ_ONCE(f6i->trap) == trap &&
6235             READ_ONCE(f6i->offload_failed) == offload_failed)
6236                 return;
6237
6238         WRITE_ONCE(f6i->offload, offload);
6239         WRITE_ONCE(f6i->trap, trap);
6240
6241         /* 2 means send notifications only if offload_failed was changed. */
6242         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6243             READ_ONCE(f6i->offload_failed) == offload_failed)
6244                 return;
6245
6246         WRITE_ONCE(f6i->offload_failed, offload_failed);
6247
6248         if (!rcu_access_pointer(f6i->fib6_node))
6249                 /* The route was removed from the tree, do not send
6250                  * notification.
6251                  */
6252                 return;
6253
6254         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6255                 return;
6256
6257         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6258         if (!skb) {
6259                 err = -ENOBUFS;
6260                 goto errout;
6261         }
6262
6263         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6264                             0, 0);
6265         if (err < 0) {
6266                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6267                 WARN_ON(err == -EMSGSIZE);
6268                 kfree_skb(skb);
6269                 goto errout;
6270         }
6271
6272         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6273         return;
6274
6275 errout:
6276         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6277 }
6278 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6279
6280 static int ip6_route_dev_notify(struct notifier_block *this,
6281                                 unsigned long event, void *ptr)
6282 {
6283         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6284         struct net *net = dev_net(dev);
6285
6286         if (!(dev->flags & IFF_LOOPBACK))
6287                 return NOTIFY_OK;
6288
6289         if (event == NETDEV_REGISTER) {
6290                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6291                 net->ipv6.ip6_null_entry->dst.dev = dev;
6292                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6293 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6294                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6295                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6296                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6297                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6298 #endif
6299          } else if (event == NETDEV_UNREGISTER &&
6300                     dev->reg_state != NETREG_UNREGISTERED) {
6301                 /* NETDEV_UNREGISTER could be fired for multiple times by
6302                  * netdev_wait_allrefs(). Make sure we only call this once.
6303                  */
6304                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6305 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6306                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6307                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6308 #endif
6309         }
6310
6311         return NOTIFY_OK;
6312 }
6313
6314 /*
6315  *      /proc
6316  */
6317
6318 #ifdef CONFIG_PROC_FS
6319 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6320 {
6321         struct net *net = (struct net *)seq->private;
6322         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6323                    net->ipv6.rt6_stats->fib_nodes,
6324                    net->ipv6.rt6_stats->fib_route_nodes,
6325                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6326                    net->ipv6.rt6_stats->fib_rt_entries,
6327                    net->ipv6.rt6_stats->fib_rt_cache,
6328                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6329                    net->ipv6.rt6_stats->fib_discarded_routes);
6330
6331         return 0;
6332 }
6333 #endif  /* CONFIG_PROC_FS */
6334
6335 #ifdef CONFIG_SYSCTL
6336
6337 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6338                               void *buffer, size_t *lenp, loff_t *ppos)
6339 {
6340         struct net *net;
6341         int delay;
6342         int ret;
6343         if (!write)
6344                 return -EINVAL;
6345
6346         net = (struct net *)ctl->extra1;
6347         delay = net->ipv6.sysctl.flush_delay;
6348         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6349         if (ret)
6350                 return ret;
6351
6352         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6353         return 0;
6354 }
6355
6356 static struct ctl_table ipv6_route_table_template[] = {
6357         {
6358                 .procname       =       "max_size",
6359                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6360                 .maxlen         =       sizeof(int),
6361                 .mode           =       0644,
6362                 .proc_handler   =       proc_dointvec,
6363         },
6364         {
6365                 .procname       =       "gc_thresh",
6366                 .data           =       &ip6_dst_ops_template.gc_thresh,
6367                 .maxlen         =       sizeof(int),
6368                 .mode           =       0644,
6369                 .proc_handler   =       proc_dointvec,
6370         },
6371         {
6372                 .procname       =       "flush",
6373                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6374                 .maxlen         =       sizeof(int),
6375                 .mode           =       0200,
6376                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6377         },
6378         {
6379                 .procname       =       "gc_min_interval",
6380                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6381                 .maxlen         =       sizeof(int),
6382                 .mode           =       0644,
6383                 .proc_handler   =       proc_dointvec_jiffies,
6384         },
6385         {
6386                 .procname       =       "gc_timeout",
6387                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6388                 .maxlen         =       sizeof(int),
6389                 .mode           =       0644,
6390                 .proc_handler   =       proc_dointvec_jiffies,
6391         },
6392         {
6393                 .procname       =       "gc_interval",
6394                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6395                 .maxlen         =       sizeof(int),
6396                 .mode           =       0644,
6397                 .proc_handler   =       proc_dointvec_jiffies,
6398         },
6399         {
6400                 .procname       =       "gc_elasticity",
6401                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6402                 .maxlen         =       sizeof(int),
6403                 .mode           =       0644,
6404                 .proc_handler   =       proc_dointvec,
6405         },
6406         {
6407                 .procname       =       "mtu_expires",
6408                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6409                 .maxlen         =       sizeof(int),
6410                 .mode           =       0644,
6411                 .proc_handler   =       proc_dointvec_jiffies,
6412         },
6413         {
6414                 .procname       =       "min_adv_mss",
6415                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6416                 .maxlen         =       sizeof(int),
6417                 .mode           =       0644,
6418                 .proc_handler   =       proc_dointvec,
6419         },
6420         {
6421                 .procname       =       "gc_min_interval_ms",
6422                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6423                 .maxlen         =       sizeof(int),
6424                 .mode           =       0644,
6425                 .proc_handler   =       proc_dointvec_ms_jiffies,
6426         },
6427         {
6428                 .procname       =       "skip_notify_on_dev_down",
6429                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6430                 .maxlen         =       sizeof(u8),
6431                 .mode           =       0644,
6432                 .proc_handler   =       proc_dou8vec_minmax,
6433                 .extra1         =       SYSCTL_ZERO,
6434                 .extra2         =       SYSCTL_ONE,
6435         },
6436         { }
6437 };
6438
6439 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6440 {
6441         struct ctl_table *table;
6442
6443         table = kmemdup(ipv6_route_table_template,
6444                         sizeof(ipv6_route_table_template),
6445                         GFP_KERNEL);
6446
6447         if (table) {
6448                 table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6449                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6450                 table[2].data = &net->ipv6.sysctl.flush_delay;
6451                 table[2].extra1 = net;
6452                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6453                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6454                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6455                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6456                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6457                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6458                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6459                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6460
6461                 /* Don't export sysctls to unprivileged users */
6462                 if (net->user_ns != &init_user_ns)
6463                         table[1].procname = NULL;
6464         }
6465
6466         return table;
6467 }
6468
6469 size_t ipv6_route_sysctl_table_size(struct net *net)
6470 {
6471         /* Don't export sysctls to unprivileged users */
6472         if (net->user_ns != &init_user_ns)
6473                 return 1;
6474
6475         return ARRAY_SIZE(ipv6_route_table_template);
6476 }
6477 #endif
6478
6479 static int __net_init ip6_route_net_init(struct net *net)
6480 {
6481         int ret = -ENOMEM;
6482
6483         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6484                sizeof(net->ipv6.ip6_dst_ops));
6485
6486         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6487                 goto out_ip6_dst_ops;
6488
6489         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6490         if (!net->ipv6.fib6_null_entry)
6491                 goto out_ip6_dst_entries;
6492         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6493                sizeof(*net->ipv6.fib6_null_entry));
6494
6495         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6496                                            sizeof(*net->ipv6.ip6_null_entry),
6497                                            GFP_KERNEL);
6498         if (!net->ipv6.ip6_null_entry)
6499                 goto out_fib6_null_entry;
6500         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6501         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6502                          ip6_template_metrics, true);
6503         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->dst.rt_uncached);
6504
6505 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6506         net->ipv6.fib6_has_custom_rules = false;
6507         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6508                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6509                                                GFP_KERNEL);
6510         if (!net->ipv6.ip6_prohibit_entry)
6511                 goto out_ip6_null_entry;
6512         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6513         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6514                          ip6_template_metrics, true);
6515         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->dst.rt_uncached);
6516
6517         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6518                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6519                                                GFP_KERNEL);
6520         if (!net->ipv6.ip6_blk_hole_entry)
6521                 goto out_ip6_prohibit_entry;
6522         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6523         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6524                          ip6_template_metrics, true);
6525         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->dst.rt_uncached);
6526 #ifdef CONFIG_IPV6_SUBTREES
6527         net->ipv6.fib6_routes_require_src = 0;
6528 #endif
6529 #endif
6530
6531         net->ipv6.sysctl.flush_delay = 0;
6532         net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
6533         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6534         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6535         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6536         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6537         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6538         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6539         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6540
6541         atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6542
6543         ret = 0;
6544 out:
6545         return ret;
6546
6547 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6548 out_ip6_prohibit_entry:
6549         kfree(net->ipv6.ip6_prohibit_entry);
6550 out_ip6_null_entry:
6551         kfree(net->ipv6.ip6_null_entry);
6552 #endif
6553 out_fib6_null_entry:
6554         kfree(net->ipv6.fib6_null_entry);
6555 out_ip6_dst_entries:
6556         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6557 out_ip6_dst_ops:
6558         goto out;
6559 }
6560
6561 static void __net_exit ip6_route_net_exit(struct net *net)
6562 {
6563         kfree(net->ipv6.fib6_null_entry);
6564         kfree(net->ipv6.ip6_null_entry);
6565 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6566         kfree(net->ipv6.ip6_prohibit_entry);
6567         kfree(net->ipv6.ip6_blk_hole_entry);
6568 #endif
6569         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6570 }
6571
6572 static int __net_init ip6_route_net_init_late(struct net *net)
6573 {
6574 #ifdef CONFIG_PROC_FS
6575         if (!proc_create_net("ipv6_route", 0, net->proc_net,
6576                              &ipv6_route_seq_ops,
6577                              sizeof(struct ipv6_route_iter)))
6578                 return -ENOMEM;
6579
6580         if (!proc_create_net_single("rt6_stats", 0444, net->proc_net,
6581                                     rt6_stats_seq_show, NULL)) {
6582                 remove_proc_entry("ipv6_route", net->proc_net);
6583                 return -ENOMEM;
6584         }
6585 #endif
6586         return 0;
6587 }
6588
6589 static void __net_exit ip6_route_net_exit_late(struct net *net)
6590 {
6591 #ifdef CONFIG_PROC_FS
6592         remove_proc_entry("ipv6_route", net->proc_net);
6593         remove_proc_entry("rt6_stats", net->proc_net);
6594 #endif
6595 }
6596
6597 static struct pernet_operations ip6_route_net_ops = {
6598         .init = ip6_route_net_init,
6599         .exit = ip6_route_net_exit,
6600 };
6601
6602 static int __net_init ipv6_inetpeer_init(struct net *net)
6603 {
6604         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6605
6606         if (!bp)
6607                 return -ENOMEM;
6608         inet_peer_base_init(bp);
6609         net->ipv6.peers = bp;
6610         return 0;
6611 }
6612
6613 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6614 {
6615         struct inet_peer_base *bp = net->ipv6.peers;
6616
6617         net->ipv6.peers = NULL;
6618         inetpeer_invalidate_tree(bp);
6619         kfree(bp);
6620 }
6621
6622 static struct pernet_operations ipv6_inetpeer_ops = {
6623         .init   =       ipv6_inetpeer_init,
6624         .exit   =       ipv6_inetpeer_exit,
6625 };
6626
6627 static struct pernet_operations ip6_route_net_late_ops = {
6628         .init = ip6_route_net_init_late,
6629         .exit = ip6_route_net_exit_late,
6630 };
6631
6632 static struct notifier_block ip6_route_dev_notifier = {
6633         .notifier_call = ip6_route_dev_notify,
6634         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6635 };
6636
6637 void __init ip6_route_init_special_entries(void)
6638 {
6639         /* Registering of the loopback is done before this portion of code,
6640          * the loopback reference in rt6_info will not be taken, do it
6641          * manually for init_net */
6642         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6643         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6644         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6645   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6646         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6647         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6648         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6649         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6650   #endif
6651 }
6652
6653 #if IS_BUILTIN(CONFIG_IPV6)
6654 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6655 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6656
6657 BTF_ID_LIST(btf_fib6_info_id)
6658 BTF_ID(struct, fib6_info)
6659
6660 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6661         .seq_ops                = &ipv6_route_seq_ops,
6662         .init_seq_private       = bpf_iter_init_seq_net,
6663         .fini_seq_private       = bpf_iter_fini_seq_net,
6664         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6665 };
6666
6667 static struct bpf_iter_reg ipv6_route_reg_info = {
6668         .target                 = "ipv6_route",
6669         .ctx_arg_info_size      = 1,
6670         .ctx_arg_info           = {
6671                 { offsetof(struct bpf_iter__ipv6_route, rt),
6672                   PTR_TO_BTF_ID_OR_NULL },
6673         },
6674         .seq_info               = &ipv6_route_seq_info,
6675 };
6676
6677 static int __init bpf_iter_register(void)
6678 {
6679         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6680         return bpf_iter_reg_target(&ipv6_route_reg_info);
6681 }
6682
6683 static void bpf_iter_unregister(void)
6684 {
6685         bpf_iter_unreg_target(&ipv6_route_reg_info);
6686 }
6687 #endif
6688 #endif
6689
6690 int __init ip6_route_init(void)
6691 {
6692         int ret;
6693         int cpu;
6694
6695         ret = -ENOMEM;
6696         ip6_dst_ops_template.kmem_cachep =
6697                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6698                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6699         if (!ip6_dst_ops_template.kmem_cachep)
6700                 goto out;
6701
6702         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6703         if (ret)
6704                 goto out_kmem_cache;
6705
6706         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6707         if (ret)
6708                 goto out_dst_entries;
6709
6710         ret = register_pernet_subsys(&ip6_route_net_ops);
6711         if (ret)
6712                 goto out_register_inetpeer;
6713
6714         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6715
6716         ret = fib6_init();
6717         if (ret)
6718                 goto out_register_subsys;
6719
6720         ret = xfrm6_init();
6721         if (ret)
6722                 goto out_fib6_init;
6723
6724         ret = fib6_rules_init();
6725         if (ret)
6726                 goto xfrm6_init;
6727
6728         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6729         if (ret)
6730                 goto fib6_rules_init;
6731
6732         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6733                                    inet6_rtm_newroute, NULL, 0);
6734         if (ret < 0)
6735                 goto out_register_late_subsys;
6736
6737         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6738                                    inet6_rtm_delroute, NULL, 0);
6739         if (ret < 0)
6740                 goto out_register_late_subsys;
6741
6742         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6743                                    inet6_rtm_getroute, NULL,
6744                                    RTNL_FLAG_DOIT_UNLOCKED);
6745         if (ret < 0)
6746                 goto out_register_late_subsys;
6747
6748         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6749         if (ret)
6750                 goto out_register_late_subsys;
6751
6752 #if IS_BUILTIN(CONFIG_IPV6)
6753 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6754         ret = bpf_iter_register();
6755         if (ret)
6756                 goto out_register_late_subsys;
6757 #endif
6758 #endif
6759
6760         for_each_possible_cpu(cpu) {
6761                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6762
6763                 INIT_LIST_HEAD(&ul->head);
6764                 INIT_LIST_HEAD(&ul->quarantine);
6765                 spin_lock_init(&ul->lock);
6766         }
6767
6768 out:
6769         return ret;
6770
6771 out_register_late_subsys:
6772         rtnl_unregister_all(PF_INET6);
6773         unregister_pernet_subsys(&ip6_route_net_late_ops);
6774 fib6_rules_init:
6775         fib6_rules_cleanup();
6776 xfrm6_init:
6777         xfrm6_fini();
6778 out_fib6_init:
6779         fib6_gc_cleanup();
6780 out_register_subsys:
6781         unregister_pernet_subsys(&ip6_route_net_ops);
6782 out_register_inetpeer:
6783         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6784 out_dst_entries:
6785         dst_entries_destroy(&ip6_dst_blackhole_ops);
6786 out_kmem_cache:
6787         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6788         goto out;
6789 }
6790
6791 void ip6_route_cleanup(void)
6792 {
6793 #if IS_BUILTIN(CONFIG_IPV6)
6794 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6795         bpf_iter_unregister();
6796 #endif
6797 #endif
6798         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6799         unregister_pernet_subsys(&ip6_route_net_late_ops);
6800         fib6_rules_cleanup();
6801         xfrm6_fini();
6802         fib6_gc_cleanup();
6803         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6804         unregister_pernet_subsys(&ip6_route_net_ops);
6805         dst_entries_destroy(&ip6_dst_blackhole_ops);
6806         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6807 }