net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <linux/siphash.h>
  45 #include <net/net_namespace.h>
  46 #include <net/snmp.h>
  47 #include <net/ipv6.h>
  48 #include <net/ip6_fib.h>
  49 #include <net/ip6_route.h>
  50 #include <net/ndisc.h>
  51 #include <net/addrconf.h>
  52 #include <net/tcp.h>
  53 #include <linux/rtnetlink.h>
  54 #include <net/dst.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/xfrm.h>
  57 #include <net/netevent.h>
  58 #include <net/netlink.h>
  59 #include <net/rtnh.h>
  60 #include <net/lwtunnel.h>
  61 #include <net/ip_tunnels.h>
  62 #include <net/l3mdev.h>
  63 #include <net/ip.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/btf_ids.h>
  66
  67 #ifdef CONFIG_SYSCTL
  68 #include <linux/sysctl.h>
  69 #endif
  70
  71 static int ip6_rt_type_to_error(u8 fib6_type);
  72
  73 #define CREATE_TRACE_POINTS
  74 #include <trace/events/fib6.h>
  75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  76 #undef CREATE_TRACE_POINTS
  77
  78 enum rt6_nud_state {
  79         RT6_NUD_FAIL_HARD = -3,
  80         RT6_NUD_FAIL_PROBE = -2,
  81         RT6_NUD_FAIL_DO_RR = -1,
  82         RT6_NUD_SUCCEED = 1
  83 };
  84
  85 INDIRECT_CALLABLE_SCOPE
  86 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  87 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  88 INDIRECT_CALLABLE_SCOPE
  89 unsigned int            ip6_mtu(const struct dst_entry *dst);
  90 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  91 static void             ip6_dst_destroy(struct dst_entry *);
  92 static void             ip6_dst_ifdown(struct dst_entry *,
  93                                        struct net_device *dev, int how);
  94 static void              ip6_dst_gc(struct dst_ops *ops);
  95
  96 static int              ip6_pkt_discard(struct sk_buff *skb);
  97 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  98 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  99 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
 100 static void             ip6_link_failure(struct sk_buff *skb);
 101 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 102                                            struct sk_buff *skb, u32 mtu,
 103                                            bool confirm_neigh);
 104 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 105                                         struct sk_buff *skb);
 106 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 107                            int strict);
 108 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 109 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 110                          struct fib6_info *rt, struct dst_entry *dst,
 111                          struct in6_addr *dest, struct in6_addr *src,
 112                          int iif, int type, u32 portid, u32 seq,
 113                          unsigned int flags);
 114 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 115                                            const struct in6_addr *daddr,
 116                                            const struct in6_addr *saddr);
 117
 118 #ifdef CONFIG_IPV6_ROUTE_INFO
 119 static struct fib6_info *rt6_add_route_info(struct net *net,
 120                                            const struct in6_addr *prefix, int prefixlen,
 121                                            const struct in6_addr *gwaddr,
 122                                            struct net_device *dev,
 123                                            unsigned int pref);
 124 static struct fib6_info *rt6_get_route_info(struct net *net,
 125                                            const struct in6_addr *prefix, int prefixlen,
 126                                            const struct in6_addr *gwaddr,
 127                                            struct net_device *dev);
 128 #endif
 129
 130 struct uncached_list {
 131         spinlock_t              lock;
 132         struct list_head        head;
 133         struct list_head        quarantine;
 134 };
 135
 136 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 137
 138 void rt6_uncached_list_add(struct rt6_info *rt)
 139 {
 140         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 141
 142         rt->dst.rt_uncached_list = ul;
 143
 144         spin_lock_bh(&ul->lock);
 145         list_add_tail(&rt->dst.rt_uncached, &ul->head);
 146         spin_unlock_bh(&ul->lock);
 147 }
 148
 149 void rt6_uncached_list_del(struct rt6_info *rt)
 150 {
 151         if (!list_empty(&rt->dst.rt_uncached)) {
 152                 struct uncached_list *ul = rt->dst.rt_uncached_list;
 153
 154                 spin_lock_bh(&ul->lock);
 155                 list_del_init(&rt->dst.rt_uncached);
 156                 spin_unlock_bh(&ul->lock);
 157         }
 158 }
 159
 160 static void rt6_uncached_list_flush_dev(struct net_device *dev)
 161 {
 162         int cpu;
 163
 164         for_each_possible_cpu(cpu) {
 165                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 166                 struct rt6_info *rt, *safe;
 167
 168                 if (list_empty(&ul->head))
 169                         continue;
 170
 171                 spin_lock_bh(&ul->lock);
 172                 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
 173                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 174                         struct net_device *rt_dev = rt->dst.dev;
 175                         bool handled = false;
 176
 177                         if (rt_idev->dev == dev) {
 178                                 rt->rt6i_idev = in6_dev_get(blackhole_netdev);
 179                                 in6_dev_put(rt_idev);
 180                                 handled = true;
 181                         }
 182
 183                         if (rt_dev == dev) {
 184                                 rt->dst.dev = blackhole_netdev;
 185                                 netdev_ref_replace(rt_dev, blackhole_netdev,
 186                                                    &rt->dst.dev_tracker,
 187                                                    GFP_ATOMIC);
 188                                 handled = true;
 189                         }
 190                         if (handled)
 191                                 list_move(&rt->dst.rt_uncached,
 192                                           &ul->quarantine);
 193                 }
 194                 spin_unlock_bh(&ul->lock);
 195         }
 196 }
 197
 198 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 199                                              struct sk_buff *skb,
 200                                              const void *daddr)
 201 {
 202         if (!ipv6_addr_any(p))
 203                 return (const void *) p;
 204         else if (skb)
 205                 return &ipv6_hdr(skb)->daddr;
 206         return daddr;
 207 }
 208
 209 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 210                                    struct net_device *dev,
 211                                    struct sk_buff *skb,
 212                                    const void *daddr)
 213 {
 214         struct neighbour *n;
 215
 216         daddr = choose_neigh_daddr(gw, skb, daddr);
 217         n = __ipv6_neigh_lookup(dev, daddr);
 218         if (n)
 219                 return n;
 220
 221         n = neigh_create(&nd_tbl, daddr, dev);
 222         return IS_ERR(n) ? NULL : n;
 223 }
 224
 225 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 226                                               struct sk_buff *skb,
 227                                               const void *daddr)
 228 {
 229         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 230
 231         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 232                                 dst->dev, skb, daddr);
 233 }
 234
 235 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 236 {
 237         struct net_device *dev = dst->dev;
 238         struct rt6_info *rt = (struct rt6_info *)dst;
 239
 240         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 241         if (!daddr)
 242                 return;
 243         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 244                 return;
 245         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 246                 return;
 247         __ipv6_confirm_neigh(dev, daddr);
 248 }
 249
 250 static struct dst_ops ip6_dst_ops_template = {
 251         .family                 =       AF_INET6,
 252         .gc                     =       ip6_dst_gc,
 253         .gc_thresh              =       1024,
 254         .check                  =       ip6_dst_check,
 255         .default_advmss         =       ip6_default_advmss,
 256         .mtu                    =       ip6_mtu,
 257         .cow_metrics            =       dst_cow_metrics_generic,
 258         .destroy                =       ip6_dst_destroy,
 259         .ifdown                 =       ip6_dst_ifdown,
 260         .negative_advice        =       ip6_negative_advice,
 261         .link_failure           =       ip6_link_failure,
 262         .update_pmtu            =       ip6_rt_update_pmtu,
 263         .redirect               =       rt6_do_redirect,
 264         .local_out              =       __ip6_local_out,
 265         .neigh_lookup           =       ip6_dst_neigh_lookup,
 266         .confirm_neigh          =       ip6_confirm_neigh,
 267 };
 268
 269 static struct dst_ops ip6_dst_blackhole_ops = {
 270         .family                 = AF_INET6,
 271         .default_advmss         = ip6_default_advmss,
 272         .neigh_lookup           = ip6_dst_neigh_lookup,
 273         .check                  = ip6_dst_check,
 274         .destroy                = ip6_dst_destroy,
 275         .cow_metrics            = dst_cow_metrics_generic,
 276         .update_pmtu            = dst_blackhole_update_pmtu,
 277         .redirect               = dst_blackhole_redirect,
 278         .mtu                    = dst_blackhole_mtu,
 279 };
 280
 281 static const u32 ip6_template_metrics[RTAX_MAX] = {
 282         [RTAX_HOPLIMIT - 1] = 0,
 283 };
 284
 285 static const struct fib6_info fib6_null_entry_template = {
 286         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 287         .fib6_protocol  = RTPROT_KERNEL,
 288         .fib6_metric    = ~(u32)0,
 289         .fib6_ref       = REFCOUNT_INIT(1),
 290         .fib6_type      = RTN_UNREACHABLE,
 291         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 292 };
 293
 294 static const struct rt6_info ip6_null_entry_template = {
 295         .dst = {
 296                 .__rcuref       = RCUREF_INIT(1),
 297                 .__use          = 1,
 298                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 299                 .error          = -ENETUNREACH,
 300                 .input          = ip6_pkt_discard,
 301                 .output         = ip6_pkt_discard_out,
 302         },
 303         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 304 };
 305
 306 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 307
 308 static const struct rt6_info ip6_prohibit_entry_template = {
 309         .dst = {
 310                 .__rcuref       = RCUREF_INIT(1),
 311                 .__use          = 1,
 312                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 313                 .error          = -EACCES,
 314                 .input          = ip6_pkt_prohibit,
 315                 .output         = ip6_pkt_prohibit_out,
 316         },
 317         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 318 };
 319
 320 static const struct rt6_info ip6_blk_hole_entry_template = {
 321         .dst = {
 322                 .__rcuref       = RCUREF_INIT(1),
 323                 .__use          = 1,
 324                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 325                 .error          = -EINVAL,
 326                 .input          = dst_discard,
 327                 .output         = dst_discard_out,
 328         },
 329         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 330 };
 331
 332 #endif
 333
 334 static void rt6_info_init(struct rt6_info *rt)
 335 {
 336         memset_after(rt, 0, dst);
 337 }
 338
 339 /* allocate dst with ip6_dst_ops */
 340 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 341                                int flags)
 342 {
 343         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 344                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 345
 346         if (rt) {
 347                 rt6_info_init(rt);
 348                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 349         }
 350
 351         return rt;
 352 }
 353 EXPORT_SYMBOL(ip6_dst_alloc);
 354
 355 static void ip6_dst_destroy(struct dst_entry *dst)
 356 {
 357         struct rt6_info *rt = (struct rt6_info *)dst;
 358         struct fib6_info *from;
 359         struct inet6_dev *idev;
 360
 361         ip_dst_metrics_put(dst);
 362         rt6_uncached_list_del(rt);
 363
 364         idev = rt->rt6i_idev;
 365         if (idev) {
 366                 rt->rt6i_idev = NULL;
 367                 in6_dev_put(idev);
 368         }
 369
 370         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 371         fib6_info_release(from);
 372 }
 373
 374 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 375                            int how)
 376 {
 377         struct rt6_info *rt = (struct rt6_info *)dst;
 378         struct inet6_dev *idev = rt->rt6i_idev;
 379
 380         if (idev && idev->dev != blackhole_netdev) {
 381                 struct inet6_dev *blackhole_idev = in6_dev_get(blackhole_netdev);
 382
 383                 if (blackhole_idev) {
 384                         rt->rt6i_idev = blackhole_idev;
 385                         in6_dev_put(idev);
 386                 }
 387         }
 388 }
 389
 390 static bool __rt6_check_expired(const struct rt6_info *rt)
 391 {
 392         if (rt->rt6i_flags & RTF_EXPIRES)
 393                 return time_after(jiffies, rt->dst.expires);
 394         else
 395                 return false;
 396 }
 397
 398 static bool rt6_check_expired(const struct rt6_info *rt)
 399 {
 400         struct fib6_info *from;
 401
 402         from = rcu_dereference(rt->from);
 403
 404         if (rt->rt6i_flags & RTF_EXPIRES) {
 405                 if (time_after(jiffies, rt->dst.expires))
 406                         return true;
 407         } else if (from) {
 408                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 409                         fib6_check_expired(from);
 410         }
 411         return false;
 412 }
 413
 414 void fib6_select_path(const struct net *net, struct fib6_result *res,
 415                       struct flowi6 *fl6, int oif, bool have_oif_match,
 416                       const struct sk_buff *skb, int strict)
 417 {
 418         struct fib6_info *sibling, *next_sibling;
 419         struct fib6_info *match = res->f6i;
 420
 421         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 422                 goto out;
 423
 424         if (match->nh && have_oif_match && res->nh)
 425                 return;
 426
 427         /* We might have already computed the hash for ICMPv6 errors. In such
 428          * case it will always be non-zero. Otherwise now is the time to do it.
 429          */
 430         if (!fl6->mp_hash &&
 431             (!match->nh || nexthop_is_multipath(match->nh)))
 432                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 433
 434         if (unlikely(match->nh)) {
 435                 nexthop_path_fib6_result(res, fl6->mp_hash);
 436                 return;
 437         }
 438
 439         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 440                 goto out;
 441
 442         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 443                                  fib6_siblings) {
 444                 const struct fib6_nh *nh = sibling->fib6_nh;
 445                 int nh_upper_bound;
 446
 447                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 448                 if (fl6->mp_hash > nh_upper_bound)
 449                         continue;
 450                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 451                         break;
 452                 match = sibling;
 453                 break;
 454         }
 455
 456 out:
 457         res->f6i = match;
 458         res->nh = match->fib6_nh;
 459 }
 460
 461 /*
 462  *      Route lookup. rcu_read_lock() should be held.
 463  */
 464
 465 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 466                                const struct in6_addr *saddr, int oif, int flags)
 467 {
 468         const struct net_device *dev;
 469
 470         if (nh->fib_nh_flags & RTNH_F_DEAD)
 471                 return false;
 472
 473         dev = nh->fib_nh_dev;
 474         if (oif) {
 475                 if (dev->ifindex == oif)
 476                         return true;
 477         } else {
 478                 if (ipv6_chk_addr(net, saddr, dev,
 479                                   flags & RT6_LOOKUP_F_IFACE))
 480                         return true;
 481         }
 482
 483         return false;
 484 }
 485
 486 struct fib6_nh_dm_arg {
 487         struct net              *net;
 488         const struct in6_addr   *saddr;
 489         int                     oif;
 490         int                     flags;
 491         struct fib6_nh          *nh;
 492 };
 493
 494 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 495 {
 496         struct fib6_nh_dm_arg *arg = _arg;
 497
 498         arg->nh = nh;
 499         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 500                                   arg->flags);
 501 }
 502
 503 /* returns fib6_nh from nexthop or NULL */
 504 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 505                                         struct fib6_result *res,
 506                                         const struct in6_addr *saddr,
 507                                         int oif, int flags)
 508 {
 509         struct fib6_nh_dm_arg arg = {
 510                 .net   = net,
 511                 .saddr = saddr,
 512                 .oif   = oif,
 513                 .flags = flags,
 514         };
 515
 516         if (nexthop_is_blackhole(nh))
 517                 return NULL;
 518
 519         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 520                 return arg.nh;
 521
 522         return NULL;
 523 }
 524
 525 static void rt6_device_match(struct net *net, struct fib6_result *res,
 526                              const struct in6_addr *saddr, int oif, int flags)
 527 {
 528         struct fib6_info *f6i = res->f6i;
 529         struct fib6_info *spf6i;
 530         struct fib6_nh *nh;
 531
 532         if (!oif && ipv6_addr_any(saddr)) {
 533                 if (unlikely(f6i->nh)) {
 534                         nh = nexthop_fib6_nh(f6i->nh);
 535                         if (nexthop_is_blackhole(f6i->nh))
 536                                 goto out_blackhole;
 537                 } else {
 538                         nh = f6i->fib6_nh;
 539                 }
 540                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 541                         goto out;
 542         }
 543
 544         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 545                 bool matched = false;
 546
 547                 if (unlikely(spf6i->nh)) {
 548                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 549                                               oif, flags);
 550                         if (nh)
 551                                 matched = true;
 552                 } else {
 553                         nh = spf6i->fib6_nh;
 554                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 555                                 matched = true;
 556                 }
 557                 if (matched) {
 558                         res->f6i = spf6i;
 559                         goto out;
 560                 }
 561         }
 562
 563         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 564                 res->f6i = net->ipv6.fib6_null_entry;
 565                 nh = res->f6i->fib6_nh;
 566                 goto out;
 567         }
 568
 569         if (unlikely(f6i->nh)) {
 570                 nh = nexthop_fib6_nh(f6i->nh);
 571                 if (nexthop_is_blackhole(f6i->nh))
 572                         goto out_blackhole;
 573         } else {
 574                 nh = f6i->fib6_nh;
 575         }
 576
 577         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 578                 res->f6i = net->ipv6.fib6_null_entry;
 579                 nh = res->f6i->fib6_nh;
 580         }
 581 out:
 582         res->nh = nh;
 583         res->fib6_type = res->f6i->fib6_type;
 584         res->fib6_flags = res->f6i->fib6_flags;
 585         return;
 586
 587 out_blackhole:
 588         res->fib6_flags |= RTF_REJECT;
 589         res->fib6_type = RTN_BLACKHOLE;
 590         res->nh = nh;
 591 }
 592
 593 #ifdef CONFIG_IPV6_ROUTER_PREF
 594 struct __rt6_probe_work {
 595         struct work_struct work;
 596         struct in6_addr target;
 597         struct net_device *dev;
 598         netdevice_tracker dev_tracker;
 599 };
 600
 601 static void rt6_probe_deferred(struct work_struct *w)
 602 {
 603         struct in6_addr mcaddr;
 604         struct __rt6_probe_work *work =
 605                 container_of(w, struct __rt6_probe_work, work);
 606
 607         addrconf_addr_solict_mult(&work->target, &mcaddr);
 608         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 609         netdev_put(work->dev, &work->dev_tracker);
 610         kfree(work);
 611 }
 612
 613 static void rt6_probe(struct fib6_nh *fib6_nh)
 614 {
 615         struct __rt6_probe_work *work = NULL;
 616         const struct in6_addr *nh_gw;
 617         unsigned long last_probe;
 618         struct neighbour *neigh;
 619         struct net_device *dev;
 620         struct inet6_dev *idev;
 621
 622         /*
 623          * Okay, this does not seem to be appropriate
 624          * for now, however, we need to check if it
 625          * is really so; aka Router Reachability Probing.
 626          *
 627          * Router Reachability Probe MUST be rate-limited
 628          * to no more than one per minute.
 629          */
 630         if (!fib6_nh->fib_nh_gw_family)
 631                 return;
 632
 633         nh_gw = &fib6_nh->fib_nh_gw6;
 634         dev = fib6_nh->fib_nh_dev;
 635         rcu_read_lock();
 636         last_probe = READ_ONCE(fib6_nh->last_probe);
 637         idev = __in6_dev_get(dev);
 638         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 639         if (neigh) {
 640                 if (READ_ONCE(neigh->nud_state) & NUD_VALID)
 641                         goto out;
 642
 643                 write_lock_bh(&neigh->lock);
 644                 if (!(neigh->nud_state & NUD_VALID) &&
 645                     time_after(jiffies,
 646                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 647                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 648                         if (work)
 649                                 __neigh_set_probe_once(neigh);
 650                 }
 651                 write_unlock_bh(&neigh->lock);
 652         } else if (time_after(jiffies, last_probe +
 653                                        idev->cnf.rtr_probe_interval)) {
 654                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 655         }
 656
 657         if (!work || cmpxchg(&fib6_nh->last_probe,
 658                              last_probe, jiffies) != last_probe) {
 659                 kfree(work);
 660         } else {
 661                 INIT_WORK(&work->work, rt6_probe_deferred);
 662                 work->target = *nh_gw;
 663                 netdev_hold(dev, &work->dev_tracker, GFP_ATOMIC);
 664                 work->dev = dev;
 665                 schedule_work(&work->work);
 666         }
 667
 668 out:
 669         rcu_read_unlock();
 670 }
 671 #else
 672 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 673 {
 674 }
 675 #endif
 676
 677 /*
 678  * Default Router Selection (RFC 2461 6.3.6)
 679  */
 680 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 681 {
 682         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 683         struct neighbour *neigh;
 684
 685         rcu_read_lock();
 686         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 687                                           &fib6_nh->fib_nh_gw6);
 688         if (neigh) {
 689                 u8 nud_state = READ_ONCE(neigh->nud_state);
 690
 691                 if (nud_state & NUD_VALID)
 692                         ret = RT6_NUD_SUCCEED;
 693 #ifdef CONFIG_IPV6_ROUTER_PREF
 694                 else if (!(nud_state & NUD_FAILED))
 695                         ret = RT6_NUD_SUCCEED;
 696                 else
 697                         ret = RT6_NUD_FAIL_PROBE;
 698 #endif
 699         } else {
 700                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 701                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 702         }
 703         rcu_read_unlock();
 704
 705         return ret;
 706 }
 707
 708 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 709                            int strict)
 710 {
 711         int m = 0;
 712
 713         if (!oif || nh->fib_nh_dev->ifindex == oif)
 714                 m = 2;
 715
 716         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 717                 return RT6_NUD_FAIL_HARD;
 718 #ifdef CONFIG_IPV6_ROUTER_PREF
 719         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 720 #endif
 721         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 722             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 723                 int n = rt6_check_neigh(nh);
 724                 if (n < 0)
 725                         return n;
 726         }
 727         return m;
 728 }
 729
 730 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 731                        int oif, int strict, int *mpri, bool *do_rr)
 732 {
 733         bool match_do_rr = false;
 734         bool rc = false;
 735         int m;
 736
 737         if (nh->fib_nh_flags & RTNH_F_DEAD)
 738                 goto out;
 739
 740         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 741             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 742             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 743                 goto out;
 744
 745         m = rt6_score_route(nh, fib6_flags, oif, strict);
 746         if (m == RT6_NUD_FAIL_DO_RR) {
 747                 match_do_rr = true;
 748                 m = 0; /* lowest valid score */
 749         } else if (m == RT6_NUD_FAIL_HARD) {
 750                 goto out;
 751         }
 752
 753         if (strict & RT6_LOOKUP_F_REACHABLE)
 754                 rt6_probe(nh);
 755
 756         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 757         if (m > *mpri) {
 758                 *do_rr = match_do_rr;
 759                 *mpri = m;
 760                 rc = true;
 761         }
 762 out:
 763         return rc;
 764 }
 765
 766 struct fib6_nh_frl_arg {
 767         u32             flags;
 768         int             oif;
 769         int             strict;
 770         int             *mpri;
 771         bool            *do_rr;
 772         struct fib6_nh  *nh;
 773 };
 774
 775 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 776 {
 777         struct fib6_nh_frl_arg *arg = _arg;
 778
 779         arg->nh = nh;
 780         return find_match(nh, arg->flags, arg->oif, arg->strict,
 781                           arg->mpri, arg->do_rr);
 782 }
 783
 784 static void __find_rr_leaf(struct fib6_info *f6i_start,
 785                            struct fib6_info *nomatch, u32 metric,
 786                            struct fib6_result *res, struct fib6_info **cont,
 787                            int oif, int strict, bool *do_rr, int *mpri)
 788 {
 789         struct fib6_info *f6i;
 790
 791         for (f6i = f6i_start;
 792              f6i && f6i != nomatch;
 793              f6i = rcu_dereference(f6i->fib6_next)) {
 794                 bool matched = false;
 795                 struct fib6_nh *nh;
 796
 797                 if (cont && f6i->fib6_metric != metric) {
 798                         *cont = f6i;
 799                         return;
 800                 }
 801
 802                 if (fib6_check_expired(f6i))
 803                         continue;
 804
 805                 if (unlikely(f6i->nh)) {
 806                         struct fib6_nh_frl_arg arg = {
 807                                 .flags  = f6i->fib6_flags,
 808                                 .oif    = oif,
 809                                 .strict = strict,
 810                                 .mpri   = mpri,
 811                                 .do_rr  = do_rr
 812                         };
 813
 814                         if (nexthop_is_blackhole(f6i->nh)) {
 815                                 res->fib6_flags = RTF_REJECT;
 816                                 res->fib6_type = RTN_BLACKHOLE;
 817                                 res->f6i = f6i;
 818                                 res->nh = nexthop_fib6_nh(f6i->nh);
 819                                 return;
 820                         }
 821                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 822                                                      &arg)) {
 823                                 matched = true;
 824                                 nh = arg.nh;
 825                         }
 826                 } else {
 827                         nh = f6i->fib6_nh;
 828                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 829                                        mpri, do_rr))
 830                                 matched = true;
 831                 }
 832                 if (matched) {
 833                         res->f6i = f6i;
 834                         res->nh = nh;
 835                         res->fib6_flags = f6i->fib6_flags;
 836                         res->fib6_type = f6i->fib6_type;
 837                 }
 838         }
 839 }
 840
 841 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 842                          struct fib6_info *rr_head, int oif, int strict,
 843                          bool *do_rr, struct fib6_result *res)
 844 {
 845         u32 metric = rr_head->fib6_metric;
 846         struct fib6_info *cont = NULL;
 847         int mpri = -1;
 848
 849         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 850                        oif, strict, do_rr, &mpri);
 851
 852         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 853                        oif, strict, do_rr, &mpri);
 854
 855         if (res->f6i || !cont)
 856                 return;
 857
 858         __find_rr_leaf(cont, NULL, metric, res, NULL,
 859                        oif, strict, do_rr, &mpri);
 860 }
 861
 862 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 863                        struct fib6_result *res, int strict)
 864 {
 865         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 866         struct fib6_info *rt0;
 867         bool do_rr = false;
 868         int key_plen;
 869
 870         /* make sure this function or its helpers sets f6i */
 871         res->f6i = NULL;
 872
 873         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 874                 goto out;
 875
 876         rt0 = rcu_dereference(fn->rr_ptr);
 877         if (!rt0)
 878                 rt0 = leaf;
 879
 880         /* Double check to make sure fn is not an intermediate node
 881          * and fn->leaf does not points to its child's leaf
 882          * (This might happen if all routes under fn are deleted from
 883          * the tree and fib6_repair_tree() is called on the node.)
 884          */
 885         key_plen = rt0->fib6_dst.plen;
 886 #ifdef CONFIG_IPV6_SUBTREES
 887         if (rt0->fib6_src.plen)
 888                 key_plen = rt0->fib6_src.plen;
 889 #endif
 890         if (fn->fn_bit != key_plen)
 891                 goto out;
 892
 893         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 894         if (do_rr) {
 895                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 896
 897                 /* no entries matched; do round-robin */
 898                 if (!next || next->fib6_metric != rt0->fib6_metric)
 899                         next = leaf;
 900
 901                 if (next != rt0) {
 902                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 903                         /* make sure next is not being deleted from the tree */
 904                         if (next->fib6_node)
 905                                 rcu_assign_pointer(fn->rr_ptr, next);
 906                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 907                 }
 908         }
 909
 910 out:
 911         if (!res->f6i) {
 912                 res->f6i = net->ipv6.fib6_null_entry;
 913                 res->nh = res->f6i->fib6_nh;
 914                 res->fib6_flags = res->f6i->fib6_flags;
 915                 res->fib6_type = res->f6i->fib6_type;
 916         }
 917 }
 918
 919 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 920 {
 921         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 922                res->nh->fib_nh_gw_family;
 923 }
 924
 925 #ifdef CONFIG_IPV6_ROUTE_INFO
 926 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 927                   const struct in6_addr *gwaddr)
 928 {
 929         struct net *net = dev_net(dev);
 930         struct route_info *rinfo = (struct route_info *) opt;
 931         struct in6_addr prefix_buf, *prefix;
 932         unsigned int pref;
 933         unsigned long lifetime;
 934         struct fib6_info *rt;
 935
 936         if (len < sizeof(struct route_info)) {
 937                 return -EINVAL;
 938         }
 939
 940         /* Sanity check for prefix_len and length */
 941         if (rinfo->length > 3) {
 942                 return -EINVAL;
 943         } else if (rinfo->prefix_len > 128) {
 944                 return -EINVAL;
 945         } else if (rinfo->prefix_len > 64) {
 946                 if (rinfo->length < 2) {
 947                         return -EINVAL;
 948                 }
 949         } else if (rinfo->prefix_len > 0) {
 950                 if (rinfo->length < 1) {
 951                         return -EINVAL;
 952                 }
 953         }
 954
 955         pref = rinfo->route_pref;
 956         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 957                 return -EINVAL;
 958
 959         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 960
 961         if (rinfo->length == 3)
 962                 prefix = (struct in6_addr *)rinfo->prefix;
 963         else {
 964                 /* this function is safe */
 965                 ipv6_addr_prefix(&prefix_buf,
 966                                  (struct in6_addr *)rinfo->prefix,
 967                                  rinfo->prefix_len);
 968                 prefix = &prefix_buf;
 969         }
 970
 971         if (rinfo->prefix_len == 0)
 972                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 973         else
 974                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 975                                         gwaddr, dev);
 976
 977         if (rt && !lifetime) {
 978                 ip6_del_rt(net, rt, false);
 979                 rt = NULL;
 980         }
 981
 982         if (!rt && lifetime)
 983                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 984                                         dev, pref);
 985         else if (rt)
 986                 rt->fib6_flags = RTF_ROUTEINFO |
 987                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 988
 989         if (rt) {
 990                 if (!addrconf_finite_timeout(lifetime))
 991                         fib6_clean_expires(rt);
 992                 else
 993                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 994
 995                 fib6_info_release(rt);
 996         }
 997         return 0;
 998 }
 999 #endif
1000
1001 /*
1002  *      Misc support functions
1003  */
1004
1005 /* called with rcu_lock held */
1006 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1007 {
1008         struct net_device *dev = res->nh->fib_nh_dev;
1009
1010         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1011                 /* for copies of local routes, dst->dev needs to be the
1012                  * device if it is a master device, the master device if
1013                  * device is enslaved, and the loopback as the default
1014                  */
1015                 if (netif_is_l3_slave(dev) &&
1016                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1017                         dev = l3mdev_master_dev_rcu(dev);
1018                 else if (!netif_is_l3_master(dev))
1019                         dev = dev_net(dev)->loopback_dev;
1020                 /* last case is netif_is_l3_master(dev) is true in which
1021                  * case we want dev returned to be dev
1022                  */
1023         }
1024
1025         return dev;
1026 }
1027
1028 static const int fib6_prop[RTN_MAX + 1] = {
1029         [RTN_UNSPEC]    = 0,
1030         [RTN_UNICAST]   = 0,
1031         [RTN_LOCAL]     = 0,
1032         [RTN_BROADCAST] = 0,
1033         [RTN_ANYCAST]   = 0,
1034         [RTN_MULTICAST] = 0,
1035         [RTN_BLACKHOLE] = -EINVAL,
1036         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1037         [RTN_PROHIBIT]  = -EACCES,
1038         [RTN_THROW]     = -EAGAIN,
1039         [RTN_NAT]       = -EINVAL,
1040         [RTN_XRESOLVE]  = -EINVAL,
1041 };
1042
1043 static int ip6_rt_type_to_error(u8 fib6_type)
1044 {
1045         return fib6_prop[fib6_type];
1046 }
1047
1048 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1049 {
1050         unsigned short flags = 0;
1051
1052         if (rt->dst_nocount)
1053                 flags |= DST_NOCOUNT;
1054         if (rt->dst_nopolicy)
1055                 flags |= DST_NOPOLICY;
1056
1057         return flags;
1058 }
1059
1060 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1061 {
1062         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1063
1064         switch (fib6_type) {
1065         case RTN_BLACKHOLE:
1066                 rt->dst.output = dst_discard_out;
1067                 rt->dst.input = dst_discard;
1068                 break;
1069         case RTN_PROHIBIT:
1070                 rt->dst.output = ip6_pkt_prohibit_out;
1071                 rt->dst.input = ip6_pkt_prohibit;
1072                 break;
1073         case RTN_THROW:
1074         case RTN_UNREACHABLE:
1075         default:
1076                 rt->dst.output = ip6_pkt_discard_out;
1077                 rt->dst.input = ip6_pkt_discard;
1078                 break;
1079         }
1080 }
1081
1082 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1083 {
1084         struct fib6_info *f6i = res->f6i;
1085
1086         if (res->fib6_flags & RTF_REJECT) {
1087                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1088                 return;
1089         }
1090
1091         rt->dst.error = 0;
1092         rt->dst.output = ip6_output;
1093
1094         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1095                 rt->dst.input = ip6_input;
1096         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1097                 rt->dst.input = ip6_mc_input;
1098         } else {
1099                 rt->dst.input = ip6_forward;
1100         }
1101
1102         if (res->nh->fib_nh_lws) {
1103                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1104                 lwtunnel_set_redirect(&rt->dst);
1105         }
1106
1107         rt->dst.lastuse = jiffies;
1108 }
1109
1110 /* Caller must already hold reference to @from */
1111 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1112 {
1113         rt->rt6i_flags &= ~RTF_EXPIRES;
1114         rcu_assign_pointer(rt->from, from);
1115         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1116 }
1117
1118 /* Caller must already hold reference to f6i in result */
1119 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1120 {
1121         const struct fib6_nh *nh = res->nh;
1122         const struct net_device *dev = nh->fib_nh_dev;
1123         struct fib6_info *f6i = res->f6i;
1124
1125         ip6_rt_init_dst(rt, res);
1126
1127         rt->rt6i_dst = f6i->fib6_dst;
1128         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1129         rt->rt6i_flags = res->fib6_flags;
1130         if (nh->fib_nh_gw_family) {
1131                 rt->rt6i_gateway = nh->fib_nh_gw6;
1132                 rt->rt6i_flags |= RTF_GATEWAY;
1133         }
1134         rt6_set_from(rt, f6i);
1135 #ifdef CONFIG_IPV6_SUBTREES
1136         rt->rt6i_src = f6i->fib6_src;
1137 #endif
1138 }
1139
1140 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1141                                         struct in6_addr *saddr)
1142 {
1143         struct fib6_node *pn, *sn;
1144         while (1) {
1145                 if (fn->fn_flags & RTN_TL_ROOT)
1146                         return NULL;
1147                 pn = rcu_dereference(fn->parent);
1148                 sn = FIB6_SUBTREE(pn);
1149                 if (sn && sn != fn)
1150                         fn = fib6_node_lookup(sn, NULL, saddr);
1151                 else
1152                         fn = pn;
1153                 if (fn->fn_flags & RTN_RTINFO)
1154                         return fn;
1155         }
1156 }
1157
1158 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1159 {
1160         struct rt6_info *rt = *prt;
1161
1162         if (dst_hold_safe(&rt->dst))
1163                 return true;
1164         if (net) {
1165                 rt = net->ipv6.ip6_null_entry;
1166                 dst_hold(&rt->dst);
1167         } else {
1168                 rt = NULL;
1169         }
1170         *prt = rt;
1171         return false;
1172 }
1173
1174 /* called with rcu_lock held */
1175 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1176 {
1177         struct net_device *dev = res->nh->fib_nh_dev;
1178         struct fib6_info *f6i = res->f6i;
1179         unsigned short flags;
1180         struct rt6_info *nrt;
1181
1182         if (!fib6_info_hold_safe(f6i))
1183                 goto fallback;
1184
1185         flags = fib6_info_dst_flags(f6i);
1186         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1187         if (!nrt) {
1188                 fib6_info_release(f6i);
1189                 goto fallback;
1190         }
1191
1192         ip6_rt_copy_init(nrt, res);
1193         return nrt;
1194
1195 fallback:
1196         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1197         dst_hold(&nrt->dst);
1198         return nrt;
1199 }
1200
1201 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1202                                              struct fib6_table *table,
1203                                              struct flowi6 *fl6,
1204                                              const struct sk_buff *skb,
1205                                              int flags)
1206 {
1207         struct fib6_result res = {};
1208         struct fib6_node *fn;
1209         struct rt6_info *rt;
1210
1211         rcu_read_lock();
1212         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1213 restart:
1214         res.f6i = rcu_dereference(fn->leaf);
1215         if (!res.f6i)
1216                 res.f6i = net->ipv6.fib6_null_entry;
1217         else
1218                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1219                                  flags);
1220
1221         if (res.f6i == net->ipv6.fib6_null_entry) {
1222                 fn = fib6_backtrack(fn, &fl6->saddr);
1223                 if (fn)
1224                         goto restart;
1225
1226                 rt = net->ipv6.ip6_null_entry;
1227                 dst_hold(&rt->dst);
1228                 goto out;
1229         } else if (res.fib6_flags & RTF_REJECT) {
1230                 goto do_create;
1231         }
1232
1233         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1234                          fl6->flowi6_oif != 0, skb, flags);
1235
1236         /* Search through exception table */
1237         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1238         if (rt) {
1239                 if (ip6_hold_safe(net, &rt))
1240                         dst_use_noref(&rt->dst, jiffies);
1241         } else {
1242 do_create:
1243                 rt = ip6_create_rt_rcu(&res);
1244         }
1245
1246 out:
1247         trace_fib6_table_lookup(net, &res, table, fl6);
1248
1249         rcu_read_unlock();
1250
1251         return rt;
1252 }
1253
1254 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1255                                    const struct sk_buff *skb, int flags)
1256 {
1257         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1258 }
1259 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1260
1261 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1262                             const struct in6_addr *saddr, int oif,
1263                             const struct sk_buff *skb, int strict)
1264 {
1265         struct flowi6 fl6 = {
1266                 .flowi6_oif = oif,
1267                 .daddr = *daddr,
1268         };
1269         struct dst_entry *dst;
1270         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1271
1272         if (saddr) {
1273                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1274                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1275         }
1276
1277         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1278         if (dst->error == 0)
1279                 return (struct rt6_info *) dst;
1280
1281         dst_release(dst);
1282
1283         return NULL;
1284 }
1285 EXPORT_SYMBOL(rt6_lookup);
1286
1287 /* ip6_ins_rt is called with FREE table->tb6_lock.
1288  * It takes new route entry, the addition fails by any reason the
1289  * route is released.
1290  * Caller must hold dst before calling it.
1291  */
1292
1293 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1294                         struct netlink_ext_ack *extack)
1295 {
1296         int err;
1297         struct fib6_table *table;
1298
1299         table = rt->fib6_table;
1300         spin_lock_bh(&table->tb6_lock);
1301         err = fib6_add(&table->tb6_root, rt, info, extack);
1302         spin_unlock_bh(&table->tb6_lock);
1303
1304         return err;
1305 }
1306
1307 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1308 {
1309         struct nl_info info = { .nl_net = net, };
1310
1311         return __ip6_ins_rt(rt, &info, NULL);
1312 }
1313
1314 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1315                                            const struct in6_addr *daddr,
1316                                            const struct in6_addr *saddr)
1317 {
1318         struct fib6_info *f6i = res->f6i;
1319         struct net_device *dev;
1320         struct rt6_info *rt;
1321
1322         /*
1323          *      Clone the route.
1324          */
1325
1326         if (!fib6_info_hold_safe(f6i))
1327                 return NULL;
1328
1329         dev = ip6_rt_get_dev_rcu(res);
1330         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1331         if (!rt) {
1332                 fib6_info_release(f6i);
1333                 return NULL;
1334         }
1335
1336         ip6_rt_copy_init(rt, res);
1337         rt->rt6i_flags |= RTF_CACHE;
1338         rt->rt6i_dst.addr = *daddr;
1339         rt->rt6i_dst.plen = 128;
1340
1341         if (!rt6_is_gw_or_nonexthop(res)) {
1342                 if (f6i->fib6_dst.plen != 128 &&
1343                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1344                         rt->rt6i_flags |= RTF_ANYCAST;
1345 #ifdef CONFIG_IPV6_SUBTREES
1346                 if (rt->rt6i_src.plen && saddr) {
1347                         rt->rt6i_src.addr = *saddr;
1348                         rt->rt6i_src.plen = 128;
1349                 }
1350 #endif
1351         }
1352
1353         return rt;
1354 }
1355
1356 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1357 {
1358         struct fib6_info *f6i = res->f6i;
1359         unsigned short flags = fib6_info_dst_flags(f6i);
1360         struct net_device *dev;
1361         struct rt6_info *pcpu_rt;
1362
1363         if (!fib6_info_hold_safe(f6i))
1364                 return NULL;
1365
1366         rcu_read_lock();
1367         dev = ip6_rt_get_dev_rcu(res);
1368         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1369         rcu_read_unlock();
1370         if (!pcpu_rt) {
1371                 fib6_info_release(f6i);
1372                 return NULL;
1373         }
1374         ip6_rt_copy_init(pcpu_rt, res);
1375         pcpu_rt->rt6i_flags |= RTF_PCPU;
1376
1377         if (f6i->nh)
1378                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1379
1380         return pcpu_rt;
1381 }
1382
1383 static bool rt6_is_valid(const struct rt6_info *rt6)
1384 {
1385         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1386 }
1387
1388 /* It should be called with rcu_read_lock() acquired */
1389 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1390 {
1391         struct rt6_info *pcpu_rt;
1392
1393         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1394
1395         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1396                 struct rt6_info *prev, **p;
1397
1398                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1399                 prev = xchg(p, NULL);
1400                 if (prev) {
1401                         dst_dev_put(&prev->dst);
1402                         dst_release(&prev->dst);
1403                 }
1404
1405                 pcpu_rt = NULL;
1406         }
1407
1408         return pcpu_rt;
1409 }
1410
1411 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1412                                             const struct fib6_result *res)
1413 {
1414         struct rt6_info *pcpu_rt, *prev, **p;
1415
1416         pcpu_rt = ip6_rt_pcpu_alloc(res);
1417         if (!pcpu_rt)
1418                 return NULL;
1419
1420         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1421         prev = cmpxchg(p, NULL, pcpu_rt);
1422         BUG_ON(prev);
1423
1424         if (res->f6i->fib6_destroying) {
1425                 struct fib6_info *from;
1426
1427                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1428                 fib6_info_release(from);
1429         }
1430
1431         return pcpu_rt;
1432 }
1433
1434 /* exception hash table implementation
1435  */
1436 static DEFINE_SPINLOCK(rt6_exception_lock);
1437
1438 /* Remove rt6_ex from hash table and free the memory
1439  * Caller must hold rt6_exception_lock
1440  */
1441 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1442                                  struct rt6_exception *rt6_ex)
1443 {
1444         struct fib6_info *from;
1445         struct net *net;
1446
1447         if (!bucket || !rt6_ex)
1448                 return;
1449
1450         net = dev_net(rt6_ex->rt6i->dst.dev);
1451         net->ipv6.rt6_stats->fib_rt_cache--;
1452
1453         /* purge completely the exception to allow releasing the held resources:
1454          * some [sk] cache may keep the dst around for unlimited time
1455          */
1456         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1457         fib6_info_release(from);
1458         dst_dev_put(&rt6_ex->rt6i->dst);
1459
1460         hlist_del_rcu(&rt6_ex->hlist);
1461         dst_release(&rt6_ex->rt6i->dst);
1462         kfree_rcu(rt6_ex, rcu);
1463         WARN_ON_ONCE(!bucket->depth);
1464         bucket->depth--;
1465 }
1466
1467 /* Remove oldest rt6_ex in bucket and free the memory
1468  * Caller must hold rt6_exception_lock
1469  */
1470 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1471 {
1472         struct rt6_exception *rt6_ex, *oldest = NULL;
1473
1474         if (!bucket)
1475                 return;
1476
1477         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1478                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1479                         oldest = rt6_ex;
1480         }
1481         rt6_remove_exception(bucket, oldest);
1482 }
1483
1484 static u32 rt6_exception_hash(const struct in6_addr *dst,
1485                               const struct in6_addr *src)
1486 {
1487         static siphash_aligned_key_t rt6_exception_key;
1488         struct {
1489                 struct in6_addr dst;
1490                 struct in6_addr src;
1491         } __aligned(SIPHASH_ALIGNMENT) combined = {
1492                 .dst = *dst,
1493         };
1494         u64 val;
1495
1496         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1497
1498 #ifdef CONFIG_IPV6_SUBTREES
1499         if (src)
1500                 combined.src = *src;
1501 #endif
1502         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1503
1504         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1505 }
1506
1507 /* Helper function to find the cached rt in the hash table
1508  * and update bucket pointer to point to the bucket for this
1509  * (daddr, saddr) pair
1510  * Caller must hold rt6_exception_lock
1511  */
1512 static struct rt6_exception *
1513 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1514                               const struct in6_addr *daddr,
1515                               const struct in6_addr *saddr)
1516 {
1517         struct rt6_exception *rt6_ex;
1518         u32 hval;
1519
1520         if (!(*bucket) || !daddr)
1521                 return NULL;
1522
1523         hval = rt6_exception_hash(daddr, saddr);
1524         *bucket += hval;
1525
1526         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1527                 struct rt6_info *rt6 = rt6_ex->rt6i;
1528                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1529
1530 #ifdef CONFIG_IPV6_SUBTREES
1531                 if (matched && saddr)
1532                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1533 #endif
1534                 if (matched)
1535                         return rt6_ex;
1536         }
1537         return NULL;
1538 }
1539
1540 /* Helper function to find the cached rt in the hash table
1541  * and update bucket pointer to point to the bucket for this
1542  * (daddr, saddr) pair
1543  * Caller must hold rcu_read_lock()
1544  */
1545 static struct rt6_exception *
1546 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1547                          const struct in6_addr *daddr,
1548                          const struct in6_addr *saddr)
1549 {
1550         struct rt6_exception *rt6_ex;
1551         u32 hval;
1552
1553         WARN_ON_ONCE(!rcu_read_lock_held());
1554
1555         if (!(*bucket) || !daddr)
1556                 return NULL;
1557
1558         hval = rt6_exception_hash(daddr, saddr);
1559         *bucket += hval;
1560
1561         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1562                 struct rt6_info *rt6 = rt6_ex->rt6i;
1563                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1564
1565 #ifdef CONFIG_IPV6_SUBTREES
1566                 if (matched && saddr)
1567                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1568 #endif
1569                 if (matched)
1570                         return rt6_ex;
1571         }
1572         return NULL;
1573 }
1574
1575 static unsigned int fib6_mtu(const struct fib6_result *res)
1576 {
1577         const struct fib6_nh *nh = res->nh;
1578         unsigned int mtu;
1579
1580         if (res->f6i->fib6_pmtu) {
1581                 mtu = res->f6i->fib6_pmtu;
1582         } else {
1583                 struct net_device *dev = nh->fib_nh_dev;
1584                 struct inet6_dev *idev;
1585
1586                 rcu_read_lock();
1587                 idev = __in6_dev_get(dev);
1588                 mtu = idev->cnf.mtu6;
1589                 rcu_read_unlock();
1590         }
1591
1592         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1593
1594         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1595 }
1596
1597 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1598
1599 /* used when the flushed bit is not relevant, only access to the bucket
1600  * (ie., all bucket users except rt6_insert_exception);
1601  *
1602  * called under rcu lock; sometimes called with rt6_exception_lock held
1603  */
1604 static
1605 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1606                                                        spinlock_t *lock)
1607 {
1608         struct rt6_exception_bucket *bucket;
1609
1610         if (lock)
1611                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1612                                                    lockdep_is_held(lock));
1613         else
1614                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1615
1616         /* remove bucket flushed bit if set */
1617         if (bucket) {
1618                 unsigned long p = (unsigned long)bucket;
1619
1620                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1621                 bucket = (struct rt6_exception_bucket *)p;
1622         }
1623
1624         return bucket;
1625 }
1626
1627 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1628 {
1629         unsigned long p = (unsigned long)bucket;
1630
1631         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1632 }
1633
1634 /* called with rt6_exception_lock held */
1635 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1636                                               spinlock_t *lock)
1637 {
1638         struct rt6_exception_bucket *bucket;
1639         unsigned long p;
1640
1641         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1642                                            lockdep_is_held(lock));
1643
1644         p = (unsigned long)bucket;
1645         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1646         bucket = (struct rt6_exception_bucket *)p;
1647         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1648 }
1649
1650 static int rt6_insert_exception(struct rt6_info *nrt,
1651                                 const struct fib6_result *res)
1652 {
1653         struct net *net = dev_net(nrt->dst.dev);
1654         struct rt6_exception_bucket *bucket;
1655         struct fib6_info *f6i = res->f6i;
1656         struct in6_addr *src_key = NULL;
1657         struct rt6_exception *rt6_ex;
1658         struct fib6_nh *nh = res->nh;
1659         int max_depth;
1660         int err = 0;
1661
1662         spin_lock_bh(&rt6_exception_lock);
1663
1664         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1665                                           lockdep_is_held(&rt6_exception_lock));
1666         if (!bucket) {
1667                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1668                                  GFP_ATOMIC);
1669                 if (!bucket) {
1670                         err = -ENOMEM;
1671                         goto out;
1672                 }
1673                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1674         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1675                 err = -EINVAL;
1676                 goto out;
1677         }
1678
1679 #ifdef CONFIG_IPV6_SUBTREES
1680         /* fib6_src.plen != 0 indicates f6i is in subtree
1681          * and exception table is indexed by a hash of
1682          * both fib6_dst and fib6_src.
1683          * Otherwise, the exception table is indexed by
1684          * a hash of only fib6_dst.
1685          */
1686         if (f6i->fib6_src.plen)
1687                 src_key = &nrt->rt6i_src.addr;
1688 #endif
1689         /* rt6_mtu_change() might lower mtu on f6i.
1690          * Only insert this exception route if its mtu
1691          * is less than f6i's mtu value.
1692          */
1693         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1694                 err = -EINVAL;
1695                 goto out;
1696         }
1697
1698         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1699                                                src_key);
1700         if (rt6_ex)
1701                 rt6_remove_exception(bucket, rt6_ex);
1702
1703         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1704         if (!rt6_ex) {
1705                 err = -ENOMEM;
1706                 goto out;
1707         }
1708         rt6_ex->rt6i = nrt;
1709         rt6_ex->stamp = jiffies;
1710         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1711         bucket->depth++;
1712         net->ipv6.rt6_stats->fib_rt_cache++;
1713
1714         /* Randomize max depth to avoid some side channels attacks. */
1715         max_depth = FIB6_MAX_DEPTH + get_random_u32_below(FIB6_MAX_DEPTH);
1716         while (bucket->depth > max_depth)
1717                 rt6_exception_remove_oldest(bucket);
1718
1719 out:
1720         spin_unlock_bh(&rt6_exception_lock);
1721
1722         /* Update fn->fn_sernum to invalidate all cached dst */
1723         if (!err) {
1724                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1725                 fib6_update_sernum(net, f6i);
1726                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1727                 fib6_force_start_gc(net);
1728         }
1729
1730         return err;
1731 }
1732
1733 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1734 {
1735         struct rt6_exception_bucket *bucket;
1736         struct rt6_exception *rt6_ex;
1737         struct hlist_node *tmp;
1738         int i;
1739
1740         spin_lock_bh(&rt6_exception_lock);
1741
1742         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1743         if (!bucket)
1744                 goto out;
1745
1746         /* Prevent rt6_insert_exception() to recreate the bucket list */
1747         if (!from)
1748                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1749
1750         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1751                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1752                         if (!from ||
1753                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1754                                 rt6_remove_exception(bucket, rt6_ex);
1755                 }
1756                 WARN_ON_ONCE(!from && bucket->depth);
1757                 bucket++;
1758         }
1759 out:
1760         spin_unlock_bh(&rt6_exception_lock);
1761 }
1762
1763 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1764 {
1765         struct fib6_info *f6i = arg;
1766
1767         fib6_nh_flush_exceptions(nh, f6i);
1768
1769         return 0;
1770 }
1771
1772 void rt6_flush_exceptions(struct fib6_info *f6i)
1773 {
1774         if (f6i->nh)
1775                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1776                                          f6i);
1777         else
1778                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1779 }
1780
1781 /* Find cached rt in the hash table inside passed in rt
1782  * Caller has to hold rcu_read_lock()
1783  */
1784 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1785                                            const struct in6_addr *daddr,
1786                                            const struct in6_addr *saddr)
1787 {
1788         const struct in6_addr *src_key = NULL;
1789         struct rt6_exception_bucket *bucket;
1790         struct rt6_exception *rt6_ex;
1791         struct rt6_info *ret = NULL;
1792
1793 #ifdef CONFIG_IPV6_SUBTREES
1794         /* fib6i_src.plen != 0 indicates f6i is in subtree
1795          * and exception table is indexed by a hash of
1796          * both fib6_dst and fib6_src.
1797          * However, the src addr used to create the hash
1798          * might not be exactly the passed in saddr which
1799          * is a /128 addr from the flow.
1800          * So we need to use f6i->fib6_src to redo lookup
1801          * if the passed in saddr does not find anything.
1802          * (See the logic in ip6_rt_cache_alloc() on how
1803          * rt->rt6i_src is updated.)
1804          */
1805         if (res->f6i->fib6_src.plen)
1806                 src_key = saddr;
1807 find_ex:
1808 #endif
1809         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1810         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1811
1812         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1813                 ret = rt6_ex->rt6i;
1814
1815 #ifdef CONFIG_IPV6_SUBTREES
1816         /* Use fib6_src as src_key and redo lookup */
1817         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1818                 src_key = &res->f6i->fib6_src.addr;
1819                 goto find_ex;
1820         }
1821 #endif
1822
1823         return ret;
1824 }
1825
1826 /* Remove the passed in cached rt from the hash table that contains it */
1827 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1828                                     const struct rt6_info *rt)
1829 {
1830         const struct in6_addr *src_key = NULL;
1831         struct rt6_exception_bucket *bucket;
1832         struct rt6_exception *rt6_ex;
1833         int err;
1834
1835         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1836                 return -ENOENT;
1837
1838         spin_lock_bh(&rt6_exception_lock);
1839         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1840
1841 #ifdef CONFIG_IPV6_SUBTREES
1842         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1843          * and exception table is indexed by a hash of
1844          * both rt6i_dst and rt6i_src.
1845          * Otherwise, the exception table is indexed by
1846          * a hash of only rt6i_dst.
1847          */
1848         if (plen)
1849                 src_key = &rt->rt6i_src.addr;
1850 #endif
1851         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1852                                                &rt->rt6i_dst.addr,
1853                                                src_key);
1854         if (rt6_ex) {
1855                 rt6_remove_exception(bucket, rt6_ex);
1856                 err = 0;
1857         } else {
1858                 err = -ENOENT;
1859         }
1860
1861         spin_unlock_bh(&rt6_exception_lock);
1862         return err;
1863 }
1864
1865 struct fib6_nh_excptn_arg {
1866         struct rt6_info *rt;
1867         int             plen;
1868 };
1869
1870 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1871 {
1872         struct fib6_nh_excptn_arg *arg = _arg;
1873         int err;
1874
1875         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1876         if (err == 0)
1877                 return 1;
1878
1879         return 0;
1880 }
1881
1882 static int rt6_remove_exception_rt(struct rt6_info *rt)
1883 {
1884         struct fib6_info *from;
1885
1886         from = rcu_dereference(rt->from);
1887         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1888                 return -EINVAL;
1889
1890         if (from->nh) {
1891                 struct fib6_nh_excptn_arg arg = {
1892                         .rt = rt,
1893                         .plen = from->fib6_src.plen
1894                 };
1895                 int rc;
1896
1897                 /* rc = 1 means an entry was found */
1898                 rc = nexthop_for_each_fib6_nh(from->nh,
1899                                               rt6_nh_remove_exception_rt,
1900                                               &arg);
1901                 return rc ? 0 : -ENOENT;
1902         }
1903
1904         return fib6_nh_remove_exception(from->fib6_nh,
1905                                         from->fib6_src.plen, rt);
1906 }
1907
1908 /* Find rt6_ex which contains the passed in rt cache and
1909  * refresh its stamp
1910  */
1911 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1912                                      const struct rt6_info *rt)
1913 {
1914         const struct in6_addr *src_key = NULL;
1915         struct rt6_exception_bucket *bucket;
1916         struct rt6_exception *rt6_ex;
1917
1918         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1919 #ifdef CONFIG_IPV6_SUBTREES
1920         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1921          * and exception table is indexed by a hash of
1922          * both rt6i_dst and rt6i_src.
1923          * Otherwise, the exception table is indexed by
1924          * a hash of only rt6i_dst.
1925          */
1926         if (plen)
1927                 src_key = &rt->rt6i_src.addr;
1928 #endif
1929         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1930         if (rt6_ex)
1931                 rt6_ex->stamp = jiffies;
1932 }
1933
1934 struct fib6_nh_match_arg {
1935         const struct net_device *dev;
1936         const struct in6_addr   *gw;
1937         struct fib6_nh          *match;
1938 };
1939
1940 /* determine if fib6_nh has given device and gateway */
1941 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1942 {
1943         struct fib6_nh_match_arg *arg = _arg;
1944
1945         if (arg->dev != nh->fib_nh_dev ||
1946             (arg->gw && !nh->fib_nh_gw_family) ||
1947             (!arg->gw && nh->fib_nh_gw_family) ||
1948             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1949                 return 0;
1950
1951         arg->match = nh;
1952
1953         /* found a match, break the loop */
1954         return 1;
1955 }
1956
1957 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1958 {
1959         struct fib6_info *from;
1960         struct fib6_nh *fib6_nh;
1961
1962         rcu_read_lock();
1963
1964         from = rcu_dereference(rt->from);
1965         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1966                 goto unlock;
1967
1968         if (from->nh) {
1969                 struct fib6_nh_match_arg arg = {
1970                         .dev = rt->dst.dev,
1971                         .gw = &rt->rt6i_gateway,
1972                 };
1973
1974                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1975
1976                 if (!arg.match)
1977                         goto unlock;
1978                 fib6_nh = arg.match;
1979         } else {
1980                 fib6_nh = from->fib6_nh;
1981         }
1982         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1983 unlock:
1984         rcu_read_unlock();
1985 }
1986
1987 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1988                                          struct rt6_info *rt, int mtu)
1989 {
1990         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1991          * lowest MTU in the path: always allow updating the route PMTU to
1992          * reflect PMTU decreases.
1993          *
1994          * If the new MTU is higher, and the route PMTU is equal to the local
1995          * MTU, this means the old MTU is the lowest in the path, so allow
1996          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1997          * handle this.
1998          */
1999
2000         if (dst_mtu(&rt->dst) >= mtu)
2001                 return true;
2002
2003         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2004                 return true;
2005
2006         return false;
2007 }
2008
2009 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2010                                        const struct fib6_nh *nh, int mtu)
2011 {
2012         struct rt6_exception_bucket *bucket;
2013         struct rt6_exception *rt6_ex;
2014         int i;
2015
2016         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2017         if (!bucket)
2018                 return;
2019
2020         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2021                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2022                         struct rt6_info *entry = rt6_ex->rt6i;
2023
2024                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2025                          * route), the metrics of its rt->from have already
2026                          * been updated.
2027                          */
2028                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2029                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2030                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2031                 }
2032                 bucket++;
2033         }
2034 }
2035
2036 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2037
2038 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2039                                             const struct in6_addr *gateway)
2040 {
2041         struct rt6_exception_bucket *bucket;
2042         struct rt6_exception *rt6_ex;
2043         struct hlist_node *tmp;
2044         int i;
2045
2046         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2047                 return;
2048
2049         spin_lock_bh(&rt6_exception_lock);
2050         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2051         if (bucket) {
2052                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2053                         hlist_for_each_entry_safe(rt6_ex, tmp,
2054                                                   &bucket->chain, hlist) {
2055                                 struct rt6_info *entry = rt6_ex->rt6i;
2056
2057                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2058                                     RTF_CACHE_GATEWAY &&
2059                                     ipv6_addr_equal(gateway,
2060                                                     &entry->rt6i_gateway)) {
2061                                         rt6_remove_exception(bucket, rt6_ex);
2062                                 }
2063                         }
2064                         bucket++;
2065                 }
2066         }
2067
2068         spin_unlock_bh(&rt6_exception_lock);
2069 }
2070
2071 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2072                                       struct rt6_exception *rt6_ex,
2073                                       struct fib6_gc_args *gc_args,
2074                                       unsigned long now)
2075 {
2076         struct rt6_info *rt = rt6_ex->rt6i;
2077
2078         /* we are pruning and obsoleting aged-out and non gateway exceptions
2079          * even if others have still references to them, so that on next
2080          * dst_check() such references can be dropped.
2081          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2082          * expired, independently from their aging, as per RFC 8201 section 4
2083          */
2084         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2085                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2086                         RT6_TRACE("aging clone %p\n", rt);
2087                         rt6_remove_exception(bucket, rt6_ex);
2088                         return;
2089                 }
2090         } else if (time_after(jiffies, rt->dst.expires)) {
2091                 RT6_TRACE("purging expired route %p\n", rt);
2092                 rt6_remove_exception(bucket, rt6_ex);
2093                 return;
2094         }
2095
2096         if (rt->rt6i_flags & RTF_GATEWAY) {
2097                 struct neighbour *neigh;
2098
2099                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2100
2101                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2102                         RT6_TRACE("purging route %p via non-router but gateway\n",
2103                                   rt);
2104                         rt6_remove_exception(bucket, rt6_ex);
2105                         return;
2106                 }
2107         }
2108
2109         gc_args->more++;
2110 }
2111
2112 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2113                                    struct fib6_gc_args *gc_args,
2114                                    unsigned long now)
2115 {
2116         struct rt6_exception_bucket *bucket;
2117         struct rt6_exception *rt6_ex;
2118         struct hlist_node *tmp;
2119         int i;
2120
2121         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2122                 return;
2123
2124         rcu_read_lock_bh();
2125         spin_lock(&rt6_exception_lock);
2126         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2127         if (bucket) {
2128                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2129                         hlist_for_each_entry_safe(rt6_ex, tmp,
2130                                                   &bucket->chain, hlist) {
2131                                 rt6_age_examine_exception(bucket, rt6_ex,
2132                                                           gc_args, now);
2133                         }
2134                         bucket++;
2135                 }
2136         }
2137         spin_unlock(&rt6_exception_lock);
2138         rcu_read_unlock_bh();
2139 }
2140
2141 struct fib6_nh_age_excptn_arg {
2142         struct fib6_gc_args     *gc_args;
2143         unsigned long           now;
2144 };
2145
2146 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2147 {
2148         struct fib6_nh_age_excptn_arg *arg = _arg;
2149
2150         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2151         return 0;
2152 }
2153
2154 void rt6_age_exceptions(struct fib6_info *f6i,
2155                         struct fib6_gc_args *gc_args,
2156                         unsigned long now)
2157 {
2158         if (f6i->nh) {
2159                 struct fib6_nh_age_excptn_arg arg = {
2160                         .gc_args = gc_args,
2161                         .now = now
2162                 };
2163
2164                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2165                                          &arg);
2166         } else {
2167                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2168         }
2169 }
2170
2171 /* must be called with rcu lock held */
2172 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2173                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2174 {
2175         struct fib6_node *fn, *saved_fn;
2176
2177         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2178         saved_fn = fn;
2179
2180 redo_rt6_select:
2181         rt6_select(net, fn, oif, res, strict);
2182         if (res->f6i == net->ipv6.fib6_null_entry) {
2183                 fn = fib6_backtrack(fn, &fl6->saddr);
2184                 if (fn)
2185                         goto redo_rt6_select;
2186                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2187                         /* also consider unreachable route */
2188                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2189                         fn = saved_fn;
2190                         goto redo_rt6_select;
2191                 }
2192         }
2193
2194         trace_fib6_table_lookup(net, res, table, fl6);
2195
2196         return 0;
2197 }
2198
2199 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2200                                int oif, struct flowi6 *fl6,
2201                                const struct sk_buff *skb, int flags)
2202 {
2203         struct fib6_result res = {};
2204         struct rt6_info *rt = NULL;
2205         int strict = 0;
2206
2207         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2208                      !rcu_read_lock_held());
2209
2210         strict |= flags & RT6_LOOKUP_F_IFACE;
2211         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2212         if (net->ipv6.devconf_all->forwarding == 0)
2213                 strict |= RT6_LOOKUP_F_REACHABLE;
2214
2215         rcu_read_lock();
2216
2217         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2218         if (res.f6i == net->ipv6.fib6_null_entry)
2219                 goto out;
2220
2221         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2222
2223         /*Search through exception table */
2224         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2225         if (rt) {
2226                 goto out;
2227         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2228                             !res.nh->fib_nh_gw_family)) {
2229                 /* Create a RTF_CACHE clone which will not be
2230                  * owned by the fib6 tree.  It is for the special case where
2231                  * the daddr in the skb during the neighbor look-up is different
2232                  * from the fl6->daddr used to look-up route here.
2233                  */
2234                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2235
2236                 if (rt) {
2237                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2238                          * As rt6_uncached_list_add() does not consume refcnt,
2239                          * this refcnt is always returned to the caller even
2240                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2241                          */
2242                         rt6_uncached_list_add(rt);
2243                         rcu_read_unlock();
2244
2245                         return rt;
2246                 }
2247         } else {
2248                 /* Get a percpu copy */
2249                 local_bh_disable();
2250                 rt = rt6_get_pcpu_route(&res);
2251
2252                 if (!rt)
2253                         rt = rt6_make_pcpu_route(net, &res);
2254
2255                 local_bh_enable();
2256         }
2257 out:
2258         if (!rt)
2259                 rt = net->ipv6.ip6_null_entry;
2260         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2261                 ip6_hold_safe(net, &rt);
2262         rcu_read_unlock();
2263
2264         return rt;
2265 }
2266 EXPORT_SYMBOL_GPL(ip6_pol_route);
2267
2268 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2269                                             struct fib6_table *table,
2270                                             struct flowi6 *fl6,
2271                                             const struct sk_buff *skb,
2272                                             int flags)
2273 {
2274         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2275 }
2276
2277 struct dst_entry *ip6_route_input_lookup(struct net *net,
2278                                          struct net_device *dev,
2279                                          struct flowi6 *fl6,
2280                                          const struct sk_buff *skb,
2281                                          int flags)
2282 {
2283         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2284                 flags |= RT6_LOOKUP_F_IFACE;
2285
2286         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2287 }
2288 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2289
2290 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2291                                   struct flow_keys *keys,
2292                                   struct flow_keys *flkeys)
2293 {
2294         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2295         const struct ipv6hdr *key_iph = outer_iph;
2296         struct flow_keys *_flkeys = flkeys;
2297         const struct ipv6hdr *inner_iph;
2298         const struct icmp6hdr *icmph;
2299         struct ipv6hdr _inner_iph;
2300         struct icmp6hdr _icmph;
2301
2302         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2303                 goto out;
2304
2305         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2306                                    sizeof(_icmph), &_icmph);
2307         if (!icmph)
2308                 goto out;
2309
2310         if (!icmpv6_is_err(icmph->icmp6_type))
2311                 goto out;
2312
2313         inner_iph = skb_header_pointer(skb,
2314                                        skb_transport_offset(skb) + sizeof(*icmph),
2315                                        sizeof(_inner_iph), &_inner_iph);
2316         if (!inner_iph)
2317                 goto out;
2318
2319         key_iph = inner_iph;
2320         _flkeys = NULL;
2321 out:
2322         if (_flkeys) {
2323                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2324                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2325                 keys->tags.flow_label = _flkeys->tags.flow_label;
2326                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2327         } else {
2328                 keys->addrs.v6addrs.src = key_iph->saddr;
2329                 keys->addrs.v6addrs.dst = key_iph->daddr;
2330                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2331                 keys->basic.ip_proto = key_iph->nexthdr;
2332         }
2333 }
2334
2335 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2336                                            const struct sk_buff *skb,
2337                                            bool *p_has_inner)
2338 {
2339         u32 hash_fields = ip6_multipath_hash_fields(net);
2340         struct flow_keys keys, hash_keys;
2341
2342         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2343                 return 0;
2344
2345         memset(&hash_keys, 0, sizeof(hash_keys));
2346         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2347
2348         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2349         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2350                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2351         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2352                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2353         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2354                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2355         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2356                 hash_keys.tags.flow_label = keys.tags.flow_label;
2357         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2358                 hash_keys.ports.src = keys.ports.src;
2359         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2360                 hash_keys.ports.dst = keys.ports.dst;
2361
2362         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2363         return flow_hash_from_keys(&hash_keys);
2364 }
2365
2366 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2367                                            const struct sk_buff *skb,
2368                                            bool has_inner)
2369 {
2370         u32 hash_fields = ip6_multipath_hash_fields(net);
2371         struct flow_keys keys, hash_keys;
2372
2373         /* We assume the packet carries an encapsulation, but if none was
2374          * encountered during dissection of the outer flow, then there is no
2375          * point in calling the flow dissector again.
2376          */
2377         if (!has_inner)
2378                 return 0;
2379
2380         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2381                 return 0;
2382
2383         memset(&hash_keys, 0, sizeof(hash_keys));
2384         skb_flow_dissect_flow_keys(skb, &keys, 0);
2385
2386         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2387                 return 0;
2388
2389         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2390                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2391                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2392                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2393                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2394                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2395         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2396                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2397                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2398                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2399                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2400                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2401                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2402                         hash_keys.tags.flow_label = keys.tags.flow_label;
2403         }
2404
2405         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2406                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2407         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2408                 hash_keys.ports.src = keys.ports.src;
2409         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2410                 hash_keys.ports.dst = keys.ports.dst;
2411
2412         return flow_hash_from_keys(&hash_keys);
2413 }
2414
2415 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2416                                          const struct sk_buff *skb)
2417 {
2418         u32 mhash, mhash_inner;
2419         bool has_inner = true;
2420
2421         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2422         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2423
2424         return jhash_2words(mhash, mhash_inner, 0);
2425 }
2426
2427 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2428                                          const struct flowi6 *fl6)
2429 {
2430         u32 hash_fields = ip6_multipath_hash_fields(net);
2431         struct flow_keys hash_keys;
2432
2433         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2434                 return 0;
2435
2436         memset(&hash_keys, 0, sizeof(hash_keys));
2437         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2438         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2439                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2440         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2441                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2442         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2443                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2444         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2445                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2446         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2447                 hash_keys.ports.src = fl6->fl6_sport;
2448         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2449                 hash_keys.ports.dst = fl6->fl6_dport;
2450
2451         return flow_hash_from_keys(&hash_keys);
2452 }
2453
2454 /* if skb is set it will be used and fl6 can be NULL */
2455 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2456                        const struct sk_buff *skb, struct flow_keys *flkeys)
2457 {
2458         struct flow_keys hash_keys;
2459         u32 mhash = 0;
2460
2461         switch (ip6_multipath_hash_policy(net)) {
2462         case 0:
2463                 memset(&hash_keys, 0, sizeof(hash_keys));
2464                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2465                 if (skb) {
2466                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2467                 } else {
2468                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2469                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2470                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2471                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2472                 }
2473                 mhash = flow_hash_from_keys(&hash_keys);
2474                 break;
2475         case 1:
2476                 if (skb) {
2477                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2478                         struct flow_keys keys;
2479
2480                         /* short-circuit if we already have L4 hash present */
2481                         if (skb->l4_hash)
2482                                 return skb_get_hash_raw(skb) >> 1;
2483
2484                         memset(&hash_keys, 0, sizeof(hash_keys));
2485
2486                         if (!flkeys) {
2487                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2488                                 flkeys = &keys;
2489                         }
2490                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2491                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2492                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2493                         hash_keys.ports.src = flkeys->ports.src;
2494                         hash_keys.ports.dst = flkeys->ports.dst;
2495                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2496                 } else {
2497                         memset(&hash_keys, 0, sizeof(hash_keys));
2498                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2499                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2500                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2501                         hash_keys.ports.src = fl6->fl6_sport;
2502                         hash_keys.ports.dst = fl6->fl6_dport;
2503                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2504                 }
2505                 mhash = flow_hash_from_keys(&hash_keys);
2506                 break;
2507         case 2:
2508                 memset(&hash_keys, 0, sizeof(hash_keys));
2509                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2510                 if (skb) {
2511                         struct flow_keys keys;
2512
2513                         if (!flkeys) {
2514                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2515                                 flkeys = &keys;
2516                         }
2517
2518                         /* Inner can be v4 or v6 */
2519                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2520                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2521                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2522                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2523                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2524                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2525                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2526                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2527                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2528                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2529                         } else {
2530                                 /* Same as case 0 */
2531                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2532                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2533                         }
2534                 } else {
2535                         /* Same as case 0 */
2536                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2537                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2538                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2539                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2540                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2541                 }
2542                 mhash = flow_hash_from_keys(&hash_keys);
2543                 break;
2544         case 3:
2545                 if (skb)
2546                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2547                 else
2548                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2549                 break;
2550         }
2551
2552         return mhash >> 1;
2553 }
2554
2555 /* Called with rcu held */
2556 void ip6_route_input(struct sk_buff *skb)
2557 {
2558         const struct ipv6hdr *iph = ipv6_hdr(skb);
2559         struct net *net = dev_net(skb->dev);
2560         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2561         struct ip_tunnel_info *tun_info;
2562         struct flowi6 fl6 = {
2563                 .flowi6_iif = skb->dev->ifindex,
2564                 .daddr = iph->daddr,
2565                 .saddr = iph->saddr,
2566                 .flowlabel = ip6_flowinfo(iph),
2567                 .flowi6_mark = skb->mark,
2568                 .flowi6_proto = iph->nexthdr,
2569         };
2570         struct flow_keys *flkeys = NULL, _flkeys;
2571
2572         tun_info = skb_tunnel_info(skb);
2573         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2574                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2575
2576         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2577                 flkeys = &_flkeys;
2578
2579         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2580                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2581         skb_dst_drop(skb);
2582         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2583                                                       &fl6, skb, flags));
2584 }
2585
2586 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2587                                              struct fib6_table *table,
2588                                              struct flowi6 *fl6,
2589                                              const struct sk_buff *skb,
2590                                              int flags)
2591 {
2592         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2593 }
2594
2595 static struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2596                                                       const struct sock *sk,
2597                                                       struct flowi6 *fl6,
2598                                                       int flags)
2599 {
2600         bool any_src;
2601
2602         if (ipv6_addr_type(&fl6->daddr) &
2603             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2604                 struct dst_entry *dst;
2605
2606                 /* This function does not take refcnt on the dst */
2607                 dst = l3mdev_link_scope_lookup(net, fl6);
2608                 if (dst)
2609                         return dst;
2610         }
2611
2612         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2613
2614         flags |= RT6_LOOKUP_F_DST_NOREF;
2615         any_src = ipv6_addr_any(&fl6->saddr);
2616         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2617             (fl6->flowi6_oif && any_src))
2618                 flags |= RT6_LOOKUP_F_IFACE;
2619
2620         if (!any_src)
2621                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2622         else if (sk)
2623                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2624
2625         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2626 }
2627
2628 struct dst_entry *ip6_route_output_flags(struct net *net,
2629                                          const struct sock *sk,
2630                                          struct flowi6 *fl6,
2631                                          int flags)
2632 {
2633         struct dst_entry *dst;
2634         struct rt6_info *rt6;
2635
2636         rcu_read_lock();
2637         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2638         rt6 = (struct rt6_info *)dst;
2639         /* For dst cached in uncached_list, refcnt is already taken. */
2640         if (list_empty(&rt6->dst.rt_uncached) && !dst_hold_safe(dst)) {
2641                 dst = &net->ipv6.ip6_null_entry->dst;
2642                 dst_hold(dst);
2643         }
2644         rcu_read_unlock();
2645
2646         return dst;
2647 }
2648 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2649
2650 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2651 {
2652         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2653         struct net_device *loopback_dev = net->loopback_dev;
2654         struct dst_entry *new = NULL;
2655
2656         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2657                        DST_OBSOLETE_DEAD, 0);
2658         if (rt) {
2659                 rt6_info_init(rt);
2660                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2661
2662                 new = &rt->dst;
2663                 new->__use = 1;
2664                 new->input = dst_discard;
2665                 new->output = dst_discard_out;
2666
2667                 dst_copy_metrics(new, &ort->dst);
2668
2669                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2670                 rt->rt6i_gateway = ort->rt6i_gateway;
2671                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2672
2673                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2674 #ifdef CONFIG_IPV6_SUBTREES
2675                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2676 #endif
2677         }
2678
2679         dst_release(dst_orig);
2680         return new ? new : ERR_PTR(-ENOMEM);
2681 }
2682
2683 /*
2684  *      Destination cache support functions
2685  */
2686
2687 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2688 {
2689         u32 rt_cookie = 0;
2690
2691         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2692                 return false;
2693
2694         if (fib6_check_expired(f6i))
2695                 return false;
2696
2697         return true;
2698 }
2699
2700 static struct dst_entry *rt6_check(struct rt6_info *rt,
2701                                    struct fib6_info *from,
2702                                    u32 cookie)
2703 {
2704         u32 rt_cookie = 0;
2705
2706         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2707             rt_cookie != cookie)
2708                 return NULL;
2709
2710         if (rt6_check_expired(rt))
2711                 return NULL;
2712
2713         return &rt->dst;
2714 }
2715
2716 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2717                                             struct fib6_info *from,
2718                                             u32 cookie)
2719 {
2720         if (!__rt6_check_expired(rt) &&
2721             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2722             fib6_check(from, cookie))
2723                 return &rt->dst;
2724         else
2725                 return NULL;
2726 }
2727
2728 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2729                                                         u32 cookie)
2730 {
2731         struct dst_entry *dst_ret;
2732         struct fib6_info *from;
2733         struct rt6_info *rt;
2734
2735         rt = container_of(dst, struct rt6_info, dst);
2736
2737         if (rt->sernum)
2738                 return rt6_is_valid(rt) ? dst : NULL;
2739
2740         rcu_read_lock();
2741
2742         /* All IPV6 dsts are created with ->obsolete set to the value
2743          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2744          * into this function always.
2745          */
2746
2747         from = rcu_dereference(rt->from);
2748
2749         if (from && (rt->rt6i_flags & RTF_PCPU ||
2750             unlikely(!list_empty(&rt->dst.rt_uncached))))
2751                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2752         else
2753                 dst_ret = rt6_check(rt, from, cookie);
2754
2755         rcu_read_unlock();
2756
2757         return dst_ret;
2758 }
2759 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2760
2761 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2762 {
2763         struct rt6_info *rt = (struct rt6_info *) dst;
2764
2765         if (rt) {
2766                 if (rt->rt6i_flags & RTF_CACHE) {
2767                         rcu_read_lock();
2768                         if (rt6_check_expired(rt)) {
2769                                 rt6_remove_exception_rt(rt);
2770                                 dst = NULL;
2771                         }
2772                         rcu_read_unlock();
2773                 } else {
2774                         dst_release(dst);
2775                         dst = NULL;
2776                 }
2777         }
2778         return dst;
2779 }
2780
2781 static void ip6_link_failure(struct sk_buff *skb)
2782 {
2783         struct rt6_info *rt;
2784
2785         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2786
2787         rt = (struct rt6_info *) skb_dst(skb);
2788         if (rt) {
2789                 rcu_read_lock();
2790                 if (rt->rt6i_flags & RTF_CACHE) {
2791                         rt6_remove_exception_rt(rt);
2792                 } else {
2793                         struct fib6_info *from;
2794                         struct fib6_node *fn;
2795
2796                         from = rcu_dereference(rt->from);
2797                         if (from) {
2798                                 fn = rcu_dereference(from->fib6_node);
2799                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2800                                         WRITE_ONCE(fn->fn_sernum, -1);
2801                         }
2802                 }
2803                 rcu_read_unlock();
2804         }
2805 }
2806
2807 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2808 {
2809         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2810                 struct fib6_info *from;
2811
2812                 rcu_read_lock();
2813                 from = rcu_dereference(rt0->from);
2814                 if (from)
2815                         rt0->dst.expires = from->expires;
2816                 rcu_read_unlock();
2817         }
2818
2819         dst_set_expires(&rt0->dst, timeout);
2820         rt0->rt6i_flags |= RTF_EXPIRES;
2821 }
2822
2823 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2824 {
2825         struct net *net = dev_net(rt->dst.dev);
2826
2827         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2828         rt->rt6i_flags |= RTF_MODIFIED;
2829         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2830 }
2831
2832 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2833 {
2834         return !(rt->rt6i_flags & RTF_CACHE) &&
2835                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2836 }
2837
2838 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2839                                  const struct ipv6hdr *iph, u32 mtu,
2840                                  bool confirm_neigh)
2841 {
2842         const struct in6_addr *daddr, *saddr;
2843         struct rt6_info *rt6 = (struct rt6_info *)dst;
2844
2845         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2846          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2847          * [see also comment in rt6_mtu_change_route()]
2848          */
2849
2850         if (iph) {
2851                 daddr = &iph->daddr;
2852                 saddr = &iph->saddr;
2853         } else if (sk) {
2854                 daddr = &sk->sk_v6_daddr;
2855                 saddr = &inet6_sk(sk)->saddr;
2856         } else {
2857                 daddr = NULL;
2858                 saddr = NULL;
2859         }
2860
2861         if (confirm_neigh)
2862                 dst_confirm_neigh(dst, daddr);
2863
2864         if (mtu < IPV6_MIN_MTU)
2865                 return;
2866         if (mtu >= dst_mtu(dst))
2867                 return;
2868
2869         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2870                 rt6_do_update_pmtu(rt6, mtu);
2871                 /* update rt6_ex->stamp for cache */
2872                 if (rt6->rt6i_flags & RTF_CACHE)
2873                         rt6_update_exception_stamp_rt(rt6);
2874         } else if (daddr) {
2875                 struct fib6_result res = {};
2876                 struct rt6_info *nrt6;
2877
2878                 rcu_read_lock();
2879                 res.f6i = rcu_dereference(rt6->from);
2880                 if (!res.f6i)
2881                         goto out_unlock;
2882
2883                 res.fib6_flags = res.f6i->fib6_flags;
2884                 res.fib6_type = res.f6i->fib6_type;
2885
2886                 if (res.f6i->nh) {
2887                         struct fib6_nh_match_arg arg = {
2888                                 .dev = dst->dev,
2889                                 .gw = &rt6->rt6i_gateway,
2890                         };
2891
2892                         nexthop_for_each_fib6_nh(res.f6i->nh,
2893                                                  fib6_nh_find_match, &arg);
2894
2895                         /* fib6_info uses a nexthop that does not have fib6_nh
2896                          * using the dst->dev + gw. Should be impossible.
2897                          */
2898                         if (!arg.match)
2899                                 goto out_unlock;
2900
2901                         res.nh = arg.match;
2902                 } else {
2903                         res.nh = res.f6i->fib6_nh;
2904                 }
2905
2906                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2907                 if (nrt6) {
2908                         rt6_do_update_pmtu(nrt6, mtu);
2909                         if (rt6_insert_exception(nrt6, &res))
2910                                 dst_release_immediate(&nrt6->dst);
2911                 }
2912 out_unlock:
2913                 rcu_read_unlock();
2914         }
2915 }
2916
2917 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2918                                struct sk_buff *skb, u32 mtu,
2919                                bool confirm_neigh)
2920 {
2921         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2922                              confirm_neigh);
2923 }
2924
2925 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2926                      int oif, u32 mark, kuid_t uid)
2927 {
2928         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2929         struct dst_entry *dst;
2930         struct flowi6 fl6 = {
2931                 .flowi6_oif = oif,
2932                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2933                 .daddr = iph->daddr,
2934                 .saddr = iph->saddr,
2935                 .flowlabel = ip6_flowinfo(iph),
2936                 .flowi6_uid = uid,
2937         };
2938
2939         dst = ip6_route_output(net, NULL, &fl6);
2940         if (!dst->error)
2941                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2942         dst_release(dst);
2943 }
2944 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2945
2946 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2947 {
2948         int oif = sk->sk_bound_dev_if;
2949         struct dst_entry *dst;
2950
2951         if (!oif && skb->dev)
2952                 oif = l3mdev_master_ifindex(skb->dev);
2953
2954         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, READ_ONCE(sk->sk_mark),
2955                         sk->sk_uid);
2956
2957         dst = __sk_dst_get(sk);
2958         if (!dst || !dst->obsolete ||
2959             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2960                 return;
2961
2962         bh_lock_sock(sk);
2963         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2964                 ip6_datagram_dst_update(sk, false);
2965         bh_unlock_sock(sk);
2966 }
2967 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2968
2969 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2970                            const struct flowi6 *fl6)
2971 {
2972 #ifdef CONFIG_IPV6_SUBTREES
2973         struct ipv6_pinfo *np = inet6_sk(sk);
2974 #endif
2975
2976         ip6_dst_store(sk, dst,
2977                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2978                       &sk->sk_v6_daddr : NULL,
2979 #ifdef CONFIG_IPV6_SUBTREES
2980                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2981                       &np->saddr :
2982 #endif
2983                       NULL);
2984 }
2985
2986 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2987                                   struct flowi6 *fl6,
2988                                   const struct in6_addr *gw,
2989                                   struct rt6_info **ret)
2990 {
2991         const struct fib6_nh *nh = res->nh;
2992
2993         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2994             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2995                 return false;
2996
2997         /* rt_cache's gateway might be different from its 'parent'
2998          * in the case of an ip redirect.
2999          * So we keep searching in the exception table if the gateway
3000          * is different.
3001          */
3002         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3003                 struct rt6_info *rt_cache;
3004
3005                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3006                 if (rt_cache &&
3007                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3008                         *ret = rt_cache;
3009                         return true;
3010                 }
3011                 return false;
3012         }
3013         return true;
3014 }
3015
3016 struct fib6_nh_rd_arg {
3017         struct fib6_result      *res;
3018         struct flowi6           *fl6;
3019         const struct in6_addr   *gw;
3020         struct rt6_info         **ret;
3021 };
3022
3023 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3024 {
3025         struct fib6_nh_rd_arg *arg = _arg;
3026
3027         arg->res->nh = nh;
3028         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3029 }
3030
3031 /* Handle redirects */
3032 struct ip6rd_flowi {
3033         struct flowi6 fl6;
3034         struct in6_addr gateway;
3035 };
3036
3037 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3038                                              struct fib6_table *table,
3039                                              struct flowi6 *fl6,
3040                                              const struct sk_buff *skb,
3041                                              int flags)
3042 {
3043         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3044         struct rt6_info *ret = NULL;
3045         struct fib6_result res = {};
3046         struct fib6_nh_rd_arg arg = {
3047                 .res = &res,
3048                 .fl6 = fl6,
3049                 .gw  = &rdfl->gateway,
3050                 .ret = &ret
3051         };
3052         struct fib6_info *rt;
3053         struct fib6_node *fn;
3054
3055         /* Get the "current" route for this destination and
3056          * check if the redirect has come from appropriate router.
3057          *
3058          * RFC 4861 specifies that redirects should only be
3059          * accepted if they come from the nexthop to the target.
3060          * Due to the way the routes are chosen, this notion
3061          * is a bit fuzzy and one might need to check all possible
3062          * routes.
3063          */
3064
3065         rcu_read_lock();
3066         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3067 restart:
3068         for_each_fib6_node_rt_rcu(fn) {
3069                 res.f6i = rt;
3070                 if (fib6_check_expired(rt))
3071                         continue;
3072                 if (rt->fib6_flags & RTF_REJECT)
3073                         break;
3074                 if (unlikely(rt->nh)) {
3075                         if (nexthop_is_blackhole(rt->nh))
3076                                 continue;
3077                         /* on match, res->nh is filled in and potentially ret */
3078                         if (nexthop_for_each_fib6_nh(rt->nh,
3079                                                      fib6_nh_redirect_match,
3080                                                      &arg))
3081                                 goto out;
3082                 } else {
3083                         res.nh = rt->fib6_nh;
3084                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3085                                                   &ret))
3086                                 goto out;
3087                 }
3088         }
3089
3090         if (!rt)
3091                 rt = net->ipv6.fib6_null_entry;
3092         else if (rt->fib6_flags & RTF_REJECT) {
3093                 ret = net->ipv6.ip6_null_entry;
3094                 goto out;
3095         }
3096
3097         if (rt == net->ipv6.fib6_null_entry) {
3098                 fn = fib6_backtrack(fn, &fl6->saddr);
3099                 if (fn)
3100                         goto restart;
3101         }
3102
3103         res.f6i = rt;
3104         res.nh = rt->fib6_nh;
3105 out:
3106         if (ret) {
3107                 ip6_hold_safe(net, &ret);
3108         } else {
3109                 res.fib6_flags = res.f6i->fib6_flags;
3110                 res.fib6_type = res.f6i->fib6_type;
3111                 ret = ip6_create_rt_rcu(&res);
3112         }
3113
3114         rcu_read_unlock();
3115
3116         trace_fib6_table_lookup(net, &res, table, fl6);
3117         return ret;
3118 };
3119
3120 static struct dst_entry *ip6_route_redirect(struct net *net,
3121                                             const struct flowi6 *fl6,
3122                                             const struct sk_buff *skb,
3123                                             const struct in6_addr *gateway)
3124 {
3125         int flags = RT6_LOOKUP_F_HAS_SADDR;
3126         struct ip6rd_flowi rdfl;
3127
3128         rdfl.fl6 = *fl6;
3129         rdfl.gateway = *gateway;
3130
3131         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3132                                 flags, __ip6_route_redirect);
3133 }
3134
3135 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3136                   kuid_t uid)
3137 {
3138         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3139         struct dst_entry *dst;
3140         struct flowi6 fl6 = {
3141                 .flowi6_iif = LOOPBACK_IFINDEX,
3142                 .flowi6_oif = oif,
3143                 .flowi6_mark = mark,
3144                 .daddr = iph->daddr,
3145                 .saddr = iph->saddr,
3146                 .flowlabel = ip6_flowinfo(iph),
3147                 .flowi6_uid = uid,
3148         };
3149
3150         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3151         rt6_do_redirect(dst, NULL, skb);
3152         dst_release(dst);
3153 }
3154 EXPORT_SYMBOL_GPL(ip6_redirect);
3155
3156 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3157 {
3158         const struct ipv6hdr *iph = ipv6_hdr(skb);
3159         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3160         struct dst_entry *dst;
3161         struct flowi6 fl6 = {
3162                 .flowi6_iif = LOOPBACK_IFINDEX,
3163                 .flowi6_oif = oif,
3164                 .daddr = msg->dest,
3165                 .saddr = iph->daddr,
3166                 .flowi6_uid = sock_net_uid(net, NULL),
3167         };
3168
3169         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3170         rt6_do_redirect(dst, NULL, skb);
3171         dst_release(dst);
3172 }
3173
3174 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3175 {
3176         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if,
3177                      READ_ONCE(sk->sk_mark), sk->sk_uid);
3178 }
3179 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3180
3181 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3182 {
3183         struct net_device *dev = dst->dev;
3184         unsigned int mtu = dst_mtu(dst);
3185         struct net *net = dev_net(dev);
3186
3187         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3188
3189         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3190                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3191
3192         /*
3193          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3194          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3195          * IPV6_MAXPLEN is also valid and means: "any MSS,
3196          * rely only on pmtu discovery"
3197          */
3198         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3199                 mtu = IPV6_MAXPLEN;
3200         return mtu;
3201 }
3202
3203 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3204 {
3205         return ip6_dst_mtu_maybe_forward(dst, false);
3206 }
3207 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3208
3209 /* MTU selection:
3210  * 1. mtu on route is locked - use it
3211  * 2. mtu from nexthop exception
3212  * 3. mtu from egress device
3213  *
3214  * based on ip6_dst_mtu_forward and exception logic of
3215  * rt6_find_cached_rt; called with rcu_read_lock
3216  */
3217 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3218                       const struct in6_addr *daddr,
3219                       const struct in6_addr *saddr)
3220 {
3221         const struct fib6_nh *nh = res->nh;
3222         struct fib6_info *f6i = res->f6i;
3223         struct inet6_dev *idev;
3224         struct rt6_info *rt;
3225         u32 mtu = 0;
3226
3227         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3228                 mtu = f6i->fib6_pmtu;
3229                 if (mtu)
3230                         goto out;
3231         }
3232
3233         rt = rt6_find_cached_rt(res, daddr, saddr);
3234         if (unlikely(rt)) {
3235                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3236         } else {
3237                 struct net_device *dev = nh->fib_nh_dev;
3238
3239                 mtu = IPV6_MIN_MTU;
3240                 idev = __in6_dev_get(dev);
3241                 if (idev && idev->cnf.mtu6 > mtu)
3242                         mtu = idev->cnf.mtu6;
3243         }
3244
3245         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3246 out:
3247         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3248 }
3249
3250 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3251                                   struct flowi6 *fl6)
3252 {
3253         struct dst_entry *dst;
3254         struct rt6_info *rt;
3255         struct inet6_dev *idev = in6_dev_get(dev);
3256         struct net *net = dev_net(dev);
3257
3258         if (unlikely(!idev))
3259                 return ERR_PTR(-ENODEV);
3260
3261         rt = ip6_dst_alloc(net, dev, 0);
3262         if (unlikely(!rt)) {
3263                 in6_dev_put(idev);
3264                 dst = ERR_PTR(-ENOMEM);
3265                 goto out;
3266         }
3267
3268         rt->dst.input = ip6_input;
3269         rt->dst.output  = ip6_output;
3270         rt->rt6i_gateway  = fl6->daddr;
3271         rt->rt6i_dst.addr = fl6->daddr;
3272         rt->rt6i_dst.plen = 128;
3273         rt->rt6i_idev     = idev;
3274         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3275
3276         /* Add this dst into uncached_list so that rt6_disable_ip() can
3277          * do proper release of the net_device
3278          */
3279         rt6_uncached_list_add(rt);
3280
3281         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3282
3283 out:
3284         return dst;
3285 }
3286
3287 static void ip6_dst_gc(struct dst_ops *ops)
3288 {
3289         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3290         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3291         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3292         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3293         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3294         unsigned int val;
3295         int entries;
3296
3297         if (time_after(rt_last_gc + rt_min_interval, jiffies))
3298                 goto out;
3299
3300         fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3301         entries = dst_entries_get_slow(ops);
3302         if (entries < ops->gc_thresh)
3303                 atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3304 out:
3305         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3306         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3307 }
3308
3309 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3310                                const struct in6_addr *gw_addr, u32 tbid,
3311                                int flags, struct fib6_result *res)
3312 {
3313         struct flowi6 fl6 = {
3314                 .flowi6_oif = cfg->fc_ifindex,
3315                 .daddr = *gw_addr,
3316                 .saddr = cfg->fc_prefsrc,
3317         };
3318         struct fib6_table *table;
3319         int err;
3320
3321         table = fib6_get_table(net, tbid);
3322         if (!table)
3323                 return -EINVAL;
3324
3325         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3326                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3327
3328         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3329
3330         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3331         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3332                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3333                                  cfg->fc_ifindex != 0, NULL, flags);
3334
3335         return err;
3336 }
3337
3338 static int ip6_route_check_nh_onlink(struct net *net,
3339                                      struct fib6_config *cfg,
3340                                      const struct net_device *dev,
3341                                      struct netlink_ext_ack *extack)
3342 {
3343         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3344         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3345         struct fib6_result res = {};
3346         int err;
3347
3348         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3349         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3350             /* ignore match if it is the default route */
3351             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3352             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3353                 NL_SET_ERR_MSG(extack,
3354                                "Nexthop has invalid gateway or device mismatch");
3355                 err = -EINVAL;
3356         }
3357
3358         return err;
3359 }
3360
3361 static int ip6_route_check_nh(struct net *net,
3362                               struct fib6_config *cfg,
3363                               struct net_device **_dev,
3364                               netdevice_tracker *dev_tracker,
3365                               struct inet6_dev **idev)
3366 {
3367         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3368         struct net_device *dev = _dev ? *_dev : NULL;
3369         int flags = RT6_LOOKUP_F_IFACE;
3370         struct fib6_result res = {};
3371         int err = -EHOSTUNREACH;
3372
3373         if (cfg->fc_table) {
3374                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3375                                           cfg->fc_table, flags, &res);
3376                 /* gw_addr can not require a gateway or resolve to a reject
3377                  * route. If a device is given, it must match the result.
3378                  */
3379                 if (err || res.fib6_flags & RTF_REJECT ||
3380                     res.nh->fib_nh_gw_family ||
3381                     (dev && dev != res.nh->fib_nh_dev))
3382                         err = -EHOSTUNREACH;
3383         }
3384
3385         if (err < 0) {
3386                 struct flowi6 fl6 = {
3387                         .flowi6_oif = cfg->fc_ifindex,
3388                         .daddr = *gw_addr,
3389                 };
3390
3391                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3392                 if (err || res.fib6_flags & RTF_REJECT ||
3393                     res.nh->fib_nh_gw_family)
3394                         err = -EHOSTUNREACH;
3395
3396                 if (err)
3397                         return err;
3398
3399                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3400                                  cfg->fc_ifindex != 0, NULL, flags);
3401         }
3402
3403         err = 0;
3404         if (dev) {
3405                 if (dev != res.nh->fib_nh_dev)
3406                         err = -EHOSTUNREACH;
3407         } else {
3408                 *_dev = dev = res.nh->fib_nh_dev;
3409                 netdev_hold(dev, dev_tracker, GFP_ATOMIC);
3410                 *idev = in6_dev_get(dev);
3411         }
3412
3413         return err;
3414 }
3415
3416 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3417                            struct net_device **_dev,
3418                            netdevice_tracker *dev_tracker,
3419                            struct inet6_dev **idev,
3420                            struct netlink_ext_ack *extack)
3421 {
3422         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3423         int gwa_type = ipv6_addr_type(gw_addr);
3424         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3425         const struct net_device *dev = *_dev;
3426         bool need_addr_check = !dev;
3427         int err = -EINVAL;
3428
3429         /* if gw_addr is local we will fail to detect this in case
3430          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3431          * will return already-added prefix route via interface that
3432          * prefix route was assigned to, which might be non-loopback.
3433          */
3434         if (dev &&
3435             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3436                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3437                 goto out;
3438         }
3439
3440         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3441                 /* IPv6 strictly inhibits using not link-local
3442                  * addresses as nexthop address.
3443                  * Otherwise, router will not able to send redirects.
3444                  * It is very good, but in some (rare!) circumstances
3445                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3446                  * some exceptions. --ANK
3447                  * We allow IPv4-mapped nexthops to support RFC4798-type
3448                  * addressing
3449                  */
3450                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3451                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3452                         goto out;
3453                 }
3454
3455                 rcu_read_lock();
3456
3457                 if (cfg->fc_flags & RTNH_F_ONLINK)
3458                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3459                 else
3460                         err = ip6_route_check_nh(net, cfg, _dev, dev_tracker,
3461                                                  idev);
3462
3463                 rcu_read_unlock();
3464
3465                 if (err)
3466                         goto out;
3467         }
3468
3469         /* reload in case device was changed */
3470         dev = *_dev;
3471
3472         err = -EINVAL;
3473         if (!dev) {
3474                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3475                 goto out;
3476         } else if (dev->flags & IFF_LOOPBACK) {
3477                 NL_SET_ERR_MSG(extack,
3478                                "Egress device can not be loopback device for this route");
3479                 goto out;
3480         }
3481
3482         /* if we did not check gw_addr above, do so now that the
3483          * egress device has been resolved.
3484          */
3485         if (need_addr_check &&
3486             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3487                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3488                 goto out;
3489         }
3490
3491         err = 0;
3492 out:
3493         return err;
3494 }
3495
3496 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3497 {
3498         if ((flags & RTF_REJECT) ||
3499             (dev && (dev->flags & IFF_LOOPBACK) &&
3500              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3501              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3502                 return true;
3503
3504         return false;
3505 }
3506
3507 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3508                  struct fib6_config *cfg, gfp_t gfp_flags,
3509                  struct netlink_ext_ack *extack)
3510 {
3511         netdevice_tracker *dev_tracker = &fib6_nh->fib_nh_dev_tracker;
3512         struct net_device *dev = NULL;
3513         struct inet6_dev *idev = NULL;
3514         int addr_type;
3515         int err;
3516
3517         fib6_nh->fib_nh_family = AF_INET6;
3518 #ifdef CONFIG_IPV6_ROUTER_PREF
3519         fib6_nh->last_probe = jiffies;
3520 #endif
3521         if (cfg->fc_is_fdb) {
3522                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3523                 fib6_nh->fib_nh_gw_family = AF_INET6;
3524                 return 0;
3525         }
3526
3527         err = -ENODEV;
3528         if (cfg->fc_ifindex) {
3529                 dev = netdev_get_by_index(net, cfg->fc_ifindex,
3530                                           dev_tracker, gfp_flags);
3531                 if (!dev)
3532                         goto out;
3533                 idev = in6_dev_get(dev);
3534                 if (!idev)
3535                         goto out;
3536         }
3537
3538         if (cfg->fc_flags & RTNH_F_ONLINK) {
3539                 if (!dev) {
3540                         NL_SET_ERR_MSG(extack,
3541                                        "Nexthop device required for onlink");
3542                         goto out;
3543                 }
3544
3545                 if (!(dev->flags & IFF_UP)) {
3546                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3547                         err = -ENETDOWN;
3548                         goto out;
3549                 }
3550
3551                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3552         }
3553
3554         fib6_nh->fib_nh_weight = 1;
3555
3556         /* We cannot add true routes via loopback here,
3557          * they would result in kernel looping; promote them to reject routes
3558          */
3559         addr_type = ipv6_addr_type(&cfg->fc_dst);
3560         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3561                 /* hold loopback dev/idev if we haven't done so. */
3562                 if (dev != net->loopback_dev) {
3563                         if (dev) {
3564                                 netdev_put(dev, dev_tracker);
3565                                 in6_dev_put(idev);
3566                         }
3567                         dev = net->loopback_dev;
3568                         netdev_hold(dev, dev_tracker, gfp_flags);
3569                         idev = in6_dev_get(dev);
3570                         if (!idev) {
3571                                 err = -ENODEV;
3572                                 goto out;
3573                         }
3574                 }
3575                 goto pcpu_alloc;
3576         }
3577
3578         if (cfg->fc_flags & RTF_GATEWAY) {
3579                 err = ip6_validate_gw(net, cfg, &dev, dev_tracker,
3580                                       &idev, extack);
3581                 if (err)
3582                         goto out;
3583
3584                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3585                 fib6_nh->fib_nh_gw_family = AF_INET6;
3586         }
3587
3588         err = -ENODEV;
3589         if (!dev)
3590                 goto out;
3591
3592         if (idev->cnf.disable_ipv6) {
3593                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3594                 err = -EACCES;
3595                 goto out;
3596         }
3597
3598         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3599                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3600                 err = -ENETDOWN;
3601                 goto out;
3602         }
3603
3604         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3605             !netif_carrier_ok(dev))
3606                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3607
3608         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3609                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3610         if (err)
3611                 goto out;
3612
3613 pcpu_alloc:
3614         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3615         if (!fib6_nh->rt6i_pcpu) {
3616                 err = -ENOMEM;
3617                 goto out;
3618         }
3619
3620         fib6_nh->fib_nh_dev = dev;
3621         fib6_nh->fib_nh_oif = dev->ifindex;
3622         err = 0;
3623 out:
3624         if (idev)
3625                 in6_dev_put(idev);
3626
3627         if (err) {
3628                 lwtstate_put(fib6_nh->fib_nh_lws);
3629                 fib6_nh->fib_nh_lws = NULL;
3630                 netdev_put(dev, dev_tracker);
3631         }
3632
3633         return err;
3634 }
3635
3636 void fib6_nh_release(struct fib6_nh *fib6_nh)
3637 {
3638         struct rt6_exception_bucket *bucket;
3639
3640         rcu_read_lock();
3641
3642         fib6_nh_flush_exceptions(fib6_nh, NULL);
3643         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3644         if (bucket) {
3645                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3646                 kfree(bucket);
3647         }
3648
3649         rcu_read_unlock();
3650
3651         fib6_nh_release_dsts(fib6_nh);
3652         free_percpu(fib6_nh->rt6i_pcpu);
3653
3654         fib_nh_common_release(&fib6_nh->nh_common);
3655 }
3656
3657 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3658 {
3659         int cpu;
3660
3661         if (!fib6_nh->rt6i_pcpu)
3662                 return;
3663
3664         for_each_possible_cpu(cpu) {
3665                 struct rt6_info *pcpu_rt, **ppcpu_rt;
3666
3667                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3668                 pcpu_rt = xchg(ppcpu_rt, NULL);
3669                 if (pcpu_rt) {
3670                         dst_dev_put(&pcpu_rt->dst);
3671                         dst_release(&pcpu_rt->dst);
3672                 }
3673         }
3674 }
3675
3676 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3677                                               gfp_t gfp_flags,
3678                                               struct netlink_ext_ack *extack)
3679 {
3680         struct net *net = cfg->fc_nlinfo.nl_net;
3681         struct fib6_info *rt = NULL;
3682         struct nexthop *nh = NULL;
3683         struct fib6_table *table;
3684         struct fib6_nh *fib6_nh;
3685         int err = -EINVAL;
3686         int addr_type;
3687
3688         /* RTF_PCPU is an internal flag; can not be set by userspace */
3689         if (cfg->fc_flags & RTF_PCPU) {
3690                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3691                 goto out;
3692         }
3693
3694         /* RTF_CACHE is an internal flag; can not be set by userspace */
3695         if (cfg->fc_flags & RTF_CACHE) {
3696                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3697                 goto out;
3698         }
3699
3700         if (cfg->fc_type > RTN_MAX) {
3701                 NL_SET_ERR_MSG(extack, "Invalid route type");
3702                 goto out;
3703         }
3704
3705         if (cfg->fc_dst_len > 128) {
3706                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3707                 goto out;
3708         }
3709         if (cfg->fc_src_len > 128) {
3710                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3711                 goto out;
3712         }
3713 #ifndef CONFIG_IPV6_SUBTREES
3714         if (cfg->fc_src_len) {
3715                 NL_SET_ERR_MSG(extack,
3716                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3717                 goto out;
3718         }
3719 #endif
3720         if (cfg->fc_nh_id) {
3721                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3722                 if (!nh) {
3723                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3724                         goto out;
3725                 }
3726                 err = fib6_check_nexthop(nh, cfg, extack);
3727                 if (err)
3728                         goto out;
3729         }
3730
3731         err = -ENOBUFS;
3732         if (cfg->fc_nlinfo.nlh &&
3733             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3734                 table = fib6_get_table(net, cfg->fc_table);
3735                 if (!table) {
3736                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3737                         table = fib6_new_table(net, cfg->fc_table);
3738                 }
3739         } else {
3740                 table = fib6_new_table(net, cfg->fc_table);
3741         }
3742
3743         if (!table)
3744                 goto out;
3745
3746         err = -ENOMEM;
3747         rt = fib6_info_alloc(gfp_flags, !nh);
3748         if (!rt)
3749                 goto out;
3750
3751         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3752                                                extack);
3753         if (IS_ERR(rt->fib6_metrics)) {
3754                 err = PTR_ERR(rt->fib6_metrics);
3755                 /* Do not leave garbage there. */
3756                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3757                 goto out_free;
3758         }
3759
3760         if (cfg->fc_flags & RTF_ADDRCONF)
3761                 rt->dst_nocount = true;
3762
3763         if (cfg->fc_flags & RTF_EXPIRES)
3764                 fib6_set_expires_locked(rt, jiffies +
3765                                         clock_t_to_jiffies(cfg->fc_expires));
3766         else
3767                 fib6_clean_expires_locked(rt);
3768
3769         if (cfg->fc_protocol == RTPROT_UNSPEC)
3770                 cfg->fc_protocol = RTPROT_BOOT;
3771         rt->fib6_protocol = cfg->fc_protocol;
3772
3773         rt->fib6_table = table;
3774         rt->fib6_metric = cfg->fc_metric;
3775         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3776         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3777
3778         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3779         rt->fib6_dst.plen = cfg->fc_dst_len;
3780
3781 #ifdef CONFIG_IPV6_SUBTREES
3782         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3783         rt->fib6_src.plen = cfg->fc_src_len;
3784 #endif
3785         if (nh) {
3786                 if (rt->fib6_src.plen) {
3787                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3788                         goto out_free;
3789                 }
3790                 if (!nexthop_get(nh)) {
3791                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3792                         goto out_free;
3793                 }
3794                 rt->nh = nh;
3795                 fib6_nh = nexthop_fib6_nh(rt->nh);
3796         } else {
3797                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3798                 if (err)
3799                         goto out;
3800
3801                 fib6_nh = rt->fib6_nh;
3802
3803                 /* We cannot add true routes via loopback here, they would
3804                  * result in kernel looping; promote them to reject routes
3805                  */
3806                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3807                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3808                                    addr_type))
3809                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3810         }
3811
3812         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3813                 struct net_device *dev = fib6_nh->fib_nh_dev;
3814
3815                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3816                         NL_SET_ERR_MSG(extack, "Invalid source address");
3817                         err = -EINVAL;
3818                         goto out;
3819                 }
3820                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3821                 rt->fib6_prefsrc.plen = 128;
3822         } else
3823                 rt->fib6_prefsrc.plen = 0;
3824
3825         return rt;
3826 out:
3827         fib6_info_release(rt);
3828         return ERR_PTR(err);
3829 out_free:
3830         ip_fib_metrics_put(rt->fib6_metrics);
3831         kfree(rt);
3832         return ERR_PTR(err);
3833 }
3834
3835 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3836                   struct netlink_ext_ack *extack)
3837 {
3838         struct fib6_info *rt;
3839         int err;
3840
3841         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3842         if (IS_ERR(rt))
3843                 return PTR_ERR(rt);
3844
3845         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3846         fib6_info_release(rt);
3847
3848         return err;
3849 }
3850
3851 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3852 {
3853         struct net *net = info->nl_net;
3854         struct fib6_table *table;
3855         int err;
3856
3857         if (rt == net->ipv6.fib6_null_entry) {
3858                 err = -ENOENT;
3859                 goto out;
3860         }
3861
3862         table = rt->fib6_table;
3863         spin_lock_bh(&table->tb6_lock);
3864         err = fib6_del(rt, info);
3865         spin_unlock_bh(&table->tb6_lock);
3866
3867 out:
3868         fib6_info_release(rt);
3869         return err;
3870 }
3871
3872 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3873 {
3874         struct nl_info info = {
3875                 .nl_net = net,
3876                 .skip_notify = skip_notify
3877         };
3878
3879         return __ip6_del_rt(rt, &info);
3880 }
3881
3882 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3883 {
3884         struct nl_info *info = &cfg->fc_nlinfo;
3885         struct net *net = info->nl_net;
3886         struct sk_buff *skb = NULL;
3887         struct fib6_table *table;
3888         int err = -ENOENT;
3889
3890         if (rt == net->ipv6.fib6_null_entry)
3891                 goto out_put;
3892         table = rt->fib6_table;
3893         spin_lock_bh(&table->tb6_lock);
3894
3895         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3896                 struct fib6_info *sibling, *next_sibling;
3897                 struct fib6_node *fn;
3898
3899                 /* prefer to send a single notification with all hops */
3900                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3901                 if (skb) {
3902                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3903
3904                         if (rt6_fill_node(net, skb, rt, NULL,
3905                                           NULL, NULL, 0, RTM_DELROUTE,
3906                                           info->portid, seq, 0) < 0) {
3907                                 kfree_skb(skb);
3908                                 skb = NULL;
3909                         } else
3910                                 info->skip_notify = 1;
3911                 }
3912
3913                 /* 'rt' points to the first sibling route. If it is not the
3914                  * leaf, then we do not need to send a notification. Otherwise,
3915                  * we need to check if the last sibling has a next route or not
3916                  * and emit a replace or delete notification, respectively.
3917                  */
3918                 info->skip_notify_kernel = 1;
3919                 fn = rcu_dereference_protected(rt->fib6_node,
3920                                             lockdep_is_held(&table->tb6_lock));
3921                 if (rcu_access_pointer(fn->leaf) == rt) {
3922                         struct fib6_info *last_sibling, *replace_rt;
3923
3924                         last_sibling = list_last_entry(&rt->fib6_siblings,
3925                                                        struct fib6_info,
3926                                                        fib6_siblings);
3927                         replace_rt = rcu_dereference_protected(
3928                                             last_sibling->fib6_next,
3929                                             lockdep_is_held(&table->tb6_lock));
3930                         if (replace_rt)
3931                                 call_fib6_entry_notifiers_replace(net,
3932                                                                   replace_rt);
3933                         else
3934                                 call_fib6_multipath_entry_notifiers(net,
3935                                                        FIB_EVENT_ENTRY_DEL,
3936                                                        rt, rt->fib6_nsiblings,
3937                                                        NULL);
3938                 }
3939                 list_for_each_entry_safe(sibling, next_sibling,
3940                                          &rt->fib6_siblings,
3941                                          fib6_siblings) {
3942                         err = fib6_del(sibling, info);
3943                         if (err)
3944                                 goto out_unlock;
3945                 }
3946         }
3947
3948         err = fib6_del(rt, info);
3949 out_unlock:
3950         spin_unlock_bh(&table->tb6_lock);
3951 out_put:
3952         fib6_info_release(rt);
3953
3954         if (skb) {
3955                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3956                             info->nlh, gfp_any());
3957         }
3958         return err;
3959 }
3960
3961 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3962 {
3963         int rc = -ESRCH;
3964
3965         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3966                 goto out;
3967
3968         if (cfg->fc_flags & RTF_GATEWAY &&
3969             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3970                 goto out;
3971
3972         rc = rt6_remove_exception_rt(rt);
3973 out:
3974         return rc;
3975 }
3976
3977 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3978                              struct fib6_nh *nh)
3979 {
3980         struct fib6_result res = {
3981                 .f6i = rt,
3982                 .nh = nh,
3983         };
3984         struct rt6_info *rt_cache;
3985
3986         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3987         if (rt_cache)
3988                 return __ip6_del_cached_rt(rt_cache, cfg);
3989
3990         return 0;
3991 }
3992
3993 struct fib6_nh_del_cached_rt_arg {
3994         struct fib6_config *cfg;
3995         struct fib6_info *f6i;
3996 };
3997
3998 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3999 {
4000         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4001         int rc;
4002
4003         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4004         return rc != -ESRCH ? rc : 0;
4005 }
4006
4007 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4008 {
4009         struct fib6_nh_del_cached_rt_arg arg = {
4010                 .cfg = cfg,
4011                 .f6i = f6i
4012         };
4013
4014         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4015 }
4016
4017 static int ip6_route_del(struct fib6_config *cfg,
4018                          struct netlink_ext_ack *extack)
4019 {
4020         struct fib6_table *table;
4021         struct fib6_info *rt;
4022         struct fib6_node *fn;
4023         int err = -ESRCH;
4024
4025         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4026         if (!table) {
4027                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4028                 return err;
4029         }
4030
4031         rcu_read_lock();
4032
4033         fn = fib6_locate(&table->tb6_root,
4034                          &cfg->fc_dst, cfg->fc_dst_len,
4035                          &cfg->fc_src, cfg->fc_src_len,
4036                          !(cfg->fc_flags & RTF_CACHE));
4037
4038         if (fn) {
4039                 for_each_fib6_node_rt_rcu(fn) {
4040                         struct fib6_nh *nh;
4041
4042                         if (rt->nh && cfg->fc_nh_id &&
4043                             rt->nh->id != cfg->fc_nh_id)
4044                                 continue;
4045
4046                         if (cfg->fc_flags & RTF_CACHE) {
4047                                 int rc = 0;
4048
4049                                 if (rt->nh) {
4050                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4051                                 } else if (cfg->fc_nh_id) {
4052                                         continue;
4053                                 } else {
4054                                         nh = rt->fib6_nh;
4055                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4056                                 }
4057                                 if (rc != -ESRCH) {
4058                                         rcu_read_unlock();
4059                                         return rc;
4060                                 }
4061                                 continue;
4062                         }
4063
4064                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4065                                 continue;
4066                         if (cfg->fc_protocol &&
4067                             cfg->fc_protocol != rt->fib6_protocol)
4068                                 continue;
4069
4070                         if (rt->nh) {
4071                                 if (!fib6_info_hold_safe(rt))
4072                                         continue;
4073                                 rcu_read_unlock();
4074
4075                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4076                         }
4077                         if (cfg->fc_nh_id)
4078                                 continue;
4079
4080                         nh = rt->fib6_nh;
4081                         if (cfg->fc_ifindex &&
4082                             (!nh->fib_nh_dev ||
4083                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4084                                 continue;
4085                         if (cfg->fc_flags & RTF_GATEWAY &&
4086                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4087                                 continue;
4088                         if (!fib6_info_hold_safe(rt))
4089                                 continue;
4090                         rcu_read_unlock();
4091
4092                         /* if gateway was specified only delete the one hop */
4093                         if (cfg->fc_flags & RTF_GATEWAY)
4094                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4095
4096                         return __ip6_del_rt_siblings(rt, cfg);
4097                 }
4098         }
4099         rcu_read_unlock();
4100
4101         return err;
4102 }
4103
4104 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4105 {
4106         struct netevent_redirect netevent;
4107         struct rt6_info *rt, *nrt = NULL;
4108         struct fib6_result res = {};
4109         struct ndisc_options ndopts;
4110         struct inet6_dev *in6_dev;
4111         struct neighbour *neigh;
4112         struct rd_msg *msg;
4113         int optlen, on_link;
4114         u8 *lladdr;
4115
4116         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4117         optlen -= sizeof(*msg);
4118
4119         if (optlen < 0) {
4120                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4121                 return;
4122         }
4123
4124         msg = (struct rd_msg *)icmp6_hdr(skb);
4125
4126         if (ipv6_addr_is_multicast(&msg->dest)) {
4127                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4128                 return;
4129         }
4130
4131         on_link = 0;
4132         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4133                 on_link = 1;
4134         } else if (ipv6_addr_type(&msg->target) !=
4135                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4136                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4137                 return;
4138         }
4139
4140         in6_dev = __in6_dev_get(skb->dev);
4141         if (!in6_dev)
4142                 return;
4143         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4144                 return;
4145
4146         /* RFC2461 8.1:
4147          *      The IP source address of the Redirect MUST be the same as the current
4148          *      first-hop router for the specified ICMP Destination Address.
4149          */
4150
4151         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4152                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4153                 return;
4154         }
4155
4156         lladdr = NULL;
4157         if (ndopts.nd_opts_tgt_lladdr) {
4158                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4159                                              skb->dev);
4160                 if (!lladdr) {
4161                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4162                         return;
4163                 }
4164         }
4165
4166         rt = (struct rt6_info *) dst;
4167         if (rt->rt6i_flags & RTF_REJECT) {
4168                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4169                 return;
4170         }
4171
4172         /* Redirect received -> path was valid.
4173          * Look, redirects are sent only in response to data packets,
4174          * so that this nexthop apparently is reachable. --ANK
4175          */
4176         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4177
4178         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4179         if (!neigh)
4180                 return;
4181
4182         /*
4183          *      We have finally decided to accept it.
4184          */
4185
4186         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4187                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4188                      NEIGH_UPDATE_F_OVERRIDE|
4189                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4190                                      NEIGH_UPDATE_F_ISROUTER)),
4191                      NDISC_REDIRECT, &ndopts);
4192
4193         rcu_read_lock();
4194         res.f6i = rcu_dereference(rt->from);
4195         if (!res.f6i)
4196                 goto out;
4197
4198         if (res.f6i->nh) {
4199                 struct fib6_nh_match_arg arg = {
4200                         .dev = dst->dev,
4201                         .gw = &rt->rt6i_gateway,
4202                 };
4203
4204                 nexthop_for_each_fib6_nh(res.f6i->nh,
4205                                          fib6_nh_find_match, &arg);
4206
4207                 /* fib6_info uses a nexthop that does not have fib6_nh
4208                  * using the dst->dev. Should be impossible
4209                  */
4210                 if (!arg.match)
4211                         goto out;
4212                 res.nh = arg.match;
4213         } else {
4214                 res.nh = res.f6i->fib6_nh;
4215         }
4216
4217         res.fib6_flags = res.f6i->fib6_flags;
4218         res.fib6_type = res.f6i->fib6_type;
4219         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4220         if (!nrt)
4221                 goto out;
4222
4223         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4224         if (on_link)
4225                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4226
4227         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4228
4229         /* rt6_insert_exception() will take care of duplicated exceptions */
4230         if (rt6_insert_exception(nrt, &res)) {
4231                 dst_release_immediate(&nrt->dst);
4232                 goto out;
4233         }
4234
4235         netevent.old = &rt->dst;
4236         netevent.new = &nrt->dst;
4237         netevent.daddr = &msg->dest;
4238         netevent.neigh = neigh;
4239         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4240
4241 out:
4242         rcu_read_unlock();
4243         neigh_release(neigh);
4244 }
4245
4246 #ifdef CONFIG_IPV6_ROUTE_INFO
4247 static struct fib6_info *rt6_get_route_info(struct net *net,
4248                                            const struct in6_addr *prefix, int prefixlen,
4249                                            const struct in6_addr *gwaddr,
4250                                            struct net_device *dev)
4251 {
4252         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4253         int ifindex = dev->ifindex;
4254         struct fib6_node *fn;
4255         struct fib6_info *rt = NULL;
4256         struct fib6_table *table;
4257
4258         table = fib6_get_table(net, tb_id);
4259         if (!table)
4260                 return NULL;
4261
4262         rcu_read_lock();
4263         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4264         if (!fn)
4265                 goto out;
4266
4267         for_each_fib6_node_rt_rcu(fn) {
4268                 /* these routes do not use nexthops */
4269                 if (rt->nh)
4270                         continue;
4271                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4272                         continue;
4273                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4274                     !rt->fib6_nh->fib_nh_gw_family)
4275                         continue;
4276                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4277                         continue;
4278                 if (!fib6_info_hold_safe(rt))
4279                         continue;
4280                 break;
4281         }
4282 out:
4283         rcu_read_unlock();
4284         return rt;
4285 }
4286
4287 static struct fib6_info *rt6_add_route_info(struct net *net,
4288                                            const struct in6_addr *prefix, int prefixlen,
4289                                            const struct in6_addr *gwaddr,
4290                                            struct net_device *dev,
4291                                            unsigned int pref)
4292 {
4293         struct fib6_config cfg = {
4294                 .fc_metric      = IP6_RT_PRIO_USER,
4295                 .fc_ifindex     = dev->ifindex,
4296                 .fc_dst_len     = prefixlen,
4297                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4298                                   RTF_UP | RTF_PREF(pref),
4299                 .fc_protocol = RTPROT_RA,
4300                 .fc_type = RTN_UNICAST,
4301                 .fc_nlinfo.portid = 0,
4302                 .fc_nlinfo.nlh = NULL,
4303                 .fc_nlinfo.nl_net = net,
4304         };
4305
4306         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4307         cfg.fc_dst = *prefix;
4308         cfg.fc_gateway = *gwaddr;
4309
4310         /* We should treat it as a default route if prefix length is 0. */
4311         if (!prefixlen)
4312                 cfg.fc_flags |= RTF_DEFAULT;
4313
4314         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4315
4316         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4317 }
4318 #endif
4319
4320 struct fib6_info *rt6_get_dflt_router(struct net *net,
4321                                      const struct in6_addr *addr,
4322                                      struct net_device *dev)
4323 {
4324         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4325         struct fib6_info *rt;
4326         struct fib6_table *table;
4327
4328         table = fib6_get_table(net, tb_id);
4329         if (!table)
4330                 return NULL;
4331
4332         rcu_read_lock();
4333         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4334                 struct fib6_nh *nh;
4335
4336                 /* RA routes do not use nexthops */
4337                 if (rt->nh)
4338                         continue;
4339
4340                 nh = rt->fib6_nh;
4341                 if (dev == nh->fib_nh_dev &&
4342                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4343                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4344                         break;
4345         }
4346         if (rt && !fib6_info_hold_safe(rt))
4347                 rt = NULL;
4348         rcu_read_unlock();
4349         return rt;
4350 }
4351
4352 struct fib6_info *rt6_add_dflt_router(struct net *net,
4353                                      const struct in6_addr *gwaddr,
4354                                      struct net_device *dev,
4355                                      unsigned int pref,
4356                                      u32 defrtr_usr_metric)
4357 {
4358         struct fib6_config cfg = {
4359                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4360                 .fc_metric      = defrtr_usr_metric,
4361                 .fc_ifindex     = dev->ifindex,
4362                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4363                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4364                 .fc_protocol = RTPROT_RA,
4365                 .fc_type = RTN_UNICAST,
4366                 .fc_nlinfo.portid = 0,
4367                 .fc_nlinfo.nlh = NULL,
4368                 .fc_nlinfo.nl_net = net,
4369         };
4370
4371         cfg.fc_gateway = *gwaddr;
4372
4373         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4374                 struct fib6_table *table;
4375
4376                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4377                 if (table)
4378                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4379         }
4380
4381         return rt6_get_dflt_router(net, gwaddr, dev);
4382 }
4383
4384 static void __rt6_purge_dflt_routers(struct net *net,
4385                                      struct fib6_table *table)
4386 {
4387         struct fib6_info *rt;
4388
4389 restart:
4390         rcu_read_lock();
4391         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4392                 struct net_device *dev = fib6_info_nh_dev(rt);
4393                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4394
4395                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4396                     (!idev || idev->cnf.accept_ra != 2) &&
4397                     fib6_info_hold_safe(rt)) {
4398                         rcu_read_unlock();
4399                         ip6_del_rt(net, rt, false);
4400                         goto restart;
4401                 }
4402         }
4403         rcu_read_unlock();
4404
4405         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4406 }
4407
4408 void rt6_purge_dflt_routers(struct net *net)
4409 {
4410         struct fib6_table *table;
4411         struct hlist_head *head;
4412         unsigned int h;
4413
4414         rcu_read_lock();
4415
4416         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4417                 head = &net->ipv6.fib_table_hash[h];
4418                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4419                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4420                                 __rt6_purge_dflt_routers(net, table);
4421                 }
4422         }
4423
4424         rcu_read_unlock();
4425 }
4426
4427 static void rtmsg_to_fib6_config(struct net *net,
4428                                  struct in6_rtmsg *rtmsg,
4429                                  struct fib6_config *cfg)
4430 {
4431         *cfg = (struct fib6_config){
4432                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4433                          : RT6_TABLE_MAIN,
4434                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4435                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4436                 .fc_expires = rtmsg->rtmsg_info,
4437                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4438                 .fc_src_len = rtmsg->rtmsg_src_len,
4439                 .fc_flags = rtmsg->rtmsg_flags,
4440                 .fc_type = rtmsg->rtmsg_type,
4441
4442                 .fc_nlinfo.nl_net = net,
4443
4444                 .fc_dst = rtmsg->rtmsg_dst,
4445                 .fc_src = rtmsg->rtmsg_src,
4446                 .fc_gateway = rtmsg->rtmsg_gateway,
4447         };
4448 }
4449
4450 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4451 {
4452         struct fib6_config cfg;
4453         int err;
4454
4455         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4456                 return -EINVAL;
4457         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4458                 return -EPERM;
4459
4460         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4461
4462         rtnl_lock();
4463         switch (cmd) {
4464         case SIOCADDRT:
4465                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4466                 break;
4467         case SIOCDELRT:
4468                 err = ip6_route_del(&cfg, NULL);
4469                 break;
4470         }
4471         rtnl_unlock();
4472         return err;
4473 }
4474
4475 /*
4476  *      Drop the packet on the floor
4477  */
4478
4479 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4480 {
4481         struct dst_entry *dst = skb_dst(skb);
4482         struct net *net = dev_net(dst->dev);
4483         struct inet6_dev *idev;
4484         SKB_DR(reason);
4485         int type;
4486
4487         if (netif_is_l3_master(skb->dev) ||
4488             dst->dev == net->loopback_dev)
4489                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4490         else
4491                 idev = ip6_dst_idev(dst);
4492
4493         switch (ipstats_mib_noroutes) {
4494         case IPSTATS_MIB_INNOROUTES:
4495                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4496                 if (type == IPV6_ADDR_ANY) {
4497                         SKB_DR_SET(reason, IP_INADDRERRORS);
4498                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4499                         break;
4500                 }
4501                 SKB_DR_SET(reason, IP_INNOROUTES);
4502                 fallthrough;
4503         case IPSTATS_MIB_OUTNOROUTES:
4504                 SKB_DR_OR(reason, IP_OUTNOROUTES);
4505                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4506                 break;
4507         }
4508
4509         /* Start over by dropping the dst for l3mdev case */
4510         if (netif_is_l3_master(skb->dev))
4511                 skb_dst_drop(skb);
4512
4513         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4514         kfree_skb_reason(skb, reason);
4515         return 0;
4516 }
4517
4518 static int ip6_pkt_discard(struct sk_buff *skb)
4519 {
4520         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4521 }
4522
4523 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4524 {
4525         skb->dev = skb_dst(skb)->dev;
4526         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4527 }
4528
4529 static int ip6_pkt_prohibit(struct sk_buff *skb)
4530 {
4531         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4532 }
4533
4534 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4535 {
4536         skb->dev = skb_dst(skb)->dev;
4537         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4538 }
4539
4540 /*
4541  *      Allocate a dst for local (unicast / anycast) address.
4542  */
4543
4544 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4545                                      struct inet6_dev *idev,
4546                                      const struct in6_addr *addr,
4547                                      bool anycast, gfp_t gfp_flags,
4548                                      struct netlink_ext_ack *extack)
4549 {
4550         struct fib6_config cfg = {
4551                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4552                 .fc_ifindex = idev->dev->ifindex,
4553                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4554                 .fc_dst = *addr,
4555                 .fc_dst_len = 128,
4556                 .fc_protocol = RTPROT_KERNEL,
4557                 .fc_nlinfo.nl_net = net,
4558                 .fc_ignore_dev_down = true,
4559         };
4560         struct fib6_info *f6i;
4561
4562         if (anycast) {
4563                 cfg.fc_type = RTN_ANYCAST;
4564                 cfg.fc_flags |= RTF_ANYCAST;
4565         } else {
4566                 cfg.fc_type = RTN_LOCAL;
4567                 cfg.fc_flags |= RTF_LOCAL;
4568         }
4569
4570         f6i = ip6_route_info_create(&cfg, gfp_flags, extack);
4571         if (!IS_ERR(f6i)) {
4572                 f6i->dst_nocount = true;
4573
4574                 if (!anycast &&
4575                     (net->ipv6.devconf_all->disable_policy ||
4576                      idev->cnf.disable_policy))
4577                         f6i->dst_nopolicy = true;
4578         }
4579
4580         return f6i;
4581 }
4582
4583 /* remove deleted ip from prefsrc entries */
4584 struct arg_dev_net_ip {
4585         struct net *net;
4586         struct in6_addr *addr;
4587 };
4588
4589 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4590 {
4591         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4592         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4593
4594         if (!rt->nh &&
4595             rt != net->ipv6.fib6_null_entry &&
4596             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr) &&
4597             !ipv6_chk_addr(net, addr, rt->fib6_nh->fib_nh_dev, 0)) {
4598                 spin_lock_bh(&rt6_exception_lock);
4599                 /* remove prefsrc entry */
4600                 rt->fib6_prefsrc.plen = 0;
4601                 spin_unlock_bh(&rt6_exception_lock);
4602         }
4603         return 0;
4604 }
4605
4606 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4607 {
4608         struct net *net = dev_net(ifp->idev->dev);
4609         struct arg_dev_net_ip adni = {
4610                 .net = net,
4611                 .addr = &ifp->addr,
4612         };
4613         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4614 }
4615
4616 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4617
4618 /* Remove routers and update dst entries when gateway turn into host. */
4619 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4620 {
4621         struct in6_addr *gateway = (struct in6_addr *)arg;
4622         struct fib6_nh *nh;
4623
4624         /* RA routes do not use nexthops */
4625         if (rt->nh)
4626                 return 0;
4627
4628         nh = rt->fib6_nh;
4629         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4630             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4631                 return -1;
4632
4633         /* Further clean up cached routes in exception table.
4634          * This is needed because cached route may have a different
4635          * gateway than its 'parent' in the case of an ip redirect.
4636          */
4637         fib6_nh_exceptions_clean_tohost(nh, gateway);
4638
4639         return 0;
4640 }
4641
4642 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4643 {
4644         fib6_clean_all(net, fib6_clean_tohost, gateway);
4645 }
4646
4647 struct arg_netdev_event {
4648         const struct net_device *dev;
4649         union {
4650                 unsigned char nh_flags;
4651                 unsigned long event;
4652         };
4653 };
4654
4655 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4656 {
4657         struct fib6_info *iter;
4658         struct fib6_node *fn;
4659
4660         fn = rcu_dereference_protected(rt->fib6_node,
4661                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4662         iter = rcu_dereference_protected(fn->leaf,
4663                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4664         while (iter) {
4665                 if (iter->fib6_metric == rt->fib6_metric &&
4666                     rt6_qualify_for_ecmp(iter))
4667                         return iter;
4668                 iter = rcu_dereference_protected(iter->fib6_next,
4669                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4670         }
4671
4672         return NULL;
4673 }
4674
4675 /* only called for fib entries with builtin fib6_nh */
4676 static bool rt6_is_dead(const struct fib6_info *rt)
4677 {
4678         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4679             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4680              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4681                 return true;
4682
4683         return false;
4684 }
4685
4686 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4687 {
4688         struct fib6_info *iter;
4689         int total = 0;
4690
4691         if (!rt6_is_dead(rt))
4692                 total += rt->fib6_nh->fib_nh_weight;
4693
4694         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4695                 if (!rt6_is_dead(iter))
4696                         total += iter->fib6_nh->fib_nh_weight;
4697         }
4698
4699         return total;
4700 }
4701
4702 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4703 {
4704         int upper_bound = -1;
4705
4706         if (!rt6_is_dead(rt)) {
4707                 *weight += rt->fib6_nh->fib_nh_weight;
4708                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4709                                                     total) - 1;
4710         }
4711         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4712 }
4713
4714 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4715 {
4716         struct fib6_info *iter;
4717         int weight = 0;
4718
4719         rt6_upper_bound_set(rt, &weight, total);
4720
4721         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4722                 rt6_upper_bound_set(iter, &weight, total);
4723 }
4724
4725 void rt6_multipath_rebalance(struct fib6_info *rt)
4726 {
4727         struct fib6_info *first;
4728         int total;
4729
4730         /* In case the entire multipath route was marked for flushing,
4731          * then there is no need to rebalance upon the removal of every
4732          * sibling route.
4733          */
4734         if (!rt->fib6_nsiblings || rt->should_flush)
4735                 return;
4736
4737         /* During lookup routes are evaluated in order, so we need to
4738          * make sure upper bounds are assigned from the first sibling
4739          * onwards.
4740          */
4741         first = rt6_multipath_first_sibling(rt);
4742         if (WARN_ON_ONCE(!first))
4743                 return;
4744
4745         total = rt6_multipath_total_weight(first);
4746         rt6_multipath_upper_bound_set(first, total);
4747 }
4748
4749 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4750 {
4751         const struct arg_netdev_event *arg = p_arg;
4752         struct net *net = dev_net(arg->dev);
4753
4754         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4755             rt->fib6_nh->fib_nh_dev == arg->dev) {
4756                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4757                 fib6_update_sernum_upto_root(net, rt);
4758                 rt6_multipath_rebalance(rt);
4759         }
4760
4761         return 0;
4762 }
4763
4764 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4765 {
4766         struct arg_netdev_event arg = {
4767                 .dev = dev,
4768                 {
4769                         .nh_flags = nh_flags,
4770                 },
4771         };
4772
4773         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4774                 arg.nh_flags |= RTNH_F_LINKDOWN;
4775
4776         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4777 }
4778
4779 /* only called for fib entries with inline fib6_nh */
4780 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4781                                    const struct net_device *dev)
4782 {
4783         struct fib6_info *iter;
4784
4785         if (rt->fib6_nh->fib_nh_dev == dev)
4786                 return true;
4787         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4788                 if (iter->fib6_nh->fib_nh_dev == dev)
4789                         return true;
4790
4791         return false;
4792 }
4793
4794 static void rt6_multipath_flush(struct fib6_info *rt)
4795 {
4796         struct fib6_info *iter;
4797
4798         rt->should_flush = 1;
4799         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4800                 iter->should_flush = 1;
4801 }
4802
4803 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4804                                              const struct net_device *down_dev)
4805 {
4806         struct fib6_info *iter;
4807         unsigned int dead = 0;
4808
4809         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4810             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4811                 dead++;
4812         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4813                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4814                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4815                         dead++;
4816
4817         return dead;
4818 }
4819
4820 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4821                                        const struct net_device *dev,
4822                                        unsigned char nh_flags)
4823 {
4824         struct fib6_info *iter;
4825
4826         if (rt->fib6_nh->fib_nh_dev == dev)
4827                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4828         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4829                 if (iter->fib6_nh->fib_nh_dev == dev)
4830                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4831 }
4832
4833 /* called with write lock held for table with rt */
4834 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4835 {
4836         const struct arg_netdev_event *arg = p_arg;
4837         const struct net_device *dev = arg->dev;
4838         struct net *net = dev_net(dev);
4839
4840         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4841                 return 0;
4842
4843         switch (arg->event) {
4844         case NETDEV_UNREGISTER:
4845                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4846         case NETDEV_DOWN:
4847                 if (rt->should_flush)
4848                         return -1;
4849                 if (!rt->fib6_nsiblings)
4850                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4851                 if (rt6_multipath_uses_dev(rt, dev)) {
4852                         unsigned int count;
4853
4854                         count = rt6_multipath_dead_count(rt, dev);
4855                         if (rt->fib6_nsiblings + 1 == count) {
4856                                 rt6_multipath_flush(rt);
4857                                 return -1;
4858                         }
4859                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4860                                                    RTNH_F_LINKDOWN);
4861                         fib6_update_sernum(net, rt);
4862                         rt6_multipath_rebalance(rt);
4863                 }
4864                 return -2;
4865         case NETDEV_CHANGE:
4866                 if (rt->fib6_nh->fib_nh_dev != dev ||
4867                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4868                         break;
4869                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4870                 rt6_multipath_rebalance(rt);
4871                 break;
4872         }
4873
4874         return 0;
4875 }
4876
4877 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4878 {
4879         struct arg_netdev_event arg = {
4880                 .dev = dev,
4881                 {
4882                         .event = event,
4883                 },
4884         };
4885         struct net *net = dev_net(dev);
4886
4887         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4888                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4889         else
4890                 fib6_clean_all(net, fib6_ifdown, &arg);
4891 }
4892
4893 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4894 {
4895         rt6_sync_down_dev(dev, event);
4896         rt6_uncached_list_flush_dev(dev);
4897         neigh_ifdown(&nd_tbl, dev);
4898 }
4899
4900 struct rt6_mtu_change_arg {
4901         struct net_device *dev;
4902         unsigned int mtu;
4903         struct fib6_info *f6i;
4904 };
4905
4906 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4907 {
4908         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4909         struct fib6_info *f6i = arg->f6i;
4910
4911         /* For administrative MTU increase, there is no way to discover
4912          * IPv6 PMTU increase, so PMTU increase should be updated here.
4913          * Since RFC 1981 doesn't include administrative MTU increase
4914          * update PMTU increase is a MUST. (i.e. jumbo frame)
4915          */
4916         if (nh->fib_nh_dev == arg->dev) {
4917                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4918                 u32 mtu = f6i->fib6_pmtu;
4919
4920                 if (mtu >= arg->mtu ||
4921                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4922                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4923
4924                 spin_lock_bh(&rt6_exception_lock);
4925                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4926                 spin_unlock_bh(&rt6_exception_lock);
4927         }
4928
4929         return 0;
4930 }
4931
4932 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4933 {
4934         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4935         struct inet6_dev *idev;
4936
4937         /* In IPv6 pmtu discovery is not optional,
4938            so that RTAX_MTU lock cannot disable it.
4939            We still use this lock to block changes
4940            caused by addrconf/ndisc.
4941         */
4942
4943         idev = __in6_dev_get(arg->dev);
4944         if (!idev)
4945                 return 0;
4946
4947         if (fib6_metric_locked(f6i, RTAX_MTU))
4948                 return 0;
4949
4950         arg->f6i = f6i;
4951         if (f6i->nh) {
4952                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4953                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4954                                                 arg);
4955         }
4956
4957         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4958 }
4959
4960 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4961 {
4962         struct rt6_mtu_change_arg arg = {
4963                 .dev = dev,
4964                 .mtu = mtu,
4965         };
4966
4967         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4968 }
4969
4970 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4971         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4972         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4973         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4974         [RTA_OIF]               = { .type = NLA_U32 },
4975         [RTA_IIF]               = { .type = NLA_U32 },
4976         [RTA_PRIORITY]          = { .type = NLA_U32 },
4977         [RTA_METRICS]           = { .type = NLA_NESTED },
4978         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4979         [RTA_PREF]              = { .type = NLA_U8 },
4980         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4981         [RTA_ENCAP]             = { .type = NLA_NESTED },
4982         [RTA_EXPIRES]           = { .type = NLA_U32 },
4983         [RTA_UID]               = { .type = NLA_U32 },
4984         [RTA_MARK]              = { .type = NLA_U32 },
4985         [RTA_TABLE]             = { .type = NLA_U32 },
4986         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4987         [RTA_SPORT]             = { .type = NLA_U16 },
4988         [RTA_DPORT]             = { .type = NLA_U16 },
4989         [RTA_NH_ID]             = { .type = NLA_U32 },
4990 };
4991
4992 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4993                               struct fib6_config *cfg,
4994                               struct netlink_ext_ack *extack)
4995 {
4996         struct rtmsg *rtm;
4997         struct nlattr *tb[RTA_MAX+1];
4998         unsigned int pref;
4999         int err;
5000
5001         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5002                                      rtm_ipv6_policy, extack);
5003         if (err < 0)
5004                 goto errout;
5005
5006         err = -EINVAL;
5007         rtm = nlmsg_data(nlh);
5008
5009         if (rtm->rtm_tos) {
5010                 NL_SET_ERR_MSG(extack,
5011                                "Invalid dsfield (tos): option not available for IPv6");
5012                 goto errout;
5013         }
5014
5015         *cfg = (struct fib6_config){
5016                 .fc_table = rtm->rtm_table,
5017                 .fc_dst_len = rtm->rtm_dst_len,
5018                 .fc_src_len = rtm->rtm_src_len,
5019                 .fc_flags = RTF_UP,
5020                 .fc_protocol = rtm->rtm_protocol,
5021                 .fc_type = rtm->rtm_type,
5022
5023                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5024                 .fc_nlinfo.nlh = nlh,
5025                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5026         };
5027
5028         if (rtm->rtm_type == RTN_UNREACHABLE ||
5029             rtm->rtm_type == RTN_BLACKHOLE ||
5030             rtm->rtm_type == RTN_PROHIBIT ||
5031             rtm->rtm_type == RTN_THROW)
5032                 cfg->fc_flags |= RTF_REJECT;
5033
5034         if (rtm->rtm_type == RTN_LOCAL)
5035                 cfg->fc_flags |= RTF_LOCAL;
5036
5037         if (rtm->rtm_flags & RTM_F_CLONED)
5038                 cfg->fc_flags |= RTF_CACHE;
5039
5040         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5041
5042         if (tb[RTA_NH_ID]) {
5043                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5044                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5045                         NL_SET_ERR_MSG(extack,
5046                                        "Nexthop specification and nexthop id are mutually exclusive");
5047                         goto errout;
5048                 }
5049                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5050         }
5051
5052         if (tb[RTA_GATEWAY]) {
5053                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5054                 cfg->fc_flags |= RTF_GATEWAY;
5055         }
5056         if (tb[RTA_VIA]) {
5057                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5058                 goto errout;
5059         }
5060
5061         if (tb[RTA_DST]) {
5062                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5063
5064                 if (nla_len(tb[RTA_DST]) < plen)
5065                         goto errout;
5066
5067                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5068         }
5069
5070         if (tb[RTA_SRC]) {
5071                 int plen = (rtm->rtm_src_len + 7) >> 3;
5072
5073                 if (nla_len(tb[RTA_SRC]) < plen)
5074                         goto errout;
5075
5076                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5077         }
5078
5079         if (tb[RTA_PREFSRC])
5080                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5081
5082         if (tb[RTA_OIF])
5083                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5084
5085         if (tb[RTA_PRIORITY])
5086                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5087
5088         if (tb[RTA_METRICS]) {
5089                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5090                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5091         }
5092
5093         if (tb[RTA_TABLE])
5094                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5095
5096         if (tb[RTA_MULTIPATH]) {
5097                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5098                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5099
5100                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5101                                                      cfg->fc_mp_len, extack);
5102                 if (err < 0)
5103                         goto errout;
5104         }
5105
5106         if (tb[RTA_PREF]) {
5107                 pref = nla_get_u8(tb[RTA_PREF]);
5108                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5109                     pref != ICMPV6_ROUTER_PREF_HIGH)
5110                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5111                 cfg->fc_flags |= RTF_PREF(pref);
5112         }
5113
5114         if (tb[RTA_ENCAP])
5115                 cfg->fc_encap = tb[RTA_ENCAP];
5116
5117         if (tb[RTA_ENCAP_TYPE]) {
5118                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5119
5120                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5121                 if (err < 0)
5122                         goto errout;
5123         }
5124
5125         if (tb[RTA_EXPIRES]) {
5126                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5127
5128                 if (addrconf_finite_timeout(timeout)) {
5129                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5130                         cfg->fc_flags |= RTF_EXPIRES;
5131                 }
5132         }
5133
5134         err = 0;
5135 errout:
5136         return err;
5137 }
5138
5139 struct rt6_nh {
5140         struct fib6_info *fib6_info;
5141         struct fib6_config r_cfg;
5142         struct list_head next;
5143 };
5144
5145 static int ip6_route_info_append(struct net *net,
5146                                  struct list_head *rt6_nh_list,
5147                                  struct fib6_info *rt,
5148                                  struct fib6_config *r_cfg)
5149 {
5150         struct rt6_nh *nh;
5151         int err = -EEXIST;
5152
5153         list_for_each_entry(nh, rt6_nh_list, next) {
5154                 /* check if fib6_info already exists */
5155                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5156                         return err;
5157         }
5158
5159         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5160         if (!nh)
5161                 return -ENOMEM;
5162         nh->fib6_info = rt;
5163         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5164         list_add_tail(&nh->next, rt6_nh_list);
5165
5166         return 0;
5167 }
5168
5169 static void ip6_route_mpath_notify(struct fib6_info *rt,
5170                                    struct fib6_info *rt_last,
5171                                    struct nl_info *info,
5172                                    __u16 nlflags)
5173 {
5174         /* if this is an APPEND route, then rt points to the first route
5175          * inserted and rt_last points to last route inserted. Userspace
5176          * wants a consistent dump of the route which starts at the first
5177          * nexthop. Since sibling routes are always added at the end of
5178          * the list, find the first sibling of the last route appended
5179          */
5180         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5181                 rt = list_first_entry(&rt_last->fib6_siblings,
5182                                       struct fib6_info,
5183                                       fib6_siblings);
5184         }
5185
5186         if (rt)
5187                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5188 }
5189
5190 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5191 {
5192         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5193         bool should_notify = false;
5194         struct fib6_info *leaf;
5195         struct fib6_node *fn;
5196
5197         rcu_read_lock();
5198         fn = rcu_dereference(rt->fib6_node);
5199         if (!fn)
5200                 goto out;
5201
5202         leaf = rcu_dereference(fn->leaf);
5203         if (!leaf)
5204                 goto out;
5205
5206         if (rt == leaf ||
5207             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5208              rt6_qualify_for_ecmp(leaf)))
5209                 should_notify = true;
5210 out:
5211         rcu_read_unlock();
5212
5213         return should_notify;
5214 }
5215
5216 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5217                              struct netlink_ext_ack *extack)
5218 {
5219         if (nla_len(nla) < sizeof(*gw)) {
5220                 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5221                 return -EINVAL;
5222         }
5223
5224         *gw = nla_get_in6_addr(nla);
5225
5226         return 0;
5227 }
5228
5229 static int ip6_route_multipath_add(struct fib6_config *cfg,
5230                                    struct netlink_ext_ack *extack)
5231 {
5232         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5233         struct nl_info *info = &cfg->fc_nlinfo;
5234         struct fib6_config r_cfg;
5235         struct rtnexthop *rtnh;
5236         struct fib6_info *rt;
5237         struct rt6_nh *err_nh;
5238         struct rt6_nh *nh, *nh_safe;
5239         __u16 nlflags;
5240         int remaining;
5241         int attrlen;
5242         int err = 1;
5243         int nhn = 0;
5244         int replace = (cfg->fc_nlinfo.nlh &&
5245                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5246         LIST_HEAD(rt6_nh_list);
5247
5248         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5249         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5250                 nlflags |= NLM_F_APPEND;
5251
5252         remaining = cfg->fc_mp_len;
5253         rtnh = (struct rtnexthop *)cfg->fc_mp;
5254
5255         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5256          * fib6_info structs per nexthop
5257          */
5258         while (rtnh_ok(rtnh, remaining)) {
5259                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5260                 if (rtnh->rtnh_ifindex)
5261                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5262
5263                 attrlen = rtnh_attrlen(rtnh);
5264                 if (attrlen > 0) {
5265                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5266
5267                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5268                         if (nla) {
5269                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5270                                                         extack);
5271                                 if (err)
5272                                         goto cleanup;
5273
5274                                 r_cfg.fc_flags |= RTF_GATEWAY;
5275                         }
5276                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5277
5278                         /* RTA_ENCAP_TYPE length checked in
5279                          * lwtunnel_valid_encap_type_attr
5280                          */
5281                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5282                         if (nla)
5283                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5284                 }
5285
5286                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5287                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5288                 if (IS_ERR(rt)) {
5289                         err = PTR_ERR(rt);
5290                         rt = NULL;
5291                         goto cleanup;
5292                 }
5293                 if (!rt6_qualify_for_ecmp(rt)) {
5294                         err = -EINVAL;
5295                         NL_SET_ERR_MSG(extack,
5296                                        "Device only routes can not be added for IPv6 using the multipath API.");
5297                         fib6_info_release(rt);
5298                         goto cleanup;
5299                 }
5300
5301                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5302
5303                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5304                                             rt, &r_cfg);
5305                 if (err) {
5306                         fib6_info_release(rt);
5307                         goto cleanup;
5308                 }
5309
5310                 rtnh = rtnh_next(rtnh, &remaining);
5311         }
5312
5313         if (list_empty(&rt6_nh_list)) {
5314                 NL_SET_ERR_MSG(extack,
5315                                "Invalid nexthop configuration - no valid nexthops");
5316                 return -EINVAL;
5317         }
5318
5319         /* for add and replace send one notification with all nexthops.
5320          * Skip the notification in fib6_add_rt2node and send one with
5321          * the full route when done
5322          */
5323         info->skip_notify = 1;
5324
5325         /* For add and replace, send one notification with all nexthops. For
5326          * append, send one notification with all appended nexthops.
5327          */
5328         info->skip_notify_kernel = 1;
5329
5330         err_nh = NULL;
5331         list_for_each_entry(nh, &rt6_nh_list, next) {
5332                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5333                 fib6_info_release(nh->fib6_info);
5334
5335                 if (!err) {
5336                         /* save reference to last route successfully inserted */
5337                         rt_last = nh->fib6_info;
5338
5339                         /* save reference to first route for notification */
5340                         if (!rt_notif)
5341                                 rt_notif = nh->fib6_info;
5342                 }
5343
5344                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5345                 nh->fib6_info = NULL;
5346                 if (err) {
5347                         if (replace && nhn)
5348                                 NL_SET_ERR_MSG_MOD(extack,
5349                                                    "multipath route replace failed (check consistency of installed routes)");
5350                         err_nh = nh;
5351                         goto add_errout;
5352                 }
5353
5354                 /* Because each route is added like a single route we remove
5355                  * these flags after the first nexthop: if there is a collision,
5356                  * we have already failed to add the first nexthop:
5357                  * fib6_add_rt2node() has rejected it; when replacing, old
5358                  * nexthops have been replaced by first new, the rest should
5359                  * be added to it.
5360                  */
5361                 if (cfg->fc_nlinfo.nlh) {
5362                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5363                                                              NLM_F_REPLACE);
5364                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5365                 }
5366                 nhn++;
5367         }
5368
5369         /* An in-kernel notification should only be sent in case the new
5370          * multipath route is added as the first route in the node, or if
5371          * it was appended to it. We pass 'rt_notif' since it is the first
5372          * sibling and might allow us to skip some checks in the replace case.
5373          */
5374         if (ip6_route_mpath_should_notify(rt_notif)) {
5375                 enum fib_event_type fib_event;
5376
5377                 if (rt_notif->fib6_nsiblings != nhn - 1)
5378                         fib_event = FIB_EVENT_ENTRY_APPEND;
5379                 else
5380                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5381
5382                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5383                                                           fib_event, rt_notif,
5384                                                           nhn - 1, extack);
5385                 if (err) {
5386                         /* Delete all the siblings that were just added */
5387                         err_nh = NULL;
5388                         goto add_errout;
5389                 }
5390         }
5391
5392         /* success ... tell user about new route */
5393         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5394         goto cleanup;
5395
5396 add_errout:
5397         /* send notification for routes that were added so that
5398          * the delete notifications sent by ip6_route_del are
5399          * coherent
5400          */
5401         if (rt_notif)
5402                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5403
5404         /* Delete routes that were already added */
5405         list_for_each_entry(nh, &rt6_nh_list, next) {
5406                 if (err_nh == nh)
5407                         break;
5408                 ip6_route_del(&nh->r_cfg, extack);
5409         }
5410
5411 cleanup:
5412         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5413                 if (nh->fib6_info)
5414                         fib6_info_release(nh->fib6_info);
5415                 list_del(&nh->next);
5416                 kfree(nh);
5417         }
5418
5419         return err;
5420 }
5421
5422 static int ip6_route_multipath_del(struct fib6_config *cfg,
5423                                    struct netlink_ext_ack *extack)
5424 {
5425         struct fib6_config r_cfg;
5426         struct rtnexthop *rtnh;
5427         int last_err = 0;
5428         int remaining;
5429         int attrlen;
5430         int err;
5431
5432         remaining = cfg->fc_mp_len;
5433         rtnh = (struct rtnexthop *)cfg->fc_mp;
5434
5435         /* Parse a Multipath Entry */
5436         while (rtnh_ok(rtnh, remaining)) {
5437                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5438                 if (rtnh->rtnh_ifindex)
5439                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5440
5441                 attrlen = rtnh_attrlen(rtnh);
5442                 if (attrlen > 0) {
5443                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5444
5445                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5446                         if (nla) {
5447                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5448                                                         extack);
5449                                 if (err) {
5450                                         last_err = err;
5451                                         goto next_rtnh;
5452                                 }
5453
5454                                 r_cfg.fc_flags |= RTF_GATEWAY;
5455                         }
5456                 }
5457                 err = ip6_route_del(&r_cfg, extack);
5458                 if (err)
5459                         last_err = err;
5460
5461 next_rtnh:
5462                 rtnh = rtnh_next(rtnh, &remaining);
5463         }
5464
5465         return last_err;
5466 }
5467
5468 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5469                               struct netlink_ext_ack *extack)
5470 {
5471         struct fib6_config cfg;
5472         int err;
5473
5474         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5475         if (err < 0)
5476                 return err;
5477
5478         if (cfg.fc_nh_id &&
5479             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5480                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5481                 return -EINVAL;
5482         }
5483
5484         if (cfg.fc_mp)
5485                 return ip6_route_multipath_del(&cfg, extack);
5486         else {
5487                 cfg.fc_delete_all_nh = 1;
5488                 return ip6_route_del(&cfg, extack);
5489         }
5490 }
5491
5492 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5493                               struct netlink_ext_ack *extack)
5494 {
5495         struct fib6_config cfg;
5496         int err;
5497
5498         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5499         if (err < 0)
5500                 return err;
5501
5502         if (cfg.fc_metric == 0)
5503                 cfg.fc_metric = IP6_RT_PRIO_USER;
5504
5505         if (cfg.fc_mp)
5506                 return ip6_route_multipath_add(&cfg, extack);
5507         else
5508                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5509 }
5510
5511 /* add the overhead of this fib6_nh to nexthop_len */
5512 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5513 {
5514         int *nexthop_len = arg;
5515
5516         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5517                      + NLA_ALIGN(sizeof(struct rtnexthop))
5518                      + nla_total_size(16); /* RTA_GATEWAY */
5519
5520         if (nh->fib_nh_lws) {
5521                 /* RTA_ENCAP_TYPE */
5522                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5523                 /* RTA_ENCAP */
5524                 *nexthop_len += nla_total_size(2);
5525         }
5526
5527         return 0;
5528 }
5529
5530 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5531 {
5532         int nexthop_len;
5533
5534         if (f6i->nh) {
5535                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5536                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5537                                          &nexthop_len);
5538         } else {
5539                 struct fib6_info *sibling, *next_sibling;
5540                 struct fib6_nh *nh = f6i->fib6_nh;
5541
5542                 nexthop_len = 0;
5543                 if (f6i->fib6_nsiblings) {
5544                         rt6_nh_nlmsg_size(nh, &nexthop_len);
5545
5546                         list_for_each_entry_safe(sibling, next_sibling,
5547                                                  &f6i->fib6_siblings, fib6_siblings) {
5548                                 rt6_nh_nlmsg_size(sibling->fib6_nh, &nexthop_len);
5549                         }
5550                 }
5551                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5552         }
5553
5554         return NLMSG_ALIGN(sizeof(struct rtmsg))
5555                + nla_total_size(16) /* RTA_SRC */
5556                + nla_total_size(16) /* RTA_DST */
5557                + nla_total_size(16) /* RTA_GATEWAY */
5558                + nla_total_size(16) /* RTA_PREFSRC */
5559                + nla_total_size(4) /* RTA_TABLE */
5560                + nla_total_size(4) /* RTA_IIF */
5561                + nla_total_size(4) /* RTA_OIF */
5562                + nla_total_size(4) /* RTA_PRIORITY */
5563                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5564                + nla_total_size(sizeof(struct rta_cacheinfo))
5565                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5566                + nla_total_size(1) /* RTA_PREF */
5567                + nexthop_len;
5568 }
5569
5570 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5571                                  unsigned char *flags)
5572 {
5573         if (nexthop_is_multipath(nh)) {
5574                 struct nlattr *mp;
5575
5576                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5577                 if (!mp)
5578                         goto nla_put_failure;
5579
5580                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5581                         goto nla_put_failure;
5582
5583                 nla_nest_end(skb, mp);
5584         } else {
5585                 struct fib6_nh *fib6_nh;
5586
5587                 fib6_nh = nexthop_fib6_nh(nh);
5588                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5589                                      flags, false) < 0)
5590                         goto nla_put_failure;
5591         }
5592
5593         return 0;
5594
5595 nla_put_failure:
5596         return -EMSGSIZE;
5597 }
5598
5599 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5600                          struct fib6_info *rt, struct dst_entry *dst,
5601                          struct in6_addr *dest, struct in6_addr *src,
5602                          int iif, int type, u32 portid, u32 seq,
5603                          unsigned int flags)
5604 {
5605         struct rt6_info *rt6 = (struct rt6_info *)dst;
5606         struct rt6key *rt6_dst, *rt6_src;
5607         u32 *pmetrics, table, rt6_flags;
5608         unsigned char nh_flags = 0;
5609         struct nlmsghdr *nlh;
5610         struct rtmsg *rtm;
5611         long expires = 0;
5612
5613         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5614         if (!nlh)
5615                 return -EMSGSIZE;
5616
5617         if (rt6) {
5618                 rt6_dst = &rt6->rt6i_dst;
5619                 rt6_src = &rt6->rt6i_src;
5620                 rt6_flags = rt6->rt6i_flags;
5621         } else {
5622                 rt6_dst = &rt->fib6_dst;
5623                 rt6_src = &rt->fib6_src;
5624                 rt6_flags = rt->fib6_flags;
5625         }
5626
5627         rtm = nlmsg_data(nlh);
5628         rtm->rtm_family = AF_INET6;
5629         rtm->rtm_dst_len = rt6_dst->plen;
5630         rtm->rtm_src_len = rt6_src->plen;
5631         rtm->rtm_tos = 0;
5632         if (rt->fib6_table)
5633                 table = rt->fib6_table->tb6_id;
5634         else
5635                 table = RT6_TABLE_UNSPEC;
5636         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5637         if (nla_put_u32(skb, RTA_TABLE, table))
5638                 goto nla_put_failure;
5639
5640         rtm->rtm_type = rt->fib6_type;
5641         rtm->rtm_flags = 0;
5642         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5643         rtm->rtm_protocol = rt->fib6_protocol;
5644
5645         if (rt6_flags & RTF_CACHE)
5646                 rtm->rtm_flags |= RTM_F_CLONED;
5647
5648         if (dest) {
5649                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5650                         goto nla_put_failure;
5651                 rtm->rtm_dst_len = 128;
5652         } else if (rtm->rtm_dst_len)
5653                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5654                         goto nla_put_failure;
5655 #ifdef CONFIG_IPV6_SUBTREES
5656         if (src) {
5657                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5658                         goto nla_put_failure;
5659                 rtm->rtm_src_len = 128;
5660         } else if (rtm->rtm_src_len &&
5661                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5662                 goto nla_put_failure;
5663 #endif
5664         if (iif) {
5665 #ifdef CONFIG_IPV6_MROUTE
5666                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5667                         int err = ip6mr_get_route(net, skb, rtm, portid);
5668
5669                         if (err == 0)
5670                                 return 0;
5671                         if (err < 0)
5672                                 goto nla_put_failure;
5673                 } else
5674 #endif
5675                         if (nla_put_u32(skb, RTA_IIF, iif))
5676                                 goto nla_put_failure;
5677         } else if (dest) {
5678                 struct in6_addr saddr_buf;
5679                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5680                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5681                         goto nla_put_failure;
5682         }
5683
5684         if (rt->fib6_prefsrc.plen) {
5685                 struct in6_addr saddr_buf;
5686                 saddr_buf = rt->fib6_prefsrc.addr;
5687                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5688                         goto nla_put_failure;
5689         }
5690
5691         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5692         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5693                 goto nla_put_failure;
5694
5695         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5696                 goto nla_put_failure;
5697
5698         /* For multipath routes, walk the siblings list and add
5699          * each as a nexthop within RTA_MULTIPATH.
5700          */
5701         if (rt6) {
5702                 if (rt6_flags & RTF_GATEWAY &&
5703                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5704                         goto nla_put_failure;
5705
5706                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5707                         goto nla_put_failure;
5708
5709                 if (dst->lwtstate &&
5710                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5711                         goto nla_put_failure;
5712         } else if (rt->fib6_nsiblings) {
5713                 struct fib6_info *sibling, *next_sibling;
5714                 struct nlattr *mp;
5715
5716                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5717                 if (!mp)
5718                         goto nla_put_failure;
5719
5720                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5721                                     rt->fib6_nh->fib_nh_weight, AF_INET6,
5722                                     0) < 0)
5723                         goto nla_put_failure;
5724
5725                 list_for_each_entry_safe(sibling, next_sibling,
5726                                          &rt->fib6_siblings, fib6_siblings) {
5727                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5728                                             sibling->fib6_nh->fib_nh_weight,
5729                                             AF_INET6, 0) < 0)
5730                                 goto nla_put_failure;
5731                 }
5732
5733                 nla_nest_end(skb, mp);
5734         } else if (rt->nh) {
5735                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5736                         goto nla_put_failure;
5737
5738                 if (nexthop_is_blackhole(rt->nh))
5739                         rtm->rtm_type = RTN_BLACKHOLE;
5740
5741                 if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5742                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5743                         goto nla_put_failure;
5744
5745                 rtm->rtm_flags |= nh_flags;
5746         } else {
5747                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5748                                      &nh_flags, false) < 0)
5749                         goto nla_put_failure;
5750
5751                 rtm->rtm_flags |= nh_flags;
5752         }
5753
5754         if (rt6_flags & RTF_EXPIRES) {
5755                 expires = dst ? dst->expires : rt->expires;
5756                 expires -= jiffies;
5757         }
5758
5759         if (!dst) {
5760                 if (READ_ONCE(rt->offload))
5761                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5762                 if (READ_ONCE(rt->trap))
5763                         rtm->rtm_flags |= RTM_F_TRAP;
5764                 if (READ_ONCE(rt->offload_failed))
5765                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5766         }
5767
5768         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5769                 goto nla_put_failure;
5770
5771         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5772                 goto nla_put_failure;
5773
5774
5775         nlmsg_end(skb, nlh);
5776         return 0;
5777
5778 nla_put_failure:
5779         nlmsg_cancel(skb, nlh);
5780         return -EMSGSIZE;
5781 }
5782
5783 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5784 {
5785         const struct net_device *dev = arg;
5786
5787         if (nh->fib_nh_dev == dev)
5788                 return 1;
5789
5790         return 0;
5791 }
5792
5793 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5794                                const struct net_device *dev)
5795 {
5796         if (f6i->nh) {
5797                 struct net_device *_dev = (struct net_device *)dev;
5798
5799                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5800                                                   fib6_info_nh_uses_dev,
5801                                                   _dev);
5802         }
5803
5804         if (f6i->fib6_nh->fib_nh_dev == dev)
5805                 return true;
5806
5807         if (f6i->fib6_nsiblings) {
5808                 struct fib6_info *sibling, *next_sibling;
5809
5810                 list_for_each_entry_safe(sibling, next_sibling,
5811                                          &f6i->fib6_siblings, fib6_siblings) {
5812                         if (sibling->fib6_nh->fib_nh_dev == dev)
5813                                 return true;
5814                 }
5815         }
5816
5817         return false;
5818 }
5819
5820 struct fib6_nh_exception_dump_walker {
5821         struct rt6_rtnl_dump_arg *dump;
5822         struct fib6_info *rt;
5823         unsigned int flags;
5824         unsigned int skip;
5825         unsigned int count;
5826 };
5827
5828 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5829 {
5830         struct fib6_nh_exception_dump_walker *w = arg;
5831         struct rt6_rtnl_dump_arg *dump = w->dump;
5832         struct rt6_exception_bucket *bucket;
5833         struct rt6_exception *rt6_ex;
5834         int i, err;
5835
5836         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5837         if (!bucket)
5838                 return 0;
5839
5840         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5841                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5842                         if (w->skip) {
5843                                 w->skip--;
5844                                 continue;
5845                         }
5846
5847                         /* Expiration of entries doesn't bump sernum, insertion
5848                          * does. Removal is triggered by insertion, so we can
5849                          * rely on the fact that if entries change between two
5850                          * partial dumps, this node is scanned again completely,
5851                          * see rt6_insert_exception() and fib6_dump_table().
5852                          *
5853                          * Count expired entries we go through as handled
5854                          * entries that we'll skip next time, in case of partial
5855                          * node dump. Otherwise, if entries expire meanwhile,
5856                          * we'll skip the wrong amount.
5857                          */
5858                         if (rt6_check_expired(rt6_ex->rt6i)) {
5859                                 w->count++;
5860                                 continue;
5861                         }
5862
5863                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5864                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5865                                             RTM_NEWROUTE,
5866                                             NETLINK_CB(dump->cb->skb).portid,
5867                                             dump->cb->nlh->nlmsg_seq, w->flags);
5868                         if (err)
5869                                 return err;
5870
5871                         w->count++;
5872                 }
5873                 bucket++;
5874         }
5875
5876         return 0;
5877 }
5878
5879 /* Return -1 if done with node, number of handled routes on partial dump */
5880 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5881 {
5882         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5883         struct fib_dump_filter *filter = &arg->filter;
5884         unsigned int flags = NLM_F_MULTI;
5885         struct net *net = arg->net;
5886         int count = 0;
5887
5888         if (rt == net->ipv6.fib6_null_entry)
5889                 return -1;
5890
5891         if ((filter->flags & RTM_F_PREFIX) &&
5892             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5893                 /* success since this is not a prefix route */
5894                 return -1;
5895         }
5896         if (filter->filter_set &&
5897             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5898              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5899              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5900                 return -1;
5901         }
5902
5903         if (filter->filter_set ||
5904             !filter->dump_routes || !filter->dump_exceptions) {
5905                 flags |= NLM_F_DUMP_FILTERED;
5906         }
5907
5908         if (filter->dump_routes) {
5909                 if (skip) {
5910                         skip--;
5911                 } else {
5912                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5913                                           0, RTM_NEWROUTE,
5914                                           NETLINK_CB(arg->cb->skb).portid,
5915                                           arg->cb->nlh->nlmsg_seq, flags)) {
5916                                 return 0;
5917                         }
5918                         count++;
5919                 }
5920         }
5921
5922         if (filter->dump_exceptions) {
5923                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5924                                                            .rt = rt,
5925                                                            .flags = flags,
5926                                                            .skip = skip,
5927                                                            .count = 0 };
5928                 int err;
5929
5930                 rcu_read_lock();
5931                 if (rt->nh) {
5932                         err = nexthop_for_each_fib6_nh(rt->nh,
5933                                                        rt6_nh_dump_exceptions,
5934                                                        &w);
5935                 } else {
5936                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5937                 }
5938                 rcu_read_unlock();
5939
5940                 if (err)
5941                         return count + w.count;
5942         }
5943
5944         return -1;
5945 }
5946
5947 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5948                                         const struct nlmsghdr *nlh,
5949                                         struct nlattr **tb,
5950                                         struct netlink_ext_ack *extack)
5951 {
5952         struct rtmsg *rtm;
5953         int i, err;
5954
5955         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5956                 NL_SET_ERR_MSG_MOD(extack,
5957                                    "Invalid header for get route request");
5958                 return -EINVAL;
5959         }
5960
5961         if (!netlink_strict_get_check(skb))
5962                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5963                                               rtm_ipv6_policy, extack);
5964
5965         rtm = nlmsg_data(nlh);
5966         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5967             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5968             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5969             rtm->rtm_type) {
5970                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5971                 return -EINVAL;
5972         }
5973         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5974                 NL_SET_ERR_MSG_MOD(extack,
5975                                    "Invalid flags for get route request");
5976                 return -EINVAL;
5977         }
5978
5979         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5980                                             rtm_ipv6_policy, extack);
5981         if (err)
5982                 return err;
5983
5984         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5985             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5986                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5987                 return -EINVAL;
5988         }
5989
5990         for (i = 0; i <= RTA_MAX; i++) {
5991                 if (!tb[i])
5992                         continue;
5993
5994                 switch (i) {
5995                 case RTA_SRC:
5996                 case RTA_DST:
5997                 case RTA_IIF:
5998                 case RTA_OIF:
5999                 case RTA_MARK:
6000                 case RTA_UID:
6001                 case RTA_SPORT:
6002                 case RTA_DPORT:
6003                 case RTA_IP_PROTO:
6004                         break;
6005                 default:
6006                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6007                         return -EINVAL;
6008                 }
6009         }
6010
6011         return 0;
6012 }
6013
6014 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6015                               struct netlink_ext_ack *extack)
6016 {
6017         struct net *net = sock_net(in_skb->sk);
6018         struct nlattr *tb[RTA_MAX+1];
6019         int err, iif = 0, oif = 0;
6020         struct fib6_info *from;
6021         struct dst_entry *dst;
6022         struct rt6_info *rt;
6023         struct sk_buff *skb;
6024         struct rtmsg *rtm;
6025         struct flowi6 fl6 = {};
6026         bool fibmatch;
6027
6028         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6029         if (err < 0)
6030                 goto errout;
6031
6032         err = -EINVAL;
6033         rtm = nlmsg_data(nlh);
6034         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6035         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6036
6037         if (tb[RTA_SRC]) {
6038                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6039                         goto errout;
6040
6041                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6042         }
6043
6044         if (tb[RTA_DST]) {
6045                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6046                         goto errout;
6047
6048                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6049         }
6050
6051         if (tb[RTA_IIF])
6052                 iif = nla_get_u32(tb[RTA_IIF]);
6053
6054         if (tb[RTA_OIF])
6055                 oif = nla_get_u32(tb[RTA_OIF]);
6056
6057         if (tb[RTA_MARK])
6058                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6059
6060         if (tb[RTA_UID])
6061                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6062                                            nla_get_u32(tb[RTA_UID]));
6063         else
6064                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6065
6066         if (tb[RTA_SPORT])
6067                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6068
6069         if (tb[RTA_DPORT])
6070                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6071
6072         if (tb[RTA_IP_PROTO]) {
6073                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6074                                                   &fl6.flowi6_proto, AF_INET6,
6075                                                   extack);
6076                 if (err)
6077                         goto errout;
6078         }
6079
6080         if (iif) {
6081                 struct net_device *dev;
6082                 int flags = 0;
6083
6084                 rcu_read_lock();
6085
6086                 dev = dev_get_by_index_rcu(net, iif);
6087                 if (!dev) {
6088                         rcu_read_unlock();
6089                         err = -ENODEV;
6090                         goto errout;
6091                 }
6092
6093                 fl6.flowi6_iif = iif;
6094
6095                 if (!ipv6_addr_any(&fl6.saddr))
6096                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6097
6098                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6099
6100                 rcu_read_unlock();
6101         } else {
6102                 fl6.flowi6_oif = oif;
6103
6104                 dst = ip6_route_output(net, NULL, &fl6);
6105         }
6106
6107
6108         rt = container_of(dst, struct rt6_info, dst);
6109         if (rt->dst.error) {
6110                 err = rt->dst.error;
6111                 ip6_rt_put(rt);
6112                 goto errout;
6113         }
6114
6115         if (rt == net->ipv6.ip6_null_entry) {
6116                 err = rt->dst.error;
6117                 ip6_rt_put(rt);
6118                 goto errout;
6119         }
6120
6121         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6122         if (!skb) {
6123                 ip6_rt_put(rt);
6124                 err = -ENOBUFS;
6125                 goto errout;
6126         }
6127
6128         skb_dst_set(skb, &rt->dst);
6129
6130         rcu_read_lock();
6131         from = rcu_dereference(rt->from);
6132         if (from) {
6133                 if (fibmatch)
6134                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6135                                             iif, RTM_NEWROUTE,
6136                                             NETLINK_CB(in_skb).portid,
6137                                             nlh->nlmsg_seq, 0);
6138                 else
6139                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6140                                             &fl6.saddr, iif, RTM_NEWROUTE,
6141                                             NETLINK_CB(in_skb).portid,
6142                                             nlh->nlmsg_seq, 0);
6143         } else {
6144                 err = -ENETUNREACH;
6145         }
6146         rcu_read_unlock();
6147
6148         if (err < 0) {
6149                 kfree_skb(skb);
6150                 goto errout;
6151         }
6152
6153         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6154 errout:
6155         return err;
6156 }
6157
6158 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6159                      unsigned int nlm_flags)
6160 {
6161         struct sk_buff *skb;
6162         struct net *net = info->nl_net;
6163         u32 seq;
6164         int err;
6165
6166         err = -ENOBUFS;
6167         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6168
6169         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6170         if (!skb)
6171                 goto errout;
6172
6173         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6174                             event, info->portid, seq, nlm_flags);
6175         if (err < 0) {
6176                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6177                 WARN_ON(err == -EMSGSIZE);
6178                 kfree_skb(skb);
6179                 goto errout;
6180         }
6181         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6182                     info->nlh, gfp_any());
6183         return;
6184 errout:
6185         if (err < 0)
6186                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6187 }
6188
6189 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6190                     struct nl_info *info)
6191 {
6192         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6193         struct sk_buff *skb;
6194         int err = -ENOBUFS;
6195
6196         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6197         if (!skb)
6198                 goto errout;
6199
6200         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6201                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6202         if (err < 0) {
6203                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6204                 WARN_ON(err == -EMSGSIZE);
6205                 kfree_skb(skb);
6206                 goto errout;
6207         }
6208         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6209                     info->nlh, gfp_any());
6210         return;
6211 errout:
6212         if (err < 0)
6213                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6214 }
6215
6216 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6217                             bool offload, bool trap, bool offload_failed)
6218 {
6219         struct sk_buff *skb;
6220         int err;
6221
6222         if (READ_ONCE(f6i->offload) == offload &&
6223             READ_ONCE(f6i->trap) == trap &&
6224             READ_ONCE(f6i->offload_failed) == offload_failed)
6225                 return;
6226
6227         WRITE_ONCE(f6i->offload, offload);
6228         WRITE_ONCE(f6i->trap, trap);
6229
6230         /* 2 means send notifications only if offload_failed was changed. */
6231         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6232             READ_ONCE(f6i->offload_failed) == offload_failed)
6233                 return;
6234
6235         WRITE_ONCE(f6i->offload_failed, offload_failed);
6236
6237         if (!rcu_access_pointer(f6i->fib6_node))
6238                 /* The route was removed from the tree, do not send
6239                  * notification.
6240                  */
6241                 return;
6242
6243         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6244                 return;
6245
6246         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6247         if (!skb) {
6248                 err = -ENOBUFS;
6249                 goto errout;
6250         }
6251
6252         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6253                             0, 0);
6254         if (err < 0) {
6255                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6256                 WARN_ON(err == -EMSGSIZE);
6257                 kfree_skb(skb);
6258                 goto errout;
6259         }
6260
6261         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6262         return;
6263
6264 errout:
6265         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6266 }
6267 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6268
6269 static int ip6_route_dev_notify(struct notifier_block *this,
6270                                 unsigned long event, void *ptr)
6271 {
6272         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6273         struct net *net = dev_net(dev);
6274
6275         if (!(dev->flags & IFF_LOOPBACK))
6276                 return NOTIFY_OK;
6277
6278         if (event == NETDEV_REGISTER) {
6279                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6280                 net->ipv6.ip6_null_entry->dst.dev = dev;
6281                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6282 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6283                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6284                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6285                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6286                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6287 #endif
6288          } else if (event == NETDEV_UNREGISTER &&
6289                     dev->reg_state != NETREG_UNREGISTERED) {
6290                 /* NETDEV_UNREGISTER could be fired for multiple times by
6291                  * netdev_wait_allrefs(). Make sure we only call this once.
6292                  */
6293                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6294 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6295                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6296                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6297 #endif
6298         }
6299
6300         return NOTIFY_OK;
6301 }
6302
6303 /*
6304  *      /proc
6305  */
6306
6307 #ifdef CONFIG_PROC_FS
6308 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6309 {
6310         struct net *net = (struct net *)seq->private;
6311         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6312                    net->ipv6.rt6_stats->fib_nodes,
6313                    net->ipv6.rt6_stats->fib_route_nodes,
6314                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6315                    net->ipv6.rt6_stats->fib_rt_entries,
6316                    net->ipv6.rt6_stats->fib_rt_cache,
6317                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6318                    net->ipv6.rt6_stats->fib_discarded_routes);
6319
6320         return 0;
6321 }
6322 #endif  /* CONFIG_PROC_FS */
6323
6324 #ifdef CONFIG_SYSCTL
6325
6326 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6327                               void *buffer, size_t *lenp, loff_t *ppos)
6328 {
6329         struct net *net;
6330         int delay;
6331         int ret;
6332         if (!write)
6333                 return -EINVAL;
6334
6335         net = (struct net *)ctl->extra1;
6336         delay = net->ipv6.sysctl.flush_delay;
6337         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6338         if (ret)
6339                 return ret;
6340
6341         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6342         return 0;
6343 }
6344
6345 static struct ctl_table ipv6_route_table_template[] = {
6346         {
6347                 .procname       =       "max_size",
6348                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6349                 .maxlen         =       sizeof(int),
6350                 .mode           =       0644,
6351                 .proc_handler   =       proc_dointvec,
6352         },
6353         {
6354                 .procname       =       "gc_thresh",
6355                 .data           =       &ip6_dst_ops_template.gc_thresh,
6356                 .maxlen         =       sizeof(int),
6357                 .mode           =       0644,
6358                 .proc_handler   =       proc_dointvec,
6359         },
6360         {
6361                 .procname       =       "flush",
6362                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6363                 .maxlen         =       sizeof(int),
6364                 .mode           =       0200,
6365                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6366         },
6367         {
6368                 .procname       =       "gc_min_interval",
6369                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6370                 .maxlen         =       sizeof(int),
6371                 .mode           =       0644,
6372                 .proc_handler   =       proc_dointvec_jiffies,
6373         },
6374         {
6375                 .procname       =       "gc_timeout",
6376                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6377                 .maxlen         =       sizeof(int),
6378                 .mode           =       0644,
6379                 .proc_handler   =       proc_dointvec_jiffies,
6380         },
6381         {
6382                 .procname       =       "gc_interval",
6383                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6384                 .maxlen         =       sizeof(int),
6385                 .mode           =       0644,
6386                 .proc_handler   =       proc_dointvec_jiffies,
6387         },
6388         {
6389                 .procname       =       "gc_elasticity",
6390                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6391                 .maxlen         =       sizeof(int),
6392                 .mode           =       0644,
6393                 .proc_handler   =       proc_dointvec,
6394         },
6395         {
6396                 .procname       =       "mtu_expires",
6397                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6398                 .maxlen         =       sizeof(int),
6399                 .mode           =       0644,
6400                 .proc_handler   =       proc_dointvec_jiffies,
6401         },
6402         {
6403                 .procname       =       "min_adv_mss",
6404                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6405                 .maxlen         =       sizeof(int),
6406                 .mode           =       0644,
6407                 .proc_handler   =       proc_dointvec,
6408         },
6409         {
6410                 .procname       =       "gc_min_interval_ms",
6411                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6412                 .maxlen         =       sizeof(int),
6413                 .mode           =       0644,
6414                 .proc_handler   =       proc_dointvec_ms_jiffies,
6415         },
6416         {
6417                 .procname       =       "skip_notify_on_dev_down",
6418                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6419                 .maxlen         =       sizeof(u8),
6420                 .mode           =       0644,
6421                 .proc_handler   =       proc_dou8vec_minmax,
6422                 .extra1         =       SYSCTL_ZERO,
6423                 .extra2         =       SYSCTL_ONE,
6424         },
6425         { }
6426 };
6427
6428 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6429 {
6430         struct ctl_table *table;
6431
6432         table = kmemdup(ipv6_route_table_template,
6433                         sizeof(ipv6_route_table_template),
6434                         GFP_KERNEL);
6435
6436         if (table) {
6437                 table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6438                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6439                 table[2].data = &net->ipv6.sysctl.flush_delay;
6440                 table[2].extra1 = net;
6441                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6442                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6443                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6444                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6445                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6446                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6447                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6448                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6449
6450                 /* Don't export sysctls to unprivileged users */
6451                 if (net->user_ns != &init_user_ns)
6452                         table[1].procname = NULL;
6453         }
6454
6455         return table;
6456 }
6457 #endif
6458
6459 static int __net_init ip6_route_net_init(struct net *net)
6460 {
6461         int ret = -ENOMEM;
6462
6463         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6464                sizeof(net->ipv6.ip6_dst_ops));
6465
6466         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6467                 goto out_ip6_dst_ops;
6468
6469         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6470         if (!net->ipv6.fib6_null_entry)
6471                 goto out_ip6_dst_entries;
6472         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6473                sizeof(*net->ipv6.fib6_null_entry));
6474
6475         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6476                                            sizeof(*net->ipv6.ip6_null_entry),
6477                                            GFP_KERNEL);
6478         if (!net->ipv6.ip6_null_entry)
6479                 goto out_fib6_null_entry;
6480         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6481         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6482                          ip6_template_metrics, true);
6483         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->dst.rt_uncached);
6484
6485 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6486         net->ipv6.fib6_has_custom_rules = false;
6487         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6488                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6489                                                GFP_KERNEL);
6490         if (!net->ipv6.ip6_prohibit_entry)
6491                 goto out_ip6_null_entry;
6492         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6493         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6494                          ip6_template_metrics, true);
6495         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->dst.rt_uncached);
6496
6497         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6498                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6499                                                GFP_KERNEL);
6500         if (!net->ipv6.ip6_blk_hole_entry)
6501                 goto out_ip6_prohibit_entry;
6502         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6503         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6504                          ip6_template_metrics, true);
6505         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->dst.rt_uncached);
6506 #ifdef CONFIG_IPV6_SUBTREES
6507         net->ipv6.fib6_routes_require_src = 0;
6508 #endif
6509 #endif
6510
6511         net->ipv6.sysctl.flush_delay = 0;
6512         net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
6513         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6514         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6515         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6516         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6517         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6518         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6519         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6520
6521         atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6522
6523         ret = 0;
6524 out:
6525         return ret;
6526
6527 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6528 out_ip6_prohibit_entry:
6529         kfree(net->ipv6.ip6_prohibit_entry);
6530 out_ip6_null_entry:
6531         kfree(net->ipv6.ip6_null_entry);
6532 #endif
6533 out_fib6_null_entry:
6534         kfree(net->ipv6.fib6_null_entry);
6535 out_ip6_dst_entries:
6536         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6537 out_ip6_dst_ops:
6538         goto out;
6539 }
6540
6541 static void __net_exit ip6_route_net_exit(struct net *net)
6542 {
6543         kfree(net->ipv6.fib6_null_entry);
6544         kfree(net->ipv6.ip6_null_entry);
6545 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6546         kfree(net->ipv6.ip6_prohibit_entry);
6547         kfree(net->ipv6.ip6_blk_hole_entry);
6548 #endif
6549         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6550 }
6551
6552 static int __net_init ip6_route_net_init_late(struct net *net)
6553 {
6554 #ifdef CONFIG_PROC_FS
6555         if (!proc_create_net("ipv6_route", 0, net->proc_net,
6556                              &ipv6_route_seq_ops,
6557                              sizeof(struct ipv6_route_iter)))
6558                 return -ENOMEM;
6559
6560         if (!proc_create_net_single("rt6_stats", 0444, net->proc_net,
6561                                     rt6_stats_seq_show, NULL)) {
6562                 remove_proc_entry("ipv6_route", net->proc_net);
6563                 return -ENOMEM;
6564         }
6565 #endif
6566         return 0;
6567 }
6568
6569 static void __net_exit ip6_route_net_exit_late(struct net *net)
6570 {
6571 #ifdef CONFIG_PROC_FS
6572         remove_proc_entry("ipv6_route", net->proc_net);
6573         remove_proc_entry("rt6_stats", net->proc_net);
6574 #endif
6575 }
6576
6577 static struct pernet_operations ip6_route_net_ops = {
6578         .init = ip6_route_net_init,
6579         .exit = ip6_route_net_exit,
6580 };
6581
6582 static int __net_init ipv6_inetpeer_init(struct net *net)
6583 {
6584         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6585
6586         if (!bp)
6587                 return -ENOMEM;
6588         inet_peer_base_init(bp);
6589         net->ipv6.peers = bp;
6590         return 0;
6591 }
6592
6593 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6594 {
6595         struct inet_peer_base *bp = net->ipv6.peers;
6596
6597         net->ipv6.peers = NULL;
6598         inetpeer_invalidate_tree(bp);
6599         kfree(bp);
6600 }
6601
6602 static struct pernet_operations ipv6_inetpeer_ops = {
6603         .init   =       ipv6_inetpeer_init,
6604         .exit   =       ipv6_inetpeer_exit,
6605 };
6606
6607 static struct pernet_operations ip6_route_net_late_ops = {
6608         .init = ip6_route_net_init_late,
6609         .exit = ip6_route_net_exit_late,
6610 };
6611
6612 static struct notifier_block ip6_route_dev_notifier = {
6613         .notifier_call = ip6_route_dev_notify,
6614         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6615 };
6616
6617 void __init ip6_route_init_special_entries(void)
6618 {
6619         /* Registering of the loopback is done before this portion of code,
6620          * the loopback reference in rt6_info will not be taken, do it
6621          * manually for init_net */
6622         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6623         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6624         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6625   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6626         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6627         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6628         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6629         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6630   #endif
6631 }
6632
6633 #if IS_BUILTIN(CONFIG_IPV6)
6634 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6635 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6636
6637 BTF_ID_LIST(btf_fib6_info_id)
6638 BTF_ID(struct, fib6_info)
6639
6640 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6641         .seq_ops                = &ipv6_route_seq_ops,
6642         .init_seq_private       = bpf_iter_init_seq_net,
6643         .fini_seq_private       = bpf_iter_fini_seq_net,
6644         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6645 };
6646
6647 static struct bpf_iter_reg ipv6_route_reg_info = {
6648         .target                 = "ipv6_route",
6649         .ctx_arg_info_size      = 1,
6650         .ctx_arg_info           = {
6651                 { offsetof(struct bpf_iter__ipv6_route, rt),
6652                   PTR_TO_BTF_ID_OR_NULL },
6653         },
6654         .seq_info               = &ipv6_route_seq_info,
6655 };
6656
6657 static int __init bpf_iter_register(void)
6658 {
6659         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6660         return bpf_iter_reg_target(&ipv6_route_reg_info);
6661 }
6662
6663 static void bpf_iter_unregister(void)
6664 {
6665         bpf_iter_unreg_target(&ipv6_route_reg_info);
6666 }
6667 #endif
6668 #endif
6669
6670 int __init ip6_route_init(void)
6671 {
6672         int ret;
6673         int cpu;
6674
6675         ret = -ENOMEM;
6676         ip6_dst_ops_template.kmem_cachep =
6677                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6678                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6679         if (!ip6_dst_ops_template.kmem_cachep)
6680                 goto out;
6681
6682         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6683         if (ret)
6684                 goto out_kmem_cache;
6685
6686         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6687         if (ret)
6688                 goto out_dst_entries;
6689
6690         ret = register_pernet_subsys(&ip6_route_net_ops);
6691         if (ret)
6692                 goto out_register_inetpeer;
6693
6694         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6695
6696         ret = fib6_init();
6697         if (ret)
6698                 goto out_register_subsys;
6699
6700         ret = xfrm6_init();
6701         if (ret)
6702                 goto out_fib6_init;
6703
6704         ret = fib6_rules_init();
6705         if (ret)
6706                 goto xfrm6_init;
6707
6708         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6709         if (ret)
6710                 goto fib6_rules_init;
6711
6712         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6713                                    inet6_rtm_newroute, NULL, 0);
6714         if (ret < 0)
6715                 goto out_register_late_subsys;
6716
6717         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6718                                    inet6_rtm_delroute, NULL, 0);
6719         if (ret < 0)
6720                 goto out_register_late_subsys;
6721
6722         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6723                                    inet6_rtm_getroute, NULL,
6724                                    RTNL_FLAG_DOIT_UNLOCKED);
6725         if (ret < 0)
6726                 goto out_register_late_subsys;
6727
6728         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6729         if (ret)
6730                 goto out_register_late_subsys;
6731
6732 #if IS_BUILTIN(CONFIG_IPV6)
6733 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6734         ret = bpf_iter_register();
6735         if (ret)
6736                 goto out_register_late_subsys;
6737 #endif
6738 #endif
6739
6740         for_each_possible_cpu(cpu) {
6741                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6742
6743                 INIT_LIST_HEAD(&ul->head);
6744                 INIT_LIST_HEAD(&ul->quarantine);
6745                 spin_lock_init(&ul->lock);
6746         }
6747
6748 out:
6749         return ret;
6750
6751 out_register_late_subsys:
6752         rtnl_unregister_all(PF_INET6);
6753         unregister_pernet_subsys(&ip6_route_net_late_ops);
6754 fib6_rules_init:
6755         fib6_rules_cleanup();
6756 xfrm6_init:
6757         xfrm6_fini();
6758 out_fib6_init:
6759         fib6_gc_cleanup();
6760 out_register_subsys:
6761         unregister_pernet_subsys(&ip6_route_net_ops);
6762 out_register_inetpeer:
6763         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6764 out_dst_entries:
6765         dst_entries_destroy(&ip6_dst_blackhole_ops);
6766 out_kmem_cache:
6767         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6768         goto out;
6769 }
6770
6771 void ip6_route_cleanup(void)
6772 {
6773 #if IS_BUILTIN(CONFIG_IPV6)
6774 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6775         bpf_iter_unregister();
6776 #endif
6777 #endif
6778         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6779         unregister_pernet_subsys(&ip6_route_net_late_ops);
6780         fib6_rules_cleanup();
6781         xfrm6_fini();
6782         fib6_gc_cleanup();
6783         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6784         unregister_pernet_subsys(&ip6_route_net_ops);
6785         dst_entries_destroy(&ip6_dst_blackhole_ops);
6786         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6787 }