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             <[email protected]>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <linux/siphash.h>
  45 #include <net/net_namespace.h>
  46 #include <net/snmp.h>
  47 #include <net/ipv6.h>
  48 #include <net/ip6_fib.h>
  49 #include <net/ip6_route.h>
  50 #include <net/ndisc.h>
  51 #include <net/addrconf.h>
  52 #include <net/tcp.h>
  53 #include <linux/rtnetlink.h>
  54 #include <net/dst.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/xfrm.h>
  57 #include <net/netevent.h>
  58 #include <net/netlink.h>
  59 #include <net/rtnh.h>
  60 #include <net/lwtunnel.h>
  61 #include <net/ip_tunnels.h>
  62 #include <net/l3mdev.h>
  63 #include <net/ip.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/btf_ids.h>
  66
  67 #ifdef CONFIG_SYSCTL
  68 #include <linux/sysctl.h>
  69 #endif
  70
  71 static int ip6_rt_type_to_error(u8 fib6_type);
  72
  73 #define CREATE_TRACE_POINTS
  74 #include <trace/events/fib6.h>
  75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  76 #undef CREATE_TRACE_POINTS
  77
  78 enum rt6_nud_state {
  79         RT6_NUD_FAIL_HARD = -3,
  80         RT6_NUD_FAIL_PROBE = -2,
  81         RT6_NUD_FAIL_DO_RR = -1,
  82         RT6_NUD_SUCCEED = 1
  83 };
  84
  85 INDIRECT_CALLABLE_SCOPE
  86 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  87 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  88 INDIRECT_CALLABLE_SCOPE
  89 unsigned int            ip6_mtu(const struct dst_entry *dst);
  90 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  91 static void             ip6_dst_destroy(struct dst_entry *);
  92 static void             ip6_dst_ifdown(struct dst_entry *,
  93                                        struct net_device *dev, int how);
  94 static int               ip6_dst_gc(struct dst_ops *ops);
  95
  96 static int              ip6_pkt_discard(struct sk_buff *skb);
  97 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  98 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  99 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
 100 static void             ip6_link_failure(struct sk_buff *skb);
 101 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 102                                            struct sk_buff *skb, u32 mtu,
 103                                            bool confirm_neigh);
 104 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 105                                         struct sk_buff *skb);
 106 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 107                            int strict);
 108 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 109 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 110                          struct fib6_info *rt, struct dst_entry *dst,
 111                          struct in6_addr *dest, struct in6_addr *src,
 112                          int iif, int type, u32 portid, u32 seq,
 113                          unsigned int flags);
 114 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 115                                            const struct in6_addr *daddr,
 116                                            const struct in6_addr *saddr);
 117
 118 #ifdef CONFIG_IPV6_ROUTE_INFO
 119 static struct fib6_info *rt6_add_route_info(struct net *net,
 120                                            const struct in6_addr *prefix, int prefixlen,
 121                                            const struct in6_addr *gwaddr,
 122                                            struct net_device *dev,
 123                                            unsigned int pref);
 124 static struct fib6_info *rt6_get_route_info(struct net *net,
 125                                            const struct in6_addr *prefix, int prefixlen,
 126                                            const struct in6_addr *gwaddr,
 127                                            struct net_device *dev);
 128 #endif
 129
 130 struct uncached_list {
 131         spinlock_t              lock;
 132         struct list_head        head;
 133         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->rt6i_uncached_list = ul;
 143
 144         spin_lock_bh(&ul->lock);
 145         list_add_tail(&rt->rt6i_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->rt6i_uncached)) {
 152                 struct uncached_list *ul = rt->rt6i_uncached_list;
 153
 154                 spin_lock_bh(&ul->lock);
 155                 list_del_init(&rt->rt6i_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, rt6i_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                                 dev_replace_track(rt_dev, blackhole_netdev,
 186                                                   &rt->dst.dev_tracker,
 187                                                   GFP_ATOMIC);
 188                                 handled = true;
 189                         }
 190                         if (handled)
 191                                 list_move(&rt->rt6i_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                 .__refcnt       = ATOMIC_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                 .__refcnt       = ATOMIC_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                 .__refcnt       = ATOMIC_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         INIT_LIST_HEAD(&rt->rt6i_uncached);
 338 }
 339
 340 /* allocate dst with ip6_dst_ops */
 341 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 342                                int flags)
 343 {
 344         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 345                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 346
 347         if (rt) {
 348                 rt6_info_init(rt);
 349                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 350         }
 351
 352         return rt;
 353 }
 354 EXPORT_SYMBOL(ip6_dst_alloc);
 355
 356 static void ip6_dst_destroy(struct dst_entry *dst)
 357 {
 358         struct rt6_info *rt = (struct rt6_info *)dst;
 359         struct fib6_info *from;
 360         struct inet6_dev *idev;
 361
 362         ip_dst_metrics_put(dst);
 363         rt6_uncached_list_del(rt);
 364
 365         idev = rt->rt6i_idev;
 366         if (idev) {
 367                 rt->rt6i_idev = NULL;
 368                 in6_dev_put(idev);
 369         }
 370
 371         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 372         fib6_info_release(from);
 373 }
 374
 375 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 376                            int how)
 377 {
 378         struct rt6_info *rt = (struct rt6_info *)dst;
 379         struct inet6_dev *idev = rt->rt6i_idev;
 380
 381         if (idev && idev->dev != blackhole_netdev) {
 382                 struct inet6_dev *blackhole_idev = in6_dev_get(blackhole_netdev);
 383
 384                 if (blackhole_idev) {
 385                         rt->rt6i_idev = blackhole_idev;
 386                         in6_dev_put(idev);
 387                 }
 388         }
 389 }
 390
 391 static bool __rt6_check_expired(const struct rt6_info *rt)
 392 {
 393         if (rt->rt6i_flags & RTF_EXPIRES)
 394                 return time_after(jiffies, rt->dst.expires);
 395         else
 396                 return false;
 397 }
 398
 399 static bool rt6_check_expired(const struct rt6_info *rt)
 400 {
 401         struct fib6_info *from;
 402
 403         from = rcu_dereference(rt->from);
 404
 405         if (rt->rt6i_flags & RTF_EXPIRES) {
 406                 if (time_after(jiffies, rt->dst.expires))
 407                         return true;
 408         } else if (from) {
 409                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 410                         fib6_check_expired(from);
 411         }
 412         return false;
 413 }
 414
 415 void fib6_select_path(const struct net *net, struct fib6_result *res,
 416                       struct flowi6 *fl6, int oif, bool have_oif_match,
 417                       const struct sk_buff *skb, int strict)
 418 {
 419         struct fib6_info *sibling, *next_sibling;
 420         struct fib6_info *match = res->f6i;
 421
 422         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 423                 goto out;
 424
 425         if (match->nh && have_oif_match && res->nh)
 426                 return;
 427
 428         /* We might have already computed the hash for ICMPv6 errors. In such
 429          * case it will always be non-zero. Otherwise now is the time to do it.
 430          */
 431         if (!fl6->mp_hash &&
 432             (!match->nh || nexthop_is_multipath(match->nh)))
 433                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 434
 435         if (unlikely(match->nh)) {
 436                 nexthop_path_fib6_result(res, fl6->mp_hash);
 437                 return;
 438         }
 439
 440         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 441                 goto out;
 442
 443         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 444                                  fib6_siblings) {
 445                 const struct fib6_nh *nh = sibling->fib6_nh;
 446                 int nh_upper_bound;
 447
 448                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 449                 if (fl6->mp_hash > nh_upper_bound)
 450                         continue;
 451                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 452                         break;
 453                 match = sibling;
 454                 break;
 455         }
 456
 457 out:
 458         res->f6i = match;
 459         res->nh = match->fib6_nh;
 460 }
 461
 462 /*
 463  *      Route lookup. rcu_read_lock() should be held.
 464  */
 465
 466 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 467                                const struct in6_addr *saddr, int oif, int flags)
 468 {
 469         const struct net_device *dev;
 470
 471         if (nh->fib_nh_flags & RTNH_F_DEAD)
 472                 return false;
 473
 474         dev = nh->fib_nh_dev;
 475         if (oif) {
 476                 if (dev->ifindex == oif)
 477                         return true;
 478         } else {
 479                 if (ipv6_chk_addr(net, saddr, dev,
 480                                   flags & RT6_LOOKUP_F_IFACE))
 481                         return true;
 482         }
 483
 484         return false;
 485 }
 486
 487 struct fib6_nh_dm_arg {
 488         struct net              *net;
 489         const struct in6_addr   *saddr;
 490         int                     oif;
 491         int                     flags;
 492         struct fib6_nh          *nh;
 493 };
 494
 495 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 496 {
 497         struct fib6_nh_dm_arg *arg = _arg;
 498
 499         arg->nh = nh;
 500         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 501                                   arg->flags);
 502 }
 503
 504 /* returns fib6_nh from nexthop or NULL */
 505 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 506                                         struct fib6_result *res,
 507                                         const struct in6_addr *saddr,
 508                                         int oif, int flags)
 509 {
 510         struct fib6_nh_dm_arg arg = {
 511                 .net   = net,
 512                 .saddr = saddr,
 513                 .oif   = oif,
 514                 .flags = flags,
 515         };
 516
 517         if (nexthop_is_blackhole(nh))
 518                 return NULL;
 519
 520         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 521                 return arg.nh;
 522
 523         return NULL;
 524 }
 525
 526 static void rt6_device_match(struct net *net, struct fib6_result *res,
 527                              const struct in6_addr *saddr, int oif, int flags)
 528 {
 529         struct fib6_info *f6i = res->f6i;
 530         struct fib6_info *spf6i;
 531         struct fib6_nh *nh;
 532
 533         if (!oif && ipv6_addr_any(saddr)) {
 534                 if (unlikely(f6i->nh)) {
 535                         nh = nexthop_fib6_nh(f6i->nh);
 536                         if (nexthop_is_blackhole(f6i->nh))
 537                                 goto out_blackhole;
 538                 } else {
 539                         nh = f6i->fib6_nh;
 540                 }
 541                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 542                         goto out;
 543         }
 544
 545         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 546                 bool matched = false;
 547
 548                 if (unlikely(spf6i->nh)) {
 549                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 550                                               oif, flags);
 551                         if (nh)
 552                                 matched = true;
 553                 } else {
 554                         nh = spf6i->fib6_nh;
 555                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 556                                 matched = true;
 557                 }
 558                 if (matched) {
 559                         res->f6i = spf6i;
 560                         goto out;
 561                 }
 562         }
 563
 564         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 565                 res->f6i = net->ipv6.fib6_null_entry;
 566                 nh = res->f6i->fib6_nh;
 567                 goto out;
 568         }
 569
 570         if (unlikely(f6i->nh)) {
 571                 nh = nexthop_fib6_nh(f6i->nh);
 572                 if (nexthop_is_blackhole(f6i->nh))
 573                         goto out_blackhole;
 574         } else {
 575                 nh = f6i->fib6_nh;
 576         }
 577
 578         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 579                 res->f6i = net->ipv6.fib6_null_entry;
 580                 nh = res->f6i->fib6_nh;
 581         }
 582 out:
 583         res->nh = nh;
 584         res->fib6_type = res->f6i->fib6_type;
 585         res->fib6_flags = res->f6i->fib6_flags;
 586         return;
 587
 588 out_blackhole:
 589         res->fib6_flags |= RTF_REJECT;
 590         res->fib6_type = RTN_BLACKHOLE;
 591         res->nh = nh;
 592 }
 593
 594 #ifdef CONFIG_IPV6_ROUTER_PREF
 595 struct __rt6_probe_work {
 596         struct work_struct work;
 597         struct in6_addr target;
 598         struct net_device *dev;
 599         netdevice_tracker dev_tracker;
 600 };
 601
 602 static void rt6_probe_deferred(struct work_struct *w)
 603 {
 604         struct in6_addr mcaddr;
 605         struct __rt6_probe_work *work =
 606                 container_of(w, struct __rt6_probe_work, work);
 607
 608         addrconf_addr_solict_mult(&work->target, &mcaddr);
 609         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 610         dev_put_track(work->dev, &work->dev_tracker);
 611         kfree(work);
 612 }
 613
 614 static void rt6_probe(struct fib6_nh *fib6_nh)
 615 {
 616         struct __rt6_probe_work *work = NULL;
 617         const struct in6_addr *nh_gw;
 618         unsigned long last_probe;
 619         struct neighbour *neigh;
 620         struct net_device *dev;
 621         struct inet6_dev *idev;
 622
 623         /*
 624          * Okay, this does not seem to be appropriate
 625          * for now, however, we need to check if it
 626          * is really so; aka Router Reachability Probing.
 627          *
 628          * Router Reachability Probe MUST be rate-limited
 629          * to no more than one per minute.
 630          */
 631         if (!fib6_nh->fib_nh_gw_family)
 632                 return;
 633
 634         nh_gw = &fib6_nh->fib_nh_gw6;
 635         dev = fib6_nh->fib_nh_dev;
 636         rcu_read_lock_bh();
 637         last_probe = READ_ONCE(fib6_nh->last_probe);
 638         idev = __in6_dev_get(dev);
 639         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 640         if (neigh) {
 641                 if (neigh->nud_state & NUD_VALID)
 642                         goto out;
 643
 644                 write_lock(&neigh->lock);
 645                 if (!(neigh->nud_state & NUD_VALID) &&
 646                     time_after(jiffies,
 647                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 648                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 649                         if (work)
 650                                 __neigh_set_probe_once(neigh);
 651                 }
 652                 write_unlock(&neigh->lock);
 653         } else if (time_after(jiffies, last_probe +
 654                                        idev->cnf.rtr_probe_interval)) {
 655                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 656         }
 657
 658         if (!work || cmpxchg(&fib6_nh->last_probe,
 659                              last_probe, jiffies) != last_probe) {
 660                 kfree(work);
 661         } else {
 662                 INIT_WORK(&work->work, rt6_probe_deferred);
 663                 work->target = *nh_gw;
 664                 dev_hold_track(dev, &work->dev_tracker, GFP_ATOMIC);
 665                 work->dev = dev;
 666                 schedule_work(&work->work);
 667         }
 668
 669 out:
 670         rcu_read_unlock_bh();
 671 }
 672 #else
 673 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 674 {
 675 }
 676 #endif
 677
 678 /*
 679  * Default Router Selection (RFC 2461 6.3.6)
 680  */
 681 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 682 {
 683         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 684         struct neighbour *neigh;
 685
 686         rcu_read_lock_bh();
 687         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 688                                           &fib6_nh->fib_nh_gw6);
 689         if (neigh) {
 690                 read_lock(&neigh->lock);
 691                 if (neigh->nud_state & NUD_VALID)
 692                         ret = RT6_NUD_SUCCEED;
 693 #ifdef CONFIG_IPV6_ROUTER_PREF
 694                 else if (!(neigh->nud_state & NUD_FAILED))
 695                         ret = RT6_NUD_SUCCEED;
 696                 else
 697                         ret = RT6_NUD_FAIL_PROBE;
 698 #endif
 699                 read_unlock(&neigh->lock);
 700         } else {
 701                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 702                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 703         }
 704         rcu_read_unlock_bh();
 705
 706         return ret;
 707 }
 708
 709 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 710                            int strict)
 711 {
 712         int m = 0;
 713
 714         if (!oif || nh->fib_nh_dev->ifindex == oif)
 715                 m = 2;
 716
 717         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 718                 return RT6_NUD_FAIL_HARD;
 719 #ifdef CONFIG_IPV6_ROUTER_PREF
 720         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 721 #endif
 722         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 723             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 724                 int n = rt6_check_neigh(nh);
 725                 if (n < 0)
 726                         return n;
 727         }
 728         return m;
 729 }
 730
 731 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 732                        int oif, int strict, int *mpri, bool *do_rr)
 733 {
 734         bool match_do_rr = false;
 735         bool rc = false;
 736         int m;
 737
 738         if (nh->fib_nh_flags & RTNH_F_DEAD)
 739                 goto out;
 740
 741         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 742             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 743             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 744                 goto out;
 745
 746         m = rt6_score_route(nh, fib6_flags, oif, strict);
 747         if (m == RT6_NUD_FAIL_DO_RR) {
 748                 match_do_rr = true;
 749                 m = 0; /* lowest valid score */
 750         } else if (m == RT6_NUD_FAIL_HARD) {
 751                 goto out;
 752         }
 753
 754         if (strict & RT6_LOOKUP_F_REACHABLE)
 755                 rt6_probe(nh);
 756
 757         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 758         if (m > *mpri) {
 759                 *do_rr = match_do_rr;
 760                 *mpri = m;
 761                 rc = true;
 762         }
 763 out:
 764         return rc;
 765 }
 766
 767 struct fib6_nh_frl_arg {
 768         u32             flags;
 769         int             oif;
 770         int             strict;
 771         int             *mpri;
 772         bool            *do_rr;
 773         struct fib6_nh  *nh;
 774 };
 775
 776 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 777 {
 778         struct fib6_nh_frl_arg *arg = _arg;
 779
 780         arg->nh = nh;
 781         return find_match(nh, arg->flags, arg->oif, arg->strict,
 782                           arg->mpri, arg->do_rr);
 783 }
 784
 785 static void __find_rr_leaf(struct fib6_info *f6i_start,
 786                            struct fib6_info *nomatch, u32 metric,
 787                            struct fib6_result *res, struct fib6_info **cont,
 788                            int oif, int strict, bool *do_rr, int *mpri)
 789 {
 790         struct fib6_info *f6i;
 791
 792         for (f6i = f6i_start;
 793              f6i && f6i != nomatch;
 794              f6i = rcu_dereference(f6i->fib6_next)) {
 795                 bool matched = false;
 796                 struct fib6_nh *nh;
 797
 798                 if (cont && f6i->fib6_metric != metric) {
 799                         *cont = f6i;
 800                         return;
 801                 }
 802
 803                 if (fib6_check_expired(f6i))
 804                         continue;
 805
 806                 if (unlikely(f6i->nh)) {
 807                         struct fib6_nh_frl_arg arg = {
 808                                 .flags  = f6i->fib6_flags,
 809                                 .oif    = oif,
 810                                 .strict = strict,
 811                                 .mpri   = mpri,
 812                                 .do_rr  = do_rr
 813                         };
 814
 815                         if (nexthop_is_blackhole(f6i->nh)) {
 816                                 res->fib6_flags = RTF_REJECT;
 817                                 res->fib6_type = RTN_BLACKHOLE;
 818                                 res->f6i = f6i;
 819                                 res->nh = nexthop_fib6_nh(f6i->nh);
 820                                 return;
 821                         }
 822                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 823                                                      &arg)) {
 824                                 matched = true;
 825                                 nh = arg.nh;
 826                         }
 827                 } else {
 828                         nh = f6i->fib6_nh;
 829                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 830                                        mpri, do_rr))
 831                                 matched = true;
 832                 }
 833                 if (matched) {
 834                         res->f6i = f6i;
 835                         res->nh = nh;
 836                         res->fib6_flags = f6i->fib6_flags;
 837                         res->fib6_type = f6i->fib6_type;
 838                 }
 839         }
 840 }
 841
 842 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 843                          struct fib6_info *rr_head, int oif, int strict,
 844                          bool *do_rr, struct fib6_result *res)
 845 {
 846         u32 metric = rr_head->fib6_metric;
 847         struct fib6_info *cont = NULL;
 848         int mpri = -1;
 849
 850         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 851                        oif, strict, do_rr, &mpri);
 852
 853         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 854                        oif, strict, do_rr, &mpri);
 855
 856         if (res->f6i || !cont)
 857                 return;
 858
 859         __find_rr_leaf(cont, NULL, metric, res, NULL,
 860                        oif, strict, do_rr, &mpri);
 861 }
 862
 863 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 864                        struct fib6_result *res, int strict)
 865 {
 866         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 867         struct fib6_info *rt0;
 868         bool do_rr = false;
 869         int key_plen;
 870
 871         /* make sure this function or its helpers sets f6i */
 872         res->f6i = NULL;
 873
 874         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 875                 goto out;
 876
 877         rt0 = rcu_dereference(fn->rr_ptr);
 878         if (!rt0)
 879                 rt0 = leaf;
 880
 881         /* Double check to make sure fn is not an intermediate node
 882          * and fn->leaf does not points to its child's leaf
 883          * (This might happen if all routes under fn are deleted from
 884          * the tree and fib6_repair_tree() is called on the node.)
 885          */
 886         key_plen = rt0->fib6_dst.plen;
 887 #ifdef CONFIG_IPV6_SUBTREES
 888         if (rt0->fib6_src.plen)
 889                 key_plen = rt0->fib6_src.plen;
 890 #endif
 891         if (fn->fn_bit != key_plen)
 892                 goto out;
 893
 894         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 895         if (do_rr) {
 896                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 897
 898                 /* no entries matched; do round-robin */
 899                 if (!next || next->fib6_metric != rt0->fib6_metric)
 900                         next = leaf;
 901
 902                 if (next != rt0) {
 903                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 904                         /* make sure next is not being deleted from the tree */
 905                         if (next->fib6_node)
 906                                 rcu_assign_pointer(fn->rr_ptr, next);
 907                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 908                 }
 909         }
 910
 911 out:
 912         if (!res->f6i) {
 913                 res->f6i = net->ipv6.fib6_null_entry;
 914                 res->nh = res->f6i->fib6_nh;
 915                 res->fib6_flags = res->f6i->fib6_flags;
 916                 res->fib6_type = res->f6i->fib6_type;
 917         }
 918 }
 919
 920 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 921 {
 922         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 923                res->nh->fib_nh_gw_family;
 924 }
 925
 926 #ifdef CONFIG_IPV6_ROUTE_INFO
 927 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 928                   const struct in6_addr *gwaddr)
 929 {
 930         struct net *net = dev_net(dev);
 931         struct route_info *rinfo = (struct route_info *) opt;
 932         struct in6_addr prefix_buf, *prefix;
 933         unsigned int pref;
 934         unsigned long lifetime;
 935         struct fib6_info *rt;
 936
 937         if (len < sizeof(struct route_info)) {
 938                 return -EINVAL;
 939         }
 940
 941         /* Sanity check for prefix_len and length */
 942         if (rinfo->length > 3) {
 943                 return -EINVAL;
 944         } else if (rinfo->prefix_len > 128) {
 945                 return -EINVAL;
 946         } else if (rinfo->prefix_len > 64) {
 947                 if (rinfo->length < 2) {
 948                         return -EINVAL;
 949                 }
 950         } else if (rinfo->prefix_len > 0) {
 951                 if (rinfo->length < 1) {
 952                         return -EINVAL;
 953                 }
 954         }
 955
 956         pref = rinfo->route_pref;
 957         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 958                 return -EINVAL;
 959
 960         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 961
 962         if (rinfo->length == 3)
 963                 prefix = (struct in6_addr *)rinfo->prefix;
 964         else {
 965                 /* this function is safe */
 966                 ipv6_addr_prefix(&prefix_buf,
 967                                  (struct in6_addr *)rinfo->prefix,
 968                                  rinfo->prefix_len);
 969                 prefix = &prefix_buf;
 970         }
 971
 972         if (rinfo->prefix_len == 0)
 973                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 974         else
 975                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 976                                         gwaddr, dev);
 977
 978         if (rt && !lifetime) {
 979                 ip6_del_rt(net, rt, false);
 980                 rt = NULL;
 981         }
 982
 983         if (!rt && lifetime)
 984                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 985                                         dev, pref);
 986         else if (rt)
 987                 rt->fib6_flags = RTF_ROUTEINFO |
 988                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 989
 990         if (rt) {
 991                 if (!addrconf_finite_timeout(lifetime))
 992                         fib6_clean_expires(rt);
 993                 else
 994                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 995
 996                 fib6_info_release(rt);
 997         }
 998         return 0;
 999 }
1000 #endif
1001
1002 /*
1003  *      Misc support functions
1004  */
1005
1006 /* called with rcu_lock held */
1007 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1008 {
1009         struct net_device *dev = res->nh->fib_nh_dev;
1010
1011         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1012                 /* for copies of local routes, dst->dev needs to be the
1013                  * device if it is a master device, the master device if
1014                  * device is enslaved, and the loopback as the default
1015                  */
1016                 if (netif_is_l3_slave(dev) &&
1017                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1018                         dev = l3mdev_master_dev_rcu(dev);
1019                 else if (!netif_is_l3_master(dev))
1020                         dev = dev_net(dev)->loopback_dev;
1021                 /* last case is netif_is_l3_master(dev) is true in which
1022                  * case we want dev returned to be dev
1023                  */
1024         }
1025
1026         return dev;
1027 }
1028
1029 static const int fib6_prop[RTN_MAX + 1] = {
1030         [RTN_UNSPEC]    = 0,
1031         [RTN_UNICAST]   = 0,
1032         [RTN_LOCAL]     = 0,
1033         [RTN_BROADCAST] = 0,
1034         [RTN_ANYCAST]   = 0,
1035         [RTN_MULTICAST] = 0,
1036         [RTN_BLACKHOLE] = -EINVAL,
1037         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1038         [RTN_PROHIBIT]  = -EACCES,
1039         [RTN_THROW]     = -EAGAIN,
1040         [RTN_NAT]       = -EINVAL,
1041         [RTN_XRESOLVE]  = -EINVAL,
1042 };
1043
1044 static int ip6_rt_type_to_error(u8 fib6_type)
1045 {
1046         return fib6_prop[fib6_type];
1047 }
1048
1049 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1050 {
1051         unsigned short flags = 0;
1052
1053         if (rt->dst_nocount)
1054                 flags |= DST_NOCOUNT;
1055         if (rt->dst_nopolicy)
1056                 flags |= DST_NOPOLICY;
1057
1058         return flags;
1059 }
1060
1061 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1062 {
1063         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1064
1065         switch (fib6_type) {
1066         case RTN_BLACKHOLE:
1067                 rt->dst.output = dst_discard_out;
1068                 rt->dst.input = dst_discard;
1069                 break;
1070         case RTN_PROHIBIT:
1071                 rt->dst.output = ip6_pkt_prohibit_out;
1072                 rt->dst.input = ip6_pkt_prohibit;
1073                 break;
1074         case RTN_THROW:
1075         case RTN_UNREACHABLE:
1076         default:
1077                 rt->dst.output = ip6_pkt_discard_out;
1078                 rt->dst.input = ip6_pkt_discard;
1079                 break;
1080         }
1081 }
1082
1083 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1084 {
1085         struct fib6_info *f6i = res->f6i;
1086
1087         if (res->fib6_flags & RTF_REJECT) {
1088                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1089                 return;
1090         }
1091
1092         rt->dst.error = 0;
1093         rt->dst.output = ip6_output;
1094
1095         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1096                 rt->dst.input = ip6_input;
1097         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1098                 rt->dst.input = ip6_mc_input;
1099         } else {
1100                 rt->dst.input = ip6_forward;
1101         }
1102
1103         if (res->nh->fib_nh_lws) {
1104                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1105                 lwtunnel_set_redirect(&rt->dst);
1106         }
1107
1108         rt->dst.lastuse = jiffies;
1109 }
1110
1111 /* Caller must already hold reference to @from */
1112 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1113 {
1114         rt->rt6i_flags &= ~RTF_EXPIRES;
1115         rcu_assign_pointer(rt->from, from);
1116         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1117 }
1118
1119 /* Caller must already hold reference to f6i in result */
1120 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1121 {
1122         const struct fib6_nh *nh = res->nh;
1123         const struct net_device *dev = nh->fib_nh_dev;
1124         struct fib6_info *f6i = res->f6i;
1125
1126         ip6_rt_init_dst(rt, res);
1127
1128         rt->rt6i_dst = f6i->fib6_dst;
1129         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1130         rt->rt6i_flags = res->fib6_flags;
1131         if (nh->fib_nh_gw_family) {
1132                 rt->rt6i_gateway = nh->fib_nh_gw6;
1133                 rt->rt6i_flags |= RTF_GATEWAY;
1134         }
1135         rt6_set_from(rt, f6i);
1136 #ifdef CONFIG_IPV6_SUBTREES
1137         rt->rt6i_src = f6i->fib6_src;
1138 #endif
1139 }
1140
1141 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1142                                         struct in6_addr *saddr)
1143 {
1144         struct fib6_node *pn, *sn;
1145         while (1) {
1146                 if (fn->fn_flags & RTN_TL_ROOT)
1147                         return NULL;
1148                 pn = rcu_dereference(fn->parent);
1149                 sn = FIB6_SUBTREE(pn);
1150                 if (sn && sn != fn)
1151                         fn = fib6_node_lookup(sn, NULL, saddr);
1152                 else
1153                         fn = pn;
1154                 if (fn->fn_flags & RTN_RTINFO)
1155                         return fn;
1156         }
1157 }
1158
1159 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1160 {
1161         struct rt6_info *rt = *prt;
1162
1163         if (dst_hold_safe(&rt->dst))
1164                 return true;
1165         if (net) {
1166                 rt = net->ipv6.ip6_null_entry;
1167                 dst_hold(&rt->dst);
1168         } else {
1169                 rt = NULL;
1170         }
1171         *prt = rt;
1172         return false;
1173 }
1174
1175 /* called with rcu_lock held */
1176 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1177 {
1178         struct net_device *dev = res->nh->fib_nh_dev;
1179         struct fib6_info *f6i = res->f6i;
1180         unsigned short flags;
1181         struct rt6_info *nrt;
1182
1183         if (!fib6_info_hold_safe(f6i))
1184                 goto fallback;
1185
1186         flags = fib6_info_dst_flags(f6i);
1187         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1188         if (!nrt) {
1189                 fib6_info_release(f6i);
1190                 goto fallback;
1191         }
1192
1193         ip6_rt_copy_init(nrt, res);
1194         return nrt;
1195
1196 fallback:
1197         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1198         dst_hold(&nrt->dst);
1199         return nrt;
1200 }
1201
1202 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1203                                              struct fib6_table *table,
1204                                              struct flowi6 *fl6,
1205                                              const struct sk_buff *skb,
1206                                              int flags)
1207 {
1208         struct fib6_result res = {};
1209         struct fib6_node *fn;
1210         struct rt6_info *rt;
1211
1212         rcu_read_lock();
1213         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1214 restart:
1215         res.f6i = rcu_dereference(fn->leaf);
1216         if (!res.f6i)
1217                 res.f6i = net->ipv6.fib6_null_entry;
1218         else
1219                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1220                                  flags);
1221
1222         if (res.f6i == net->ipv6.fib6_null_entry) {
1223                 fn = fib6_backtrack(fn, &fl6->saddr);
1224                 if (fn)
1225                         goto restart;
1226
1227                 rt = net->ipv6.ip6_null_entry;
1228                 dst_hold(&rt->dst);
1229                 goto out;
1230         } else if (res.fib6_flags & RTF_REJECT) {
1231                 goto do_create;
1232         }
1233
1234         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1235                          fl6->flowi6_oif != 0, skb, flags);
1236
1237         /* Search through exception table */
1238         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1239         if (rt) {
1240                 if (ip6_hold_safe(net, &rt))
1241                         dst_use_noref(&rt->dst, jiffies);
1242         } else {
1243 do_create:
1244                 rt = ip6_create_rt_rcu(&res);
1245         }
1246
1247 out:
1248         trace_fib6_table_lookup(net, &res, table, fl6);
1249
1250         rcu_read_unlock();
1251
1252         return rt;
1253 }
1254
1255 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1256                                    const struct sk_buff *skb, int flags)
1257 {
1258         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1259 }
1260 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1261
1262 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1263                             const struct in6_addr *saddr, int oif,
1264                             const struct sk_buff *skb, int strict)
1265 {
1266         struct flowi6 fl6 = {
1267                 .flowi6_oif = oif,
1268                 .daddr = *daddr,
1269         };
1270         struct dst_entry *dst;
1271         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1272
1273         if (saddr) {
1274                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1275                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1276         }
1277
1278         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1279         if (dst->error == 0)
1280                 return (struct rt6_info *) dst;
1281
1282         dst_release(dst);
1283
1284         return NULL;
1285 }
1286 EXPORT_SYMBOL(rt6_lookup);
1287
1288 /* ip6_ins_rt is called with FREE table->tb6_lock.
1289  * It takes new route entry, the addition fails by any reason the
1290  * route is released.
1291  * Caller must hold dst before calling it.
1292  */
1293
1294 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1295                         struct netlink_ext_ack *extack)
1296 {
1297         int err;
1298         struct fib6_table *table;
1299
1300         table = rt->fib6_table;
1301         spin_lock_bh(&table->tb6_lock);
1302         err = fib6_add(&table->tb6_root, rt, info, extack);
1303         spin_unlock_bh(&table->tb6_lock);
1304
1305         return err;
1306 }
1307
1308 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1309 {
1310         struct nl_info info = { .nl_net = net, };
1311
1312         return __ip6_ins_rt(rt, &info, NULL);
1313 }
1314
1315 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1316                                            const struct in6_addr *daddr,
1317                                            const struct in6_addr *saddr)
1318 {
1319         struct fib6_info *f6i = res->f6i;
1320         struct net_device *dev;
1321         struct rt6_info *rt;
1322
1323         /*
1324          *      Clone the route.
1325          */
1326
1327         if (!fib6_info_hold_safe(f6i))
1328                 return NULL;
1329
1330         dev = ip6_rt_get_dev_rcu(res);
1331         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1332         if (!rt) {
1333                 fib6_info_release(f6i);
1334                 return NULL;
1335         }
1336
1337         ip6_rt_copy_init(rt, res);
1338         rt->rt6i_flags |= RTF_CACHE;
1339         rt->rt6i_dst.addr = *daddr;
1340         rt->rt6i_dst.plen = 128;
1341
1342         if (!rt6_is_gw_or_nonexthop(res)) {
1343                 if (f6i->fib6_dst.plen != 128 &&
1344                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1345                         rt->rt6i_flags |= RTF_ANYCAST;
1346 #ifdef CONFIG_IPV6_SUBTREES
1347                 if (rt->rt6i_src.plen && saddr) {
1348                         rt->rt6i_src.addr = *saddr;
1349                         rt->rt6i_src.plen = 128;
1350                 }
1351 #endif
1352         }
1353
1354         return rt;
1355 }
1356
1357 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1358 {
1359         struct fib6_info *f6i = res->f6i;
1360         unsigned short flags = fib6_info_dst_flags(f6i);
1361         struct net_device *dev;
1362         struct rt6_info *pcpu_rt;
1363
1364         if (!fib6_info_hold_safe(f6i))
1365                 return NULL;
1366
1367         rcu_read_lock();
1368         dev = ip6_rt_get_dev_rcu(res);
1369         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1370         rcu_read_unlock();
1371         if (!pcpu_rt) {
1372                 fib6_info_release(f6i);
1373                 return NULL;
1374         }
1375         ip6_rt_copy_init(pcpu_rt, res);
1376         pcpu_rt->rt6i_flags |= RTF_PCPU;
1377
1378         if (f6i->nh)
1379                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1380
1381         return pcpu_rt;
1382 }
1383
1384 static bool rt6_is_valid(const struct rt6_info *rt6)
1385 {
1386         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1387 }
1388
1389 /* It should be called with rcu_read_lock() acquired */
1390 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1391 {
1392         struct rt6_info *pcpu_rt;
1393
1394         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1395
1396         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1397                 struct rt6_info *prev, **p;
1398
1399                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1400                 prev = xchg(p, NULL);
1401                 if (prev) {
1402                         dst_dev_put(&prev->dst);
1403                         dst_release(&prev->dst);
1404                 }
1405
1406                 pcpu_rt = NULL;
1407         }
1408
1409         return pcpu_rt;
1410 }
1411
1412 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1413                                             const struct fib6_result *res)
1414 {
1415         struct rt6_info *pcpu_rt, *prev, **p;
1416
1417         pcpu_rt = ip6_rt_pcpu_alloc(res);
1418         if (!pcpu_rt)
1419                 return NULL;
1420
1421         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1422         prev = cmpxchg(p, NULL, pcpu_rt);
1423         BUG_ON(prev);
1424
1425         if (res->f6i->fib6_destroying) {
1426                 struct fib6_info *from;
1427
1428                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1429                 fib6_info_release(from);
1430         }
1431
1432         return pcpu_rt;
1433 }
1434
1435 /* exception hash table implementation
1436  */
1437 static DEFINE_SPINLOCK(rt6_exception_lock);
1438
1439 /* Remove rt6_ex from hash table and free the memory
1440  * Caller must hold rt6_exception_lock
1441  */
1442 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1443                                  struct rt6_exception *rt6_ex)
1444 {
1445         struct fib6_info *from;
1446         struct net *net;
1447
1448         if (!bucket || !rt6_ex)
1449                 return;
1450
1451         net = dev_net(rt6_ex->rt6i->dst.dev);
1452         net->ipv6.rt6_stats->fib_rt_cache--;
1453
1454         /* purge completely the exception to allow releasing the held resources:
1455          * some [sk] cache may keep the dst around for unlimited time
1456          */
1457         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1458         fib6_info_release(from);
1459         dst_dev_put(&rt6_ex->rt6i->dst);
1460
1461         hlist_del_rcu(&rt6_ex->hlist);
1462         dst_release(&rt6_ex->rt6i->dst);
1463         kfree_rcu(rt6_ex, rcu);
1464         WARN_ON_ONCE(!bucket->depth);
1465         bucket->depth--;
1466 }
1467
1468 /* Remove oldest rt6_ex in bucket and free the memory
1469  * Caller must hold rt6_exception_lock
1470  */
1471 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1472 {
1473         struct rt6_exception *rt6_ex, *oldest = NULL;
1474
1475         if (!bucket)
1476                 return;
1477
1478         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1479                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1480                         oldest = rt6_ex;
1481         }
1482         rt6_remove_exception(bucket, oldest);
1483 }
1484
1485 static u32 rt6_exception_hash(const struct in6_addr *dst,
1486                               const struct in6_addr *src)
1487 {
1488         static siphash_aligned_key_t rt6_exception_key;
1489         struct {
1490                 struct in6_addr dst;
1491                 struct in6_addr src;
1492         } __aligned(SIPHASH_ALIGNMENT) combined = {
1493                 .dst = *dst,
1494         };
1495         u64 val;
1496
1497         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1498
1499 #ifdef CONFIG_IPV6_SUBTREES
1500         if (src)
1501                 combined.src = *src;
1502 #endif
1503         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1504
1505         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1506 }
1507
1508 /* Helper function to find the cached rt in the hash table
1509  * and update bucket pointer to point to the bucket for this
1510  * (daddr, saddr) pair
1511  * Caller must hold rt6_exception_lock
1512  */
1513 static struct rt6_exception *
1514 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1515                               const struct in6_addr *daddr,
1516                               const struct in6_addr *saddr)
1517 {
1518         struct rt6_exception *rt6_ex;
1519         u32 hval;
1520
1521         if (!(*bucket) || !daddr)
1522                 return NULL;
1523
1524         hval = rt6_exception_hash(daddr, saddr);
1525         *bucket += hval;
1526
1527         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1528                 struct rt6_info *rt6 = rt6_ex->rt6i;
1529                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1530
1531 #ifdef CONFIG_IPV6_SUBTREES
1532                 if (matched && saddr)
1533                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1534 #endif
1535                 if (matched)
1536                         return rt6_ex;
1537         }
1538         return NULL;
1539 }
1540
1541 /* Helper function to find the cached rt in the hash table
1542  * and update bucket pointer to point to the bucket for this
1543  * (daddr, saddr) pair
1544  * Caller must hold rcu_read_lock()
1545  */
1546 static struct rt6_exception *
1547 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1548                          const struct in6_addr *daddr,
1549                          const struct in6_addr *saddr)
1550 {
1551         struct rt6_exception *rt6_ex;
1552         u32 hval;
1553
1554         WARN_ON_ONCE(!rcu_read_lock_held());
1555
1556         if (!(*bucket) || !daddr)
1557                 return NULL;
1558
1559         hval = rt6_exception_hash(daddr, saddr);
1560         *bucket += hval;
1561
1562         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1563                 struct rt6_info *rt6 = rt6_ex->rt6i;
1564                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1565
1566 #ifdef CONFIG_IPV6_SUBTREES
1567                 if (matched && saddr)
1568                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1569 #endif
1570                 if (matched)
1571                         return rt6_ex;
1572         }
1573         return NULL;
1574 }
1575
1576 static unsigned int fib6_mtu(const struct fib6_result *res)
1577 {
1578         const struct fib6_nh *nh = res->nh;
1579         unsigned int mtu;
1580
1581         if (res->f6i->fib6_pmtu) {
1582                 mtu = res->f6i->fib6_pmtu;
1583         } else {
1584                 struct net_device *dev = nh->fib_nh_dev;
1585                 struct inet6_dev *idev;
1586
1587                 rcu_read_lock();
1588                 idev = __in6_dev_get(dev);
1589                 mtu = idev->cnf.mtu6;
1590                 rcu_read_unlock();
1591         }
1592
1593         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1594
1595         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1596 }
1597
1598 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1599
1600 /* used when the flushed bit is not relevant, only access to the bucket
1601  * (ie., all bucket users except rt6_insert_exception);
1602  *
1603  * called under rcu lock; sometimes called with rt6_exception_lock held
1604  */
1605 static
1606 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1607                                                        spinlock_t *lock)
1608 {
1609         struct rt6_exception_bucket *bucket;
1610
1611         if (lock)
1612                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1613                                                    lockdep_is_held(lock));
1614         else
1615                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1616
1617         /* remove bucket flushed bit if set */
1618         if (bucket) {
1619                 unsigned long p = (unsigned long)bucket;
1620
1621                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1622                 bucket = (struct rt6_exception_bucket *)p;
1623         }
1624
1625         return bucket;
1626 }
1627
1628 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1629 {
1630         unsigned long p = (unsigned long)bucket;
1631
1632         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1633 }
1634
1635 /* called with rt6_exception_lock held */
1636 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1637                                               spinlock_t *lock)
1638 {
1639         struct rt6_exception_bucket *bucket;
1640         unsigned long p;
1641
1642         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1643                                            lockdep_is_held(lock));
1644
1645         p = (unsigned long)bucket;
1646         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1647         bucket = (struct rt6_exception_bucket *)p;
1648         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1649 }
1650
1651 static int rt6_insert_exception(struct rt6_info *nrt,
1652                                 const struct fib6_result *res)
1653 {
1654         struct net *net = dev_net(nrt->dst.dev);
1655         struct rt6_exception_bucket *bucket;
1656         struct fib6_info *f6i = res->f6i;
1657         struct in6_addr *src_key = NULL;
1658         struct rt6_exception *rt6_ex;
1659         struct fib6_nh *nh = res->nh;
1660         int max_depth;
1661         int err = 0;
1662
1663         spin_lock_bh(&rt6_exception_lock);
1664
1665         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1666                                           lockdep_is_held(&rt6_exception_lock));
1667         if (!bucket) {
1668                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1669                                  GFP_ATOMIC);
1670                 if (!bucket) {
1671                         err = -ENOMEM;
1672                         goto out;
1673                 }
1674                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1675         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1676                 err = -EINVAL;
1677                 goto out;
1678         }
1679
1680 #ifdef CONFIG_IPV6_SUBTREES
1681         /* fib6_src.plen != 0 indicates f6i is in subtree
1682          * and exception table is indexed by a hash of
1683          * both fib6_dst and fib6_src.
1684          * Otherwise, the exception table is indexed by
1685          * a hash of only fib6_dst.
1686          */
1687         if (f6i->fib6_src.plen)
1688                 src_key = &nrt->rt6i_src.addr;
1689 #endif
1690         /* rt6_mtu_change() might lower mtu on f6i.
1691          * Only insert this exception route if its mtu
1692          * is less than f6i's mtu value.
1693          */
1694         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1695                 err = -EINVAL;
1696                 goto out;
1697         }
1698
1699         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1700                                                src_key);
1701         if (rt6_ex)
1702                 rt6_remove_exception(bucket, rt6_ex);
1703
1704         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1705         if (!rt6_ex) {
1706                 err = -ENOMEM;
1707                 goto out;
1708         }
1709         rt6_ex->rt6i = nrt;
1710         rt6_ex->stamp = jiffies;
1711         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1712         bucket->depth++;
1713         net->ipv6.rt6_stats->fib_rt_cache++;
1714
1715         /* Randomize max depth to avoid some side channels attacks. */
1716         max_depth = FIB6_MAX_DEPTH + prandom_u32_max(FIB6_MAX_DEPTH);
1717         while (bucket->depth > max_depth)
1718                 rt6_exception_remove_oldest(bucket);
1719
1720 out:
1721         spin_unlock_bh(&rt6_exception_lock);
1722
1723         /* Update fn->fn_sernum to invalidate all cached dst */
1724         if (!err) {
1725                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1726                 fib6_update_sernum(net, f6i);
1727                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1728                 fib6_force_start_gc(net);
1729         }
1730
1731         return err;
1732 }
1733
1734 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1735 {
1736         struct rt6_exception_bucket *bucket;
1737         struct rt6_exception *rt6_ex;
1738         struct hlist_node *tmp;
1739         int i;
1740
1741         spin_lock_bh(&rt6_exception_lock);
1742
1743         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1744         if (!bucket)
1745                 goto out;
1746
1747         /* Prevent rt6_insert_exception() to recreate the bucket list */
1748         if (!from)
1749                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1750
1751         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1752                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1753                         if (!from ||
1754                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1755                                 rt6_remove_exception(bucket, rt6_ex);
1756                 }
1757                 WARN_ON_ONCE(!from && bucket->depth);
1758                 bucket++;
1759         }
1760 out:
1761         spin_unlock_bh(&rt6_exception_lock);
1762 }
1763
1764 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1765 {
1766         struct fib6_info *f6i = arg;
1767
1768         fib6_nh_flush_exceptions(nh, f6i);
1769
1770         return 0;
1771 }
1772
1773 void rt6_flush_exceptions(struct fib6_info *f6i)
1774 {
1775         if (f6i->nh)
1776                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1777                                          f6i);
1778         else
1779                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1780 }
1781
1782 /* Find cached rt in the hash table inside passed in rt
1783  * Caller has to hold rcu_read_lock()
1784  */
1785 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1786                                            const struct in6_addr *daddr,
1787                                            const struct in6_addr *saddr)
1788 {
1789         const struct in6_addr *src_key = NULL;
1790         struct rt6_exception_bucket *bucket;
1791         struct rt6_exception *rt6_ex;
1792         struct rt6_info *ret = NULL;
1793
1794 #ifdef CONFIG_IPV6_SUBTREES
1795         /* fib6i_src.plen != 0 indicates f6i is in subtree
1796          * and exception table is indexed by a hash of
1797          * both fib6_dst and fib6_src.
1798          * However, the src addr used to create the hash
1799          * might not be exactly the passed in saddr which
1800          * is a /128 addr from the flow.
1801          * So we need to use f6i->fib6_src to redo lookup
1802          * if the passed in saddr does not find anything.
1803          * (See the logic in ip6_rt_cache_alloc() on how
1804          * rt->rt6i_src is updated.)
1805          */
1806         if (res->f6i->fib6_src.plen)
1807                 src_key = saddr;
1808 find_ex:
1809 #endif
1810         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1811         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1812
1813         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1814                 ret = rt6_ex->rt6i;
1815
1816 #ifdef CONFIG_IPV6_SUBTREES
1817         /* Use fib6_src as src_key and redo lookup */
1818         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1819                 src_key = &res->f6i->fib6_src.addr;
1820                 goto find_ex;
1821         }
1822 #endif
1823
1824         return ret;
1825 }
1826
1827 /* Remove the passed in cached rt from the hash table that contains it */
1828 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1829                                     const struct rt6_info *rt)
1830 {
1831         const struct in6_addr *src_key = NULL;
1832         struct rt6_exception_bucket *bucket;
1833         struct rt6_exception *rt6_ex;
1834         int err;
1835
1836         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1837                 return -ENOENT;
1838
1839         spin_lock_bh(&rt6_exception_lock);
1840         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1841
1842 #ifdef CONFIG_IPV6_SUBTREES
1843         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1844          * and exception table is indexed by a hash of
1845          * both rt6i_dst and rt6i_src.
1846          * Otherwise, the exception table is indexed by
1847          * a hash of only rt6i_dst.
1848          */
1849         if (plen)
1850                 src_key = &rt->rt6i_src.addr;
1851 #endif
1852         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1853                                                &rt->rt6i_dst.addr,
1854                                                src_key);
1855         if (rt6_ex) {
1856                 rt6_remove_exception(bucket, rt6_ex);
1857                 err = 0;
1858         } else {
1859                 err = -ENOENT;
1860         }
1861
1862         spin_unlock_bh(&rt6_exception_lock);
1863         return err;
1864 }
1865
1866 struct fib6_nh_excptn_arg {
1867         struct rt6_info *rt;
1868         int             plen;
1869 };
1870
1871 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1872 {
1873         struct fib6_nh_excptn_arg *arg = _arg;
1874         int err;
1875
1876         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1877         if (err == 0)
1878                 return 1;
1879
1880         return 0;
1881 }
1882
1883 static int rt6_remove_exception_rt(struct rt6_info *rt)
1884 {
1885         struct fib6_info *from;
1886
1887         from = rcu_dereference(rt->from);
1888         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1889                 return -EINVAL;
1890
1891         if (from->nh) {
1892                 struct fib6_nh_excptn_arg arg = {
1893                         .rt = rt,
1894                         .plen = from->fib6_src.plen
1895                 };
1896                 int rc;
1897
1898                 /* rc = 1 means an entry was found */
1899                 rc = nexthop_for_each_fib6_nh(from->nh,
1900                                               rt6_nh_remove_exception_rt,
1901                                               &arg);
1902                 return rc ? 0 : -ENOENT;
1903         }
1904
1905         return fib6_nh_remove_exception(from->fib6_nh,
1906                                         from->fib6_src.plen, rt);
1907 }
1908
1909 /* Find rt6_ex which contains the passed in rt cache and
1910  * refresh its stamp
1911  */
1912 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1913                                      const struct rt6_info *rt)
1914 {
1915         const struct in6_addr *src_key = NULL;
1916         struct rt6_exception_bucket *bucket;
1917         struct rt6_exception *rt6_ex;
1918
1919         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1920 #ifdef CONFIG_IPV6_SUBTREES
1921         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1922          * and exception table is indexed by a hash of
1923          * both rt6i_dst and rt6i_src.
1924          * Otherwise, the exception table is indexed by
1925          * a hash of only rt6i_dst.
1926          */
1927         if (plen)
1928                 src_key = &rt->rt6i_src.addr;
1929 #endif
1930         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1931         if (rt6_ex)
1932                 rt6_ex->stamp = jiffies;
1933 }
1934
1935 struct fib6_nh_match_arg {
1936         const struct net_device *dev;
1937         const struct in6_addr   *gw;
1938         struct fib6_nh          *match;
1939 };
1940
1941 /* determine if fib6_nh has given device and gateway */
1942 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1943 {
1944         struct fib6_nh_match_arg *arg = _arg;
1945
1946         if (arg->dev != nh->fib_nh_dev ||
1947             (arg->gw && !nh->fib_nh_gw_family) ||
1948             (!arg->gw && nh->fib_nh_gw_family) ||
1949             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1950                 return 0;
1951
1952         arg->match = nh;
1953
1954         /* found a match, break the loop */
1955         return 1;
1956 }
1957
1958 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1959 {
1960         struct fib6_info *from;
1961         struct fib6_nh *fib6_nh;
1962
1963         rcu_read_lock();
1964
1965         from = rcu_dereference(rt->from);
1966         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1967                 goto unlock;
1968
1969         if (from->nh) {
1970                 struct fib6_nh_match_arg arg = {
1971                         .dev = rt->dst.dev,
1972                         .gw = &rt->rt6i_gateway,
1973                 };
1974
1975                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1976
1977                 if (!arg.match)
1978                         goto unlock;
1979                 fib6_nh = arg.match;
1980         } else {
1981                 fib6_nh = from->fib6_nh;
1982         }
1983         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1984 unlock:
1985         rcu_read_unlock();
1986 }
1987
1988 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1989                                          struct rt6_info *rt, int mtu)
1990 {
1991         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1992          * lowest MTU in the path: always allow updating the route PMTU to
1993          * reflect PMTU decreases.
1994          *
1995          * If the new MTU is higher, and the route PMTU is equal to the local
1996          * MTU, this means the old MTU is the lowest in the path, so allow
1997          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1998          * handle this.
1999          */
2000
2001         if (dst_mtu(&rt->dst) >= mtu)
2002                 return true;
2003
2004         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2005                 return true;
2006
2007         return false;
2008 }
2009
2010 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2011                                        const struct fib6_nh *nh, int mtu)
2012 {
2013         struct rt6_exception_bucket *bucket;
2014         struct rt6_exception *rt6_ex;
2015         int i;
2016
2017         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2018         if (!bucket)
2019                 return;
2020
2021         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2022                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2023                         struct rt6_info *entry = rt6_ex->rt6i;
2024
2025                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2026                          * route), the metrics of its rt->from have already
2027                          * been updated.
2028                          */
2029                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2030                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2031                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2032                 }
2033                 bucket++;
2034         }
2035 }
2036
2037 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2038
2039 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2040                                             const struct in6_addr *gateway)
2041 {
2042         struct rt6_exception_bucket *bucket;
2043         struct rt6_exception *rt6_ex;
2044         struct hlist_node *tmp;
2045         int i;
2046
2047         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2048                 return;
2049
2050         spin_lock_bh(&rt6_exception_lock);
2051         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2052         if (bucket) {
2053                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2054                         hlist_for_each_entry_safe(rt6_ex, tmp,
2055                                                   &bucket->chain, hlist) {
2056                                 struct rt6_info *entry = rt6_ex->rt6i;
2057
2058                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2059                                     RTF_CACHE_GATEWAY &&
2060                                     ipv6_addr_equal(gateway,
2061                                                     &entry->rt6i_gateway)) {
2062                                         rt6_remove_exception(bucket, rt6_ex);
2063                                 }
2064                         }
2065                         bucket++;
2066                 }
2067         }
2068
2069         spin_unlock_bh(&rt6_exception_lock);
2070 }
2071
2072 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2073                                       struct rt6_exception *rt6_ex,
2074                                       struct fib6_gc_args *gc_args,
2075                                       unsigned long now)
2076 {
2077         struct rt6_info *rt = rt6_ex->rt6i;
2078
2079         /* we are pruning and obsoleting aged-out and non gateway exceptions
2080          * even if others have still references to them, so that on next
2081          * dst_check() such references can be dropped.
2082          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2083          * expired, independently from their aging, as per RFC 8201 section 4
2084          */
2085         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2086                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2087                         RT6_TRACE("aging clone %p\n", rt);
2088                         rt6_remove_exception(bucket, rt6_ex);
2089                         return;
2090                 }
2091         } else if (time_after(jiffies, rt->dst.expires)) {
2092                 RT6_TRACE("purging expired route %p\n", rt);
2093                 rt6_remove_exception(bucket, rt6_ex);
2094                 return;
2095         }
2096
2097         if (rt->rt6i_flags & RTF_GATEWAY) {
2098                 struct neighbour *neigh;
2099
2100                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2101
2102                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2103                         RT6_TRACE("purging route %p via non-router but gateway\n",
2104                                   rt);
2105                         rt6_remove_exception(bucket, rt6_ex);
2106                         return;
2107                 }
2108         }
2109
2110         gc_args->more++;
2111 }
2112
2113 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2114                                    struct fib6_gc_args *gc_args,
2115                                    unsigned long now)
2116 {
2117         struct rt6_exception_bucket *bucket;
2118         struct rt6_exception *rt6_ex;
2119         struct hlist_node *tmp;
2120         int i;
2121
2122         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2123                 return;
2124
2125         rcu_read_lock_bh();
2126         spin_lock(&rt6_exception_lock);
2127         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2128         if (bucket) {
2129                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2130                         hlist_for_each_entry_safe(rt6_ex, tmp,
2131                                                   &bucket->chain, hlist) {
2132                                 rt6_age_examine_exception(bucket, rt6_ex,
2133                                                           gc_args, now);
2134                         }
2135                         bucket++;
2136                 }
2137         }
2138         spin_unlock(&rt6_exception_lock);
2139         rcu_read_unlock_bh();
2140 }
2141
2142 struct fib6_nh_age_excptn_arg {
2143         struct fib6_gc_args     *gc_args;
2144         unsigned long           now;
2145 };
2146
2147 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2148 {
2149         struct fib6_nh_age_excptn_arg *arg = _arg;
2150
2151         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2152         return 0;
2153 }
2154
2155 void rt6_age_exceptions(struct fib6_info *f6i,
2156                         struct fib6_gc_args *gc_args,
2157                         unsigned long now)
2158 {
2159         if (f6i->nh) {
2160                 struct fib6_nh_age_excptn_arg arg = {
2161                         .gc_args = gc_args,
2162                         .now = now
2163                 };
2164
2165                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2166                                          &arg);
2167         } else {
2168                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2169         }
2170 }
2171
2172 /* must be called with rcu lock held */
2173 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2174                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2175 {
2176         struct fib6_node *fn, *saved_fn;
2177
2178         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2179         saved_fn = fn;
2180
2181 redo_rt6_select:
2182         rt6_select(net, fn, oif, res, strict);
2183         if (res->f6i == net->ipv6.fib6_null_entry) {
2184                 fn = fib6_backtrack(fn, &fl6->saddr);
2185                 if (fn)
2186                         goto redo_rt6_select;
2187                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2188                         /* also consider unreachable route */
2189                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2190                         fn = saved_fn;
2191                         goto redo_rt6_select;
2192                 }
2193         }
2194
2195         trace_fib6_table_lookup(net, res, table, fl6);
2196
2197         return 0;
2198 }
2199
2200 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2201                                int oif, struct flowi6 *fl6,
2202                                const struct sk_buff *skb, int flags)
2203 {
2204         struct fib6_result res = {};
2205         struct rt6_info *rt = NULL;
2206         int strict = 0;
2207
2208         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2209                      !rcu_read_lock_held());
2210
2211         strict |= flags & RT6_LOOKUP_F_IFACE;
2212         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2213         if (net->ipv6.devconf_all->forwarding == 0)
2214                 strict |= RT6_LOOKUP_F_REACHABLE;
2215
2216         rcu_read_lock();
2217
2218         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2219         if (res.f6i == net->ipv6.fib6_null_entry)
2220                 goto out;
2221
2222         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2223
2224         /*Search through exception table */
2225         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2226         if (rt) {
2227                 goto out;
2228         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2229                             !res.nh->fib_nh_gw_family)) {
2230                 /* Create a RTF_CACHE clone which will not be
2231                  * owned by the fib6 tree.  It is for the special case where
2232                  * the daddr in the skb during the neighbor look-up is different
2233                  * from the fl6->daddr used to look-up route here.
2234                  */
2235                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2236
2237                 if (rt) {
2238                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2239                          * As rt6_uncached_list_add() does not consume refcnt,
2240                          * this refcnt is always returned to the caller even
2241                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2242                          */
2243                         rt6_uncached_list_add(rt);
2244                         rcu_read_unlock();
2245
2246                         return rt;
2247                 }
2248         } else {
2249                 /* Get a percpu copy */
2250                 local_bh_disable();
2251                 rt = rt6_get_pcpu_route(&res);
2252
2253                 if (!rt)
2254                         rt = rt6_make_pcpu_route(net, &res);
2255
2256                 local_bh_enable();
2257         }
2258 out:
2259         if (!rt)
2260                 rt = net->ipv6.ip6_null_entry;
2261         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2262                 ip6_hold_safe(net, &rt);
2263         rcu_read_unlock();
2264
2265         return rt;
2266 }
2267 EXPORT_SYMBOL_GPL(ip6_pol_route);
2268
2269 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2270                                             struct fib6_table *table,
2271                                             struct flowi6 *fl6,
2272                                             const struct sk_buff *skb,
2273                                             int flags)
2274 {
2275         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2276 }
2277
2278 struct dst_entry *ip6_route_input_lookup(struct net *net,
2279                                          struct net_device *dev,
2280                                          struct flowi6 *fl6,
2281                                          const struct sk_buff *skb,
2282                                          int flags)
2283 {
2284         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2285                 flags |= RT6_LOOKUP_F_IFACE;
2286
2287         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2288 }
2289 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2290
2291 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2292                                   struct flow_keys *keys,
2293                                   struct flow_keys *flkeys)
2294 {
2295         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2296         const struct ipv6hdr *key_iph = outer_iph;
2297         struct flow_keys *_flkeys = flkeys;
2298         const struct ipv6hdr *inner_iph;
2299         const struct icmp6hdr *icmph;
2300         struct ipv6hdr _inner_iph;
2301         struct icmp6hdr _icmph;
2302
2303         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2304                 goto out;
2305
2306         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2307                                    sizeof(_icmph), &_icmph);
2308         if (!icmph)
2309                 goto out;
2310
2311         if (!icmpv6_is_err(icmph->icmp6_type))
2312                 goto out;
2313
2314         inner_iph = skb_header_pointer(skb,
2315                                        skb_transport_offset(skb) + sizeof(*icmph),
2316                                        sizeof(_inner_iph), &_inner_iph);
2317         if (!inner_iph)
2318                 goto out;
2319
2320         key_iph = inner_iph;
2321         _flkeys = NULL;
2322 out:
2323         if (_flkeys) {
2324                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2325                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2326                 keys->tags.flow_label = _flkeys->tags.flow_label;
2327                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2328         } else {
2329                 keys->addrs.v6addrs.src = key_iph->saddr;
2330                 keys->addrs.v6addrs.dst = key_iph->daddr;
2331                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2332                 keys->basic.ip_proto = key_iph->nexthdr;
2333         }
2334 }
2335
2336 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2337                                            const struct sk_buff *skb,
2338                                            bool *p_has_inner)
2339 {
2340         u32 hash_fields = ip6_multipath_hash_fields(net);
2341         struct flow_keys keys, hash_keys;
2342
2343         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2344                 return 0;
2345
2346         memset(&hash_keys, 0, sizeof(hash_keys));
2347         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2348
2349         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2350         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2351                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2352         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2353                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2354         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2355                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2356         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2357                 hash_keys.tags.flow_label = keys.tags.flow_label;
2358         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2359                 hash_keys.ports.src = keys.ports.src;
2360         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2361                 hash_keys.ports.dst = keys.ports.dst;
2362
2363         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2364         return flow_hash_from_keys(&hash_keys);
2365 }
2366
2367 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2368                                            const struct sk_buff *skb,
2369                                            bool has_inner)
2370 {
2371         u32 hash_fields = ip6_multipath_hash_fields(net);
2372         struct flow_keys keys, hash_keys;
2373
2374         /* We assume the packet carries an encapsulation, but if none was
2375          * encountered during dissection of the outer flow, then there is no
2376          * point in calling the flow dissector again.
2377          */
2378         if (!has_inner)
2379                 return 0;
2380
2381         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2382                 return 0;
2383
2384         memset(&hash_keys, 0, sizeof(hash_keys));
2385         skb_flow_dissect_flow_keys(skb, &keys, 0);
2386
2387         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2388                 return 0;
2389
2390         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2391                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2392                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2393                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2394                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2395                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2396         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2397                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2398                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2399                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2400                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2401                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2402                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2403                         hash_keys.tags.flow_label = keys.tags.flow_label;
2404         }
2405
2406         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2407                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2408         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2409                 hash_keys.ports.src = keys.ports.src;
2410         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2411                 hash_keys.ports.dst = keys.ports.dst;
2412
2413         return flow_hash_from_keys(&hash_keys);
2414 }
2415
2416 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2417                                          const struct sk_buff *skb)
2418 {
2419         u32 mhash, mhash_inner;
2420         bool has_inner = true;
2421
2422         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2423         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2424
2425         return jhash_2words(mhash, mhash_inner, 0);
2426 }
2427
2428 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2429                                          const struct flowi6 *fl6)
2430 {
2431         u32 hash_fields = ip6_multipath_hash_fields(net);
2432         struct flow_keys hash_keys;
2433
2434         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2435                 return 0;
2436
2437         memset(&hash_keys, 0, sizeof(hash_keys));
2438         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2439         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2440                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2441         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2442                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2443         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2444                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2445         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2446                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2447         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2448                 hash_keys.ports.src = fl6->fl6_sport;
2449         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2450                 hash_keys.ports.dst = fl6->fl6_dport;
2451
2452         return flow_hash_from_keys(&hash_keys);
2453 }
2454
2455 /* if skb is set it will be used and fl6 can be NULL */
2456 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2457                        const struct sk_buff *skb, struct flow_keys *flkeys)
2458 {
2459         struct flow_keys hash_keys;
2460         u32 mhash = 0;
2461
2462         switch (ip6_multipath_hash_policy(net)) {
2463         case 0:
2464                 memset(&hash_keys, 0, sizeof(hash_keys));
2465                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2466                 if (skb) {
2467                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2468                 } else {
2469                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2470                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2471                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2472                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2473                 }
2474                 mhash = flow_hash_from_keys(&hash_keys);
2475                 break;
2476         case 1:
2477                 if (skb) {
2478                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2479                         struct flow_keys keys;
2480
2481                         /* short-circuit if we already have L4 hash present */
2482                         if (skb->l4_hash)
2483                                 return skb_get_hash_raw(skb) >> 1;
2484
2485                         memset(&hash_keys, 0, sizeof(hash_keys));
2486
2487                         if (!flkeys) {
2488                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2489                                 flkeys = &keys;
2490                         }
2491                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2492                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2493                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2494                         hash_keys.ports.src = flkeys->ports.src;
2495                         hash_keys.ports.dst = flkeys->ports.dst;
2496                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2497                 } else {
2498                         memset(&hash_keys, 0, sizeof(hash_keys));
2499                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2500                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2501                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2502                         hash_keys.ports.src = fl6->fl6_sport;
2503                         hash_keys.ports.dst = fl6->fl6_dport;
2504                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2505                 }
2506                 mhash = flow_hash_from_keys(&hash_keys);
2507                 break;
2508         case 2:
2509                 memset(&hash_keys, 0, sizeof(hash_keys));
2510                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2511                 if (skb) {
2512                         struct flow_keys keys;
2513
2514                         if (!flkeys) {
2515                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2516                                 flkeys = &keys;
2517                         }
2518
2519                         /* Inner can be v4 or v6 */
2520                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2521                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2522                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2523                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2524                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2525                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2526                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2527                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2528                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2529                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2530                         } else {
2531                                 /* Same as case 0 */
2532                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2533                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2534                         }
2535                 } else {
2536                         /* Same as case 0 */
2537                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2538                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2539                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2540                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2541                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2542                 }
2543                 mhash = flow_hash_from_keys(&hash_keys);
2544                 break;
2545         case 3:
2546                 if (skb)
2547                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2548                 else
2549                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2550                 break;
2551         }
2552
2553         return mhash >> 1;
2554 }
2555
2556 /* Called with rcu held */
2557 void ip6_route_input(struct sk_buff *skb)
2558 {
2559         const struct ipv6hdr *iph = ipv6_hdr(skb);
2560         struct net *net = dev_net(skb->dev);
2561         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2562         struct ip_tunnel_info *tun_info;
2563         struct flowi6 fl6 = {
2564                 .flowi6_iif = skb->dev->ifindex,
2565                 .daddr = iph->daddr,
2566                 .saddr = iph->saddr,
2567                 .flowlabel = ip6_flowinfo(iph),
2568                 .flowi6_mark = skb->mark,
2569                 .flowi6_proto = iph->nexthdr,
2570         };
2571         struct flow_keys *flkeys = NULL, _flkeys;
2572
2573         tun_info = skb_tunnel_info(skb);
2574         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2575                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2576
2577         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2578                 flkeys = &_flkeys;
2579
2580         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2581                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2582         skb_dst_drop(skb);
2583         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2584                                                       &fl6, skb, flags));
2585 }
2586
2587 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2588                                              struct fib6_table *table,
2589                                              struct flowi6 *fl6,
2590                                              const struct sk_buff *skb,
2591                                              int flags)
2592 {
2593         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2594 }
2595
2596 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2597                                                const struct sock *sk,
2598                                                struct flowi6 *fl6, 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 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2628
2629 struct dst_entry *ip6_route_output_flags(struct net *net,
2630                                          const struct sock *sk,
2631                                          struct flowi6 *fl6,
2632                                          int flags)
2633 {
2634         struct dst_entry *dst;
2635         struct rt6_info *rt6;
2636
2637         rcu_read_lock();
2638         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2639         rt6 = (struct rt6_info *)dst;
2640         /* For dst cached in uncached_list, refcnt is already taken. */
2641         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2642                 dst = &net->ipv6.ip6_null_entry->dst;
2643                 dst_hold(dst);
2644         }
2645         rcu_read_unlock();
2646
2647         return dst;
2648 }
2649 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2650
2651 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2652 {
2653         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2654         struct net_device *loopback_dev = net->loopback_dev;
2655         struct dst_entry *new = NULL;
2656
2657         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2658                        DST_OBSOLETE_DEAD, 0);
2659         if (rt) {
2660                 rt6_info_init(rt);
2661                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2662
2663                 new = &rt->dst;
2664                 new->__use = 1;
2665                 new->input = dst_discard;
2666                 new->output = dst_discard_out;
2667
2668                 dst_copy_metrics(new, &ort->dst);
2669
2670                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2671                 rt->rt6i_gateway = ort->rt6i_gateway;
2672                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2673
2674                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2675 #ifdef CONFIG_IPV6_SUBTREES
2676                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2677 #endif
2678         }
2679
2680         dst_release(dst_orig);
2681         return new ? new : ERR_PTR(-ENOMEM);
2682 }
2683
2684 /*
2685  *      Destination cache support functions
2686  */
2687
2688 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2689 {
2690         u32 rt_cookie = 0;
2691
2692         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2693                 return false;
2694
2695         if (fib6_check_expired(f6i))
2696                 return false;
2697
2698         return true;
2699 }
2700
2701 static struct dst_entry *rt6_check(struct rt6_info *rt,
2702                                    struct fib6_info *from,
2703                                    u32 cookie)
2704 {
2705         u32 rt_cookie = 0;
2706
2707         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2708             rt_cookie != cookie)
2709                 return NULL;
2710
2711         if (rt6_check_expired(rt))
2712                 return NULL;
2713
2714         return &rt->dst;
2715 }
2716
2717 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2718                                             struct fib6_info *from,
2719                                             u32 cookie)
2720 {
2721         if (!__rt6_check_expired(rt) &&
2722             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2723             fib6_check(from, cookie))
2724                 return &rt->dst;
2725         else
2726                 return NULL;
2727 }
2728
2729 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2730                                                         u32 cookie)
2731 {
2732         struct dst_entry *dst_ret;
2733         struct fib6_info *from;
2734         struct rt6_info *rt;
2735
2736         rt = container_of(dst, struct rt6_info, dst);
2737
2738         if (rt->sernum)
2739                 return rt6_is_valid(rt) ? dst : NULL;
2740
2741         rcu_read_lock();
2742
2743         /* All IPV6 dsts are created with ->obsolete set to the value
2744          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2745          * into this function always.
2746          */
2747
2748         from = rcu_dereference(rt->from);
2749
2750         if (from && (rt->rt6i_flags & RTF_PCPU ||
2751             unlikely(!list_empty(&rt->rt6i_uncached))))
2752                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2753         else
2754                 dst_ret = rt6_check(rt, from, cookie);
2755
2756         rcu_read_unlock();
2757
2758         return dst_ret;
2759 }
2760 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2761
2762 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2763 {
2764         struct rt6_info *rt = (struct rt6_info *) dst;
2765
2766         if (rt) {
2767                 if (rt->rt6i_flags & RTF_CACHE) {
2768                         rcu_read_lock();
2769                         if (rt6_check_expired(rt)) {
2770                                 rt6_remove_exception_rt(rt);
2771                                 dst = NULL;
2772                         }
2773                         rcu_read_unlock();
2774                 } else {
2775                         dst_release(dst);
2776                         dst = NULL;
2777                 }
2778         }
2779         return dst;
2780 }
2781
2782 static void ip6_link_failure(struct sk_buff *skb)
2783 {
2784         struct rt6_info *rt;
2785
2786         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2787
2788         rt = (struct rt6_info *) skb_dst(skb);
2789         if (rt) {
2790                 rcu_read_lock();
2791                 if (rt->rt6i_flags & RTF_CACHE) {
2792                         rt6_remove_exception_rt(rt);
2793                 } else {
2794                         struct fib6_info *from;
2795                         struct fib6_node *fn;
2796
2797                         from = rcu_dereference(rt->from);
2798                         if (from) {
2799                                 fn = rcu_dereference(from->fib6_node);
2800                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2801                                         WRITE_ONCE(fn->fn_sernum, -1);
2802                         }
2803                 }
2804                 rcu_read_unlock();
2805         }
2806 }
2807
2808 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2809 {
2810         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2811                 struct fib6_info *from;
2812
2813                 rcu_read_lock();
2814                 from = rcu_dereference(rt0->from);
2815                 if (from)
2816                         rt0->dst.expires = from->expires;
2817                 rcu_read_unlock();
2818         }
2819
2820         dst_set_expires(&rt0->dst, timeout);
2821         rt0->rt6i_flags |= RTF_EXPIRES;
2822 }
2823
2824 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2825 {
2826         struct net *net = dev_net(rt->dst.dev);
2827
2828         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2829         rt->rt6i_flags |= RTF_MODIFIED;
2830         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2831 }
2832
2833 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2834 {
2835         return !(rt->rt6i_flags & RTF_CACHE) &&
2836                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2837 }
2838
2839 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2840                                  const struct ipv6hdr *iph, u32 mtu,
2841                                  bool confirm_neigh)
2842 {
2843         const struct in6_addr *daddr, *saddr;
2844         struct rt6_info *rt6 = (struct rt6_info *)dst;
2845
2846         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2847          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2848          * [see also comment in rt6_mtu_change_route()]
2849          */
2850
2851         if (iph) {
2852                 daddr = &iph->daddr;
2853                 saddr = &iph->saddr;
2854         } else if (sk) {
2855                 daddr = &sk->sk_v6_daddr;
2856                 saddr = &inet6_sk(sk)->saddr;
2857         } else {
2858                 daddr = NULL;
2859                 saddr = NULL;
2860         }
2861
2862         if (confirm_neigh)
2863                 dst_confirm_neigh(dst, daddr);
2864
2865         if (mtu < IPV6_MIN_MTU)
2866                 return;
2867         if (mtu >= dst_mtu(dst))
2868                 return;
2869
2870         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2871                 rt6_do_update_pmtu(rt6, mtu);
2872                 /* update rt6_ex->stamp for cache */
2873                 if (rt6->rt6i_flags & RTF_CACHE)
2874                         rt6_update_exception_stamp_rt(rt6);
2875         } else if (daddr) {
2876                 struct fib6_result res = {};
2877                 struct rt6_info *nrt6;
2878
2879                 rcu_read_lock();
2880                 res.f6i = rcu_dereference(rt6->from);
2881                 if (!res.f6i)
2882                         goto out_unlock;
2883
2884                 res.fib6_flags = res.f6i->fib6_flags;
2885                 res.fib6_type = res.f6i->fib6_type;
2886
2887                 if (res.f6i->nh) {
2888                         struct fib6_nh_match_arg arg = {
2889                                 .dev = dst->dev,
2890                                 .gw = &rt6->rt6i_gateway,
2891                         };
2892
2893                         nexthop_for_each_fib6_nh(res.f6i->nh,
2894                                                  fib6_nh_find_match, &arg);
2895
2896                         /* fib6_info uses a nexthop that does not have fib6_nh
2897                          * using the dst->dev + gw. Should be impossible.
2898                          */
2899                         if (!arg.match)
2900                                 goto out_unlock;
2901
2902                         res.nh = arg.match;
2903                 } else {
2904                         res.nh = res.f6i->fib6_nh;
2905                 }
2906
2907                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2908                 if (nrt6) {
2909                         rt6_do_update_pmtu(nrt6, mtu);
2910                         if (rt6_insert_exception(nrt6, &res))
2911                                 dst_release_immediate(&nrt6->dst);
2912                 }
2913 out_unlock:
2914                 rcu_read_unlock();
2915         }
2916 }
2917
2918 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2919                                struct sk_buff *skb, u32 mtu,
2920                                bool confirm_neigh)
2921 {
2922         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2923                              confirm_neigh);
2924 }
2925
2926 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2927                      int oif, u32 mark, kuid_t uid)
2928 {
2929         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2930         struct dst_entry *dst;
2931         struct flowi6 fl6 = {
2932                 .flowi6_oif = oif,
2933                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2934                 .daddr = iph->daddr,
2935                 .saddr = iph->saddr,
2936                 .flowlabel = ip6_flowinfo(iph),
2937                 .flowi6_uid = uid,
2938         };
2939
2940         dst = ip6_route_output(net, NULL, &fl6);
2941         if (!dst->error)
2942                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2943         dst_release(dst);
2944 }
2945 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2946
2947 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2948 {
2949         int oif = sk->sk_bound_dev_if;
2950         struct dst_entry *dst;
2951
2952         if (!oif && skb->dev)
2953                 oif = l3mdev_master_ifindex(skb->dev);
2954
2955         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, 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, sk->sk_mark,
3177                      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 int 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_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3292         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3293         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3294         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3295         unsigned int val;
3296         int entries;
3297
3298         entries = dst_entries_get_fast(ops);
3299         if (entries > rt_max_size)
3300                 entries = dst_entries_get_slow(ops);
3301
3302         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3303             entries <= rt_max_size)
3304                 goto out;
3305
3306         fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3307         entries = dst_entries_get_slow(ops);
3308         if (entries < ops->gc_thresh)
3309                 atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3310 out:
3311         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3312         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3313         return entries > rt_max_size;
3314 }
3315
3316 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3317                                const struct in6_addr *gw_addr, u32 tbid,
3318                                int flags, struct fib6_result *res)
3319 {
3320         struct flowi6 fl6 = {
3321                 .flowi6_oif = cfg->fc_ifindex,
3322                 .daddr = *gw_addr,
3323                 .saddr = cfg->fc_prefsrc,
3324         };
3325         struct fib6_table *table;
3326         int err;
3327
3328         table = fib6_get_table(net, tbid);
3329         if (!table)
3330                 return -EINVAL;
3331
3332         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3333                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3334
3335         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3336
3337         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3338         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3339                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3340                                  cfg->fc_ifindex != 0, NULL, flags);
3341
3342         return err;
3343 }
3344
3345 static int ip6_route_check_nh_onlink(struct net *net,
3346                                      struct fib6_config *cfg,
3347                                      const struct net_device *dev,
3348                                      struct netlink_ext_ack *extack)
3349 {
3350         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3351         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3352         struct fib6_result res = {};
3353         int err;
3354
3355         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3356         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3357             /* ignore match if it is the default route */
3358             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3359             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3360                 NL_SET_ERR_MSG(extack,
3361                                "Nexthop has invalid gateway or device mismatch");
3362                 err = -EINVAL;
3363         }
3364
3365         return err;
3366 }
3367
3368 static int ip6_route_check_nh(struct net *net,
3369                               struct fib6_config *cfg,
3370                               struct net_device **_dev,
3371                               struct inet6_dev **idev)
3372 {
3373         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3374         struct net_device *dev = _dev ? *_dev : NULL;
3375         int flags = RT6_LOOKUP_F_IFACE;
3376         struct fib6_result res = {};
3377         int err = -EHOSTUNREACH;
3378
3379         if (cfg->fc_table) {
3380                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3381                                           cfg->fc_table, flags, &res);
3382                 /* gw_addr can not require a gateway or resolve to a reject
3383                  * route. If a device is given, it must match the result.
3384                  */
3385                 if (err || res.fib6_flags & RTF_REJECT ||
3386                     res.nh->fib_nh_gw_family ||
3387                     (dev && dev != res.nh->fib_nh_dev))
3388                         err = -EHOSTUNREACH;
3389         }
3390
3391         if (err < 0) {
3392                 struct flowi6 fl6 = {
3393                         .flowi6_oif = cfg->fc_ifindex,
3394                         .daddr = *gw_addr,
3395                 };
3396
3397                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3398                 if (err || res.fib6_flags & RTF_REJECT ||
3399                     res.nh->fib_nh_gw_family)
3400                         err = -EHOSTUNREACH;
3401
3402                 if (err)
3403                         return err;
3404
3405                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3406                                  cfg->fc_ifindex != 0, NULL, flags);
3407         }
3408
3409         err = 0;
3410         if (dev) {
3411                 if (dev != res.nh->fib_nh_dev)
3412                         err = -EHOSTUNREACH;
3413         } else {
3414                 *_dev = dev = res.nh->fib_nh_dev;
3415                 dev_hold(dev);
3416                 *idev = in6_dev_get(dev);
3417         }
3418
3419         return err;
3420 }
3421
3422 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3423                            struct net_device **_dev, struct inet6_dev **idev,
3424                            struct netlink_ext_ack *extack)
3425 {
3426         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3427         int gwa_type = ipv6_addr_type(gw_addr);
3428         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3429         const struct net_device *dev = *_dev;
3430         bool need_addr_check = !dev;
3431         int err = -EINVAL;
3432
3433         /* if gw_addr is local we will fail to detect this in case
3434          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3435          * will return already-added prefix route via interface that
3436          * prefix route was assigned to, which might be non-loopback.
3437          */
3438         if (dev &&
3439             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3440                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3441                 goto out;
3442         }
3443
3444         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3445                 /* IPv6 strictly inhibits using not link-local
3446                  * addresses as nexthop address.
3447                  * Otherwise, router will not able to send redirects.
3448                  * It is very good, but in some (rare!) circumstances
3449                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3450                  * some exceptions. --ANK
3451                  * We allow IPv4-mapped nexthops to support RFC4798-type
3452                  * addressing
3453                  */
3454                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3455                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3456                         goto out;
3457                 }
3458
3459                 rcu_read_lock();
3460
3461                 if (cfg->fc_flags & RTNH_F_ONLINK)
3462                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3463                 else
3464                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3465
3466                 rcu_read_unlock();
3467
3468                 if (err)
3469                         goto out;
3470         }
3471
3472         /* reload in case device was changed */
3473         dev = *_dev;
3474
3475         err = -EINVAL;
3476         if (!dev) {
3477                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3478                 goto out;
3479         } else if (dev->flags & IFF_LOOPBACK) {
3480                 NL_SET_ERR_MSG(extack,
3481                                "Egress device can not be loopback device for this route");
3482                 goto out;
3483         }
3484
3485         /* if we did not check gw_addr above, do so now that the
3486          * egress device has been resolved.
3487          */
3488         if (need_addr_check &&
3489             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3490                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3491                 goto out;
3492         }
3493
3494         err = 0;
3495 out:
3496         return err;
3497 }
3498
3499 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3500 {
3501         if ((flags & RTF_REJECT) ||
3502             (dev && (dev->flags & IFF_LOOPBACK) &&
3503              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3504              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3505                 return true;
3506
3507         return false;
3508 }
3509
3510 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3511                  struct fib6_config *cfg, gfp_t gfp_flags,
3512                  struct netlink_ext_ack *extack)
3513 {
3514         struct net_device *dev = NULL;
3515         struct inet6_dev *idev = NULL;
3516         int addr_type;
3517         int err;
3518
3519         fib6_nh->fib_nh_family = AF_INET6;
3520 #ifdef CONFIG_IPV6_ROUTER_PREF
3521         fib6_nh->last_probe = jiffies;
3522 #endif
3523         if (cfg->fc_is_fdb) {
3524                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3525                 fib6_nh->fib_nh_gw_family = AF_INET6;
3526                 return 0;
3527         }
3528
3529         err = -ENODEV;
3530         if (cfg->fc_ifindex) {
3531                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3532                 if (!dev)
3533                         goto out;
3534                 idev = in6_dev_get(dev);
3535                 if (!idev)
3536                         goto out;
3537         }
3538
3539         if (cfg->fc_flags & RTNH_F_ONLINK) {
3540                 if (!dev) {
3541                         NL_SET_ERR_MSG(extack,
3542                                        "Nexthop device required for onlink");
3543                         goto out;
3544                 }
3545
3546                 if (!(dev->flags & IFF_UP)) {
3547                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3548                         err = -ENETDOWN;
3549                         goto out;
3550                 }
3551
3552                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3553         }
3554
3555         fib6_nh->fib_nh_weight = 1;
3556
3557         /* We cannot add true routes via loopback here,
3558          * they would result in kernel looping; promote them to reject routes
3559          */
3560         addr_type = ipv6_addr_type(&cfg->fc_dst);
3561         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3562                 /* hold loopback dev/idev if we haven't done so. */
3563                 if (dev != net->loopback_dev) {
3564                         if (dev) {
3565                                 dev_put(dev);
3566                                 in6_dev_put(idev);
3567                         }
3568                         dev = net->loopback_dev;
3569                         dev_hold(dev);
3570                         idev = in6_dev_get(dev);
3571                         if (!idev) {
3572                                 err = -ENODEV;
3573                                 goto out;
3574                         }
3575                 }
3576                 goto pcpu_alloc;
3577         }
3578
3579         if (cfg->fc_flags & RTF_GATEWAY) {
3580                 err = ip6_validate_gw(net, cfg, &dev, &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         netdev_tracker_alloc(dev, &fib6_nh->fib_nh_dev_tracker, gfp_flags);
3622
3623         fib6_nh->fib_nh_oif = dev->ifindex;
3624         err = 0;
3625 out:
3626         if (idev)
3627                 in6_dev_put(idev);
3628
3629         if (err) {
3630                 lwtstate_put(fib6_nh->fib_nh_lws);
3631                 fib6_nh->fib_nh_lws = NULL;
3632                 dev_put(dev);
3633         }
3634
3635         return err;
3636 }
3637
3638 void fib6_nh_release(struct fib6_nh *fib6_nh)
3639 {
3640         struct rt6_exception_bucket *bucket;
3641
3642         rcu_read_lock();
3643
3644         fib6_nh_flush_exceptions(fib6_nh, NULL);
3645         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3646         if (bucket) {
3647                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3648                 kfree(bucket);
3649         }
3650
3651         rcu_read_unlock();
3652
3653         fib6_nh_release_dsts(fib6_nh);
3654         free_percpu(fib6_nh->rt6i_pcpu);
3655
3656         fib_nh_common_release(&fib6_nh->nh_common);
3657 }
3658
3659 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3660 {
3661         int cpu;
3662
3663         if (!fib6_nh->rt6i_pcpu)
3664                 return;
3665
3666         for_each_possible_cpu(cpu) {
3667                 struct rt6_info *pcpu_rt, **ppcpu_rt;
3668
3669                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3670                 pcpu_rt = xchg(ppcpu_rt, NULL);
3671                 if (pcpu_rt) {
3672                         dst_dev_put(&pcpu_rt->dst);
3673                         dst_release(&pcpu_rt->dst);
3674                 }
3675         }
3676 }
3677
3678 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3679                                               gfp_t gfp_flags,
3680                                               struct netlink_ext_ack *extack)
3681 {
3682         struct net *net = cfg->fc_nlinfo.nl_net;
3683         struct fib6_info *rt = NULL;
3684         struct nexthop *nh = NULL;
3685         struct fib6_table *table;
3686         struct fib6_nh *fib6_nh;
3687         int err = -EINVAL;
3688         int addr_type;
3689
3690         /* RTF_PCPU is an internal flag; can not be set by userspace */
3691         if (cfg->fc_flags & RTF_PCPU) {
3692                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3693                 goto out;
3694         }
3695
3696         /* RTF_CACHE is an internal flag; can not be set by userspace */
3697         if (cfg->fc_flags & RTF_CACHE) {
3698                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3699                 goto out;
3700         }
3701
3702         if (cfg->fc_type > RTN_MAX) {
3703                 NL_SET_ERR_MSG(extack, "Invalid route type");
3704                 goto out;
3705         }
3706
3707         if (cfg->fc_dst_len > 128) {
3708                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3709                 goto out;
3710         }
3711         if (cfg->fc_src_len > 128) {
3712                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3713                 goto out;
3714         }
3715 #ifndef CONFIG_IPV6_SUBTREES
3716         if (cfg->fc_src_len) {
3717                 NL_SET_ERR_MSG(extack,
3718                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3719                 goto out;
3720         }
3721 #endif
3722         if (cfg->fc_nh_id) {
3723                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3724                 if (!nh) {
3725                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3726                         goto out;
3727                 }
3728                 err = fib6_check_nexthop(nh, cfg, extack);
3729                 if (err)
3730                         goto out;
3731         }
3732
3733         err = -ENOBUFS;
3734         if (cfg->fc_nlinfo.nlh &&
3735             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3736                 table = fib6_get_table(net, cfg->fc_table);
3737                 if (!table) {
3738                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3739                         table = fib6_new_table(net, cfg->fc_table);
3740                 }
3741         } else {
3742                 table = fib6_new_table(net, cfg->fc_table);
3743         }
3744
3745         if (!table)
3746                 goto out;
3747
3748         err = -ENOMEM;
3749         rt = fib6_info_alloc(gfp_flags, !nh);
3750         if (!rt)
3751                 goto out;
3752
3753         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3754                                                extack);
3755         if (IS_ERR(rt->fib6_metrics)) {
3756                 err = PTR_ERR(rt->fib6_metrics);
3757                 /* Do not leave garbage there. */
3758                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3759                 goto out_free;
3760         }
3761
3762         if (cfg->fc_flags & RTF_ADDRCONF)
3763                 rt->dst_nocount = true;
3764
3765         if (cfg->fc_flags & RTF_EXPIRES)
3766                 fib6_set_expires(rt, jiffies +
3767                                 clock_t_to_jiffies(cfg->fc_expires));
3768         else
3769                 fib6_clean_expires(rt);
3770
3771         if (cfg->fc_protocol == RTPROT_UNSPEC)
3772                 cfg->fc_protocol = RTPROT_BOOT;
3773         rt->fib6_protocol = cfg->fc_protocol;
3774
3775         rt->fib6_table = table;
3776         rt->fib6_metric = cfg->fc_metric;
3777         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3778         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3779
3780         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3781         rt->fib6_dst.plen = cfg->fc_dst_len;
3782
3783 #ifdef CONFIG_IPV6_SUBTREES
3784         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3785         rt->fib6_src.plen = cfg->fc_src_len;
3786 #endif
3787         if (nh) {
3788                 if (rt->fib6_src.plen) {
3789                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3790                         goto out_free;
3791                 }
3792                 if (!nexthop_get(nh)) {
3793                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3794                         goto out_free;
3795                 }
3796                 rt->nh = nh;
3797                 fib6_nh = nexthop_fib6_nh(rt->nh);
3798         } else {
3799                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3800                 if (err)
3801                         goto out;
3802
3803                 fib6_nh = rt->fib6_nh;
3804
3805                 /* We cannot add true routes via loopback here, they would
3806                  * result in kernel looping; promote them to reject routes
3807                  */
3808                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3809                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3810                                    addr_type))
3811                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3812         }
3813
3814         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3815                 struct net_device *dev = fib6_nh->fib_nh_dev;
3816
3817                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3818                         NL_SET_ERR_MSG(extack, "Invalid source address");
3819                         err = -EINVAL;
3820                         goto out;
3821                 }
3822                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3823                 rt->fib6_prefsrc.plen = 128;
3824         } else
3825                 rt->fib6_prefsrc.plen = 0;
3826
3827         return rt;
3828 out:
3829         fib6_info_release(rt);
3830         return ERR_PTR(err);
3831 out_free:
3832         ip_fib_metrics_put(rt->fib6_metrics);
3833         kfree(rt);
3834         return ERR_PTR(err);
3835 }
3836
3837 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3838                   struct netlink_ext_ack *extack)
3839 {
3840         struct fib6_info *rt;
3841         int err;
3842
3843         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3844         if (IS_ERR(rt))
3845                 return PTR_ERR(rt);
3846
3847         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3848         fib6_info_release(rt);
3849
3850         return err;
3851 }
3852
3853 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3854 {
3855         struct net *net = info->nl_net;
3856         struct fib6_table *table;
3857         int err;
3858
3859         if (rt == net->ipv6.fib6_null_entry) {
3860                 err = -ENOENT;
3861                 goto out;
3862         }
3863
3864         table = rt->fib6_table;
3865         spin_lock_bh(&table->tb6_lock);
3866         err = fib6_del(rt, info);
3867         spin_unlock_bh(&table->tb6_lock);
3868
3869 out:
3870         fib6_info_release(rt);
3871         return err;
3872 }
3873
3874 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3875 {
3876         struct nl_info info = {
3877                 .nl_net = net,
3878                 .skip_notify = skip_notify
3879         };
3880
3881         return __ip6_del_rt(rt, &info);
3882 }
3883
3884 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3885 {
3886         struct nl_info *info = &cfg->fc_nlinfo;
3887         struct net *net = info->nl_net;
3888         struct sk_buff *skb = NULL;
3889         struct fib6_table *table;
3890         int err = -ENOENT;
3891
3892         if (rt == net->ipv6.fib6_null_entry)
3893                 goto out_put;
3894         table = rt->fib6_table;
3895         spin_lock_bh(&table->tb6_lock);
3896
3897         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3898                 struct fib6_info *sibling, *next_sibling;
3899                 struct fib6_node *fn;
3900
3901                 /* prefer to send a single notification with all hops */
3902                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3903                 if (skb) {
3904                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3905
3906                         if (rt6_fill_node(net, skb, rt, NULL,
3907                                           NULL, NULL, 0, RTM_DELROUTE,
3908                                           info->portid, seq, 0) < 0) {
3909                                 kfree_skb(skb);
3910                                 skb = NULL;
3911                         } else
3912                                 info->skip_notify = 1;
3913                 }
3914
3915                 /* 'rt' points to the first sibling route. If it is not the
3916                  * leaf, then we do not need to send a notification. Otherwise,
3917                  * we need to check if the last sibling has a next route or not
3918                  * and emit a replace or delete notification, respectively.
3919                  */
3920                 info->skip_notify_kernel = 1;
3921                 fn = rcu_dereference_protected(rt->fib6_node,
3922                                             lockdep_is_held(&table->tb6_lock));
3923                 if (rcu_access_pointer(fn->leaf) == rt) {
3924                         struct fib6_info *last_sibling, *replace_rt;
3925
3926                         last_sibling = list_last_entry(&rt->fib6_siblings,
3927                                                        struct fib6_info,
3928                                                        fib6_siblings);
3929                         replace_rt = rcu_dereference_protected(
3930                                             last_sibling->fib6_next,
3931                                             lockdep_is_held(&table->tb6_lock));
3932                         if (replace_rt)
3933                                 call_fib6_entry_notifiers_replace(net,
3934                                                                   replace_rt);
3935                         else
3936                                 call_fib6_multipath_entry_notifiers(net,
3937                                                        FIB_EVENT_ENTRY_DEL,
3938                                                        rt, rt->fib6_nsiblings,
3939                                                        NULL);
3940                 }
3941                 list_for_each_entry_safe(sibling, next_sibling,
3942                                          &rt->fib6_siblings,
3943                                          fib6_siblings) {
3944                         err = fib6_del(sibling, info);
3945                         if (err)
3946                                 goto out_unlock;
3947                 }
3948         }
3949
3950         err = fib6_del(rt, info);
3951 out_unlock:
3952         spin_unlock_bh(&table->tb6_lock);
3953 out_put:
3954         fib6_info_release(rt);
3955
3956         if (skb) {
3957                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3958                             info->nlh, gfp_any());
3959         }
3960         return err;
3961 }
3962
3963 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3964 {
3965         int rc = -ESRCH;
3966
3967         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3968                 goto out;
3969
3970         if (cfg->fc_flags & RTF_GATEWAY &&
3971             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3972                 goto out;
3973
3974         rc = rt6_remove_exception_rt(rt);
3975 out:
3976         return rc;
3977 }
3978
3979 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3980                              struct fib6_nh *nh)
3981 {
3982         struct fib6_result res = {
3983                 .f6i = rt,
3984                 .nh = nh,
3985         };
3986         struct rt6_info *rt_cache;
3987
3988         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3989         if (rt_cache)
3990                 return __ip6_del_cached_rt(rt_cache, cfg);
3991
3992         return 0;
3993 }
3994
3995 struct fib6_nh_del_cached_rt_arg {
3996         struct fib6_config *cfg;
3997         struct fib6_info *f6i;
3998 };
3999
4000 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4001 {
4002         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4003         int rc;
4004
4005         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4006         return rc != -ESRCH ? rc : 0;
4007 }
4008
4009 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4010 {
4011         struct fib6_nh_del_cached_rt_arg arg = {
4012                 .cfg = cfg,
4013                 .f6i = f6i
4014         };
4015
4016         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4017 }
4018
4019 static int ip6_route_del(struct fib6_config *cfg,
4020                          struct netlink_ext_ack *extack)
4021 {
4022         struct fib6_table *table;
4023         struct fib6_info *rt;
4024         struct fib6_node *fn;
4025         int err = -ESRCH;
4026
4027         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4028         if (!table) {
4029                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4030                 return err;
4031         }
4032
4033         rcu_read_lock();
4034
4035         fn = fib6_locate(&table->tb6_root,
4036                          &cfg->fc_dst, cfg->fc_dst_len,
4037                          &cfg->fc_src, cfg->fc_src_len,
4038                          !(cfg->fc_flags & RTF_CACHE));
4039
4040         if (fn) {
4041                 for_each_fib6_node_rt_rcu(fn) {
4042                         struct fib6_nh *nh;
4043
4044                         if (rt->nh && cfg->fc_nh_id &&
4045                             rt->nh->id != cfg->fc_nh_id)
4046                                 continue;
4047
4048                         if (cfg->fc_flags & RTF_CACHE) {
4049                                 int rc = 0;
4050
4051                                 if (rt->nh) {
4052                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4053                                 } else if (cfg->fc_nh_id) {
4054                                         continue;
4055                                 } else {
4056                                         nh = rt->fib6_nh;
4057                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4058                                 }
4059                                 if (rc != -ESRCH) {
4060                                         rcu_read_unlock();
4061                                         return rc;
4062                                 }
4063                                 continue;
4064                         }
4065
4066                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4067                                 continue;
4068                         if (cfg->fc_protocol &&
4069                             cfg->fc_protocol != rt->fib6_protocol)
4070                                 continue;
4071
4072                         if (rt->nh) {
4073                                 if (!fib6_info_hold_safe(rt))
4074                                         continue;
4075                                 rcu_read_unlock();
4076
4077                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4078                         }
4079                         if (cfg->fc_nh_id)
4080                                 continue;
4081
4082                         nh = rt->fib6_nh;
4083                         if (cfg->fc_ifindex &&
4084                             (!nh->fib_nh_dev ||
4085                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4086                                 continue;
4087                         if (cfg->fc_flags & RTF_GATEWAY &&
4088                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4089                                 continue;
4090                         if (!fib6_info_hold_safe(rt))
4091                                 continue;
4092                         rcu_read_unlock();
4093
4094                         /* if gateway was specified only delete the one hop */
4095                         if (cfg->fc_flags & RTF_GATEWAY)
4096                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4097
4098                         return __ip6_del_rt_siblings(rt, cfg);
4099                 }
4100         }
4101         rcu_read_unlock();
4102
4103         return err;
4104 }
4105
4106 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4107 {
4108         struct netevent_redirect netevent;
4109         struct rt6_info *rt, *nrt = NULL;
4110         struct fib6_result res = {};
4111         struct ndisc_options ndopts;
4112         struct inet6_dev *in6_dev;
4113         struct neighbour *neigh;
4114         struct rd_msg *msg;
4115         int optlen, on_link;
4116         u8 *lladdr;
4117
4118         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4119         optlen -= sizeof(*msg);
4120
4121         if (optlen < 0) {
4122                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4123                 return;
4124         }
4125
4126         msg = (struct rd_msg *)icmp6_hdr(skb);
4127
4128         if (ipv6_addr_is_multicast(&msg->dest)) {
4129                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4130                 return;
4131         }
4132
4133         on_link = 0;
4134         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4135                 on_link = 1;
4136         } else if (ipv6_addr_type(&msg->target) !=
4137                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4138                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4139                 return;
4140         }
4141
4142         in6_dev = __in6_dev_get(skb->dev);
4143         if (!in6_dev)
4144                 return;
4145         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4146                 return;
4147
4148         /* RFC2461 8.1:
4149          *      The IP source address of the Redirect MUST be the same as the current
4150          *      first-hop router for the specified ICMP Destination Address.
4151          */
4152
4153         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4154                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4155                 return;
4156         }
4157
4158         lladdr = NULL;
4159         if (ndopts.nd_opts_tgt_lladdr) {
4160                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4161                                              skb->dev);
4162                 if (!lladdr) {
4163                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4164                         return;
4165                 }
4166         }
4167
4168         rt = (struct rt6_info *) dst;
4169         if (rt->rt6i_flags & RTF_REJECT) {
4170                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4171                 return;
4172         }
4173
4174         /* Redirect received -> path was valid.
4175          * Look, redirects are sent only in response to data packets,
4176          * so that this nexthop apparently is reachable. --ANK
4177          */
4178         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4179
4180         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4181         if (!neigh)
4182                 return;
4183
4184         /*
4185          *      We have finally decided to accept it.
4186          */
4187
4188         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4189                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4190                      NEIGH_UPDATE_F_OVERRIDE|
4191                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4192                                      NEIGH_UPDATE_F_ISROUTER)),
4193                      NDISC_REDIRECT, &ndopts);
4194
4195         rcu_read_lock();
4196         res.f6i = rcu_dereference(rt->from);
4197         if (!res.f6i)
4198                 goto out;
4199
4200         if (res.f6i->nh) {
4201                 struct fib6_nh_match_arg arg = {
4202                         .dev = dst->dev,
4203                         .gw = &rt->rt6i_gateway,
4204                 };
4205
4206                 nexthop_for_each_fib6_nh(res.f6i->nh,
4207                                          fib6_nh_find_match, &arg);
4208
4209                 /* fib6_info uses a nexthop that does not have fib6_nh
4210                  * using the dst->dev. Should be impossible
4211                  */
4212                 if (!arg.match)
4213                         goto out;
4214                 res.nh = arg.match;
4215         } else {
4216                 res.nh = res.f6i->fib6_nh;
4217         }
4218
4219         res.fib6_flags = res.f6i->fib6_flags;
4220         res.fib6_type = res.f6i->fib6_type;
4221         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4222         if (!nrt)
4223                 goto out;
4224
4225         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4226         if (on_link)
4227                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4228
4229         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4230
4231         /* rt6_insert_exception() will take care of duplicated exceptions */
4232         if (rt6_insert_exception(nrt, &res)) {
4233                 dst_release_immediate(&nrt->dst);
4234                 goto out;
4235         }
4236
4237         netevent.old = &rt->dst;
4238         netevent.new = &nrt->dst;
4239         netevent.daddr = &msg->dest;
4240         netevent.neigh = neigh;
4241         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4242
4243 out:
4244         rcu_read_unlock();
4245         neigh_release(neigh);
4246 }
4247
4248 #ifdef CONFIG_IPV6_ROUTE_INFO
4249 static struct fib6_info *rt6_get_route_info(struct net *net,
4250                                            const struct in6_addr *prefix, int prefixlen,
4251                                            const struct in6_addr *gwaddr,
4252                                            struct net_device *dev)
4253 {
4254         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4255         int ifindex = dev->ifindex;
4256         struct fib6_node *fn;
4257         struct fib6_info *rt = NULL;
4258         struct fib6_table *table;
4259
4260         table = fib6_get_table(net, tb_id);
4261         if (!table)
4262                 return NULL;
4263
4264         rcu_read_lock();
4265         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4266         if (!fn)
4267                 goto out;
4268
4269         for_each_fib6_node_rt_rcu(fn) {
4270                 /* these routes do not use nexthops */
4271                 if (rt->nh)
4272                         continue;
4273                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4274                         continue;
4275                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4276                     !rt->fib6_nh->fib_nh_gw_family)
4277                         continue;
4278                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4279                         continue;
4280                 if (!fib6_info_hold_safe(rt))
4281                         continue;
4282                 break;
4283         }
4284 out:
4285         rcu_read_unlock();
4286         return rt;
4287 }
4288
4289 static struct fib6_info *rt6_add_route_info(struct net *net,
4290                                            const struct in6_addr *prefix, int prefixlen,
4291                                            const struct in6_addr *gwaddr,
4292                                            struct net_device *dev,
4293                                            unsigned int pref)
4294 {
4295         struct fib6_config cfg = {
4296                 .fc_metric      = IP6_RT_PRIO_USER,
4297                 .fc_ifindex     = dev->ifindex,
4298                 .fc_dst_len     = prefixlen,
4299                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4300                                   RTF_UP | RTF_PREF(pref),
4301                 .fc_protocol = RTPROT_RA,
4302                 .fc_type = RTN_UNICAST,
4303                 .fc_nlinfo.portid = 0,
4304                 .fc_nlinfo.nlh = NULL,
4305                 .fc_nlinfo.nl_net = net,
4306         };
4307
4308         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4309         cfg.fc_dst = *prefix;
4310         cfg.fc_gateway = *gwaddr;
4311
4312         /* We should treat it as a default route if prefix length is 0. */
4313         if (!prefixlen)
4314                 cfg.fc_flags |= RTF_DEFAULT;
4315
4316         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4317
4318         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4319 }
4320 #endif
4321
4322 struct fib6_info *rt6_get_dflt_router(struct net *net,
4323                                      const struct in6_addr *addr,
4324                                      struct net_device *dev)
4325 {
4326         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4327         struct fib6_info *rt;
4328         struct fib6_table *table;
4329
4330         table = fib6_get_table(net, tb_id);
4331         if (!table)
4332                 return NULL;
4333
4334         rcu_read_lock();
4335         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4336                 struct fib6_nh *nh;
4337
4338                 /* RA routes do not use nexthops */
4339                 if (rt->nh)
4340                         continue;
4341
4342                 nh = rt->fib6_nh;
4343                 if (dev == nh->fib_nh_dev &&
4344                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4345                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4346                         break;
4347         }
4348         if (rt && !fib6_info_hold_safe(rt))
4349                 rt = NULL;
4350         rcu_read_unlock();
4351         return rt;
4352 }
4353
4354 struct fib6_info *rt6_add_dflt_router(struct net *net,
4355                                      const struct in6_addr *gwaddr,
4356                                      struct net_device *dev,
4357                                      unsigned int pref,
4358                                      u32 defrtr_usr_metric)
4359 {
4360         struct fib6_config cfg = {
4361                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4362                 .fc_metric      = defrtr_usr_metric,
4363                 .fc_ifindex     = dev->ifindex,
4364                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4365                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4366                 .fc_protocol = RTPROT_RA,
4367                 .fc_type = RTN_UNICAST,
4368                 .fc_nlinfo.portid = 0,
4369                 .fc_nlinfo.nlh = NULL,
4370                 .fc_nlinfo.nl_net = net,
4371         };
4372
4373         cfg.fc_gateway = *gwaddr;
4374
4375         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4376                 struct fib6_table *table;
4377
4378                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4379                 if (table)
4380                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4381         }
4382
4383         return rt6_get_dflt_router(net, gwaddr, dev);
4384 }
4385
4386 static void __rt6_purge_dflt_routers(struct net *net,
4387                                      struct fib6_table *table)
4388 {
4389         struct fib6_info *rt;
4390
4391 restart:
4392         rcu_read_lock();
4393         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4394                 struct net_device *dev = fib6_info_nh_dev(rt);
4395                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4396
4397                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4398                     (!idev || idev->cnf.accept_ra != 2) &&
4399                     fib6_info_hold_safe(rt)) {
4400                         rcu_read_unlock();
4401                         ip6_del_rt(net, rt, false);
4402                         goto restart;
4403                 }
4404         }
4405         rcu_read_unlock();
4406
4407         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4408 }
4409
4410 void rt6_purge_dflt_routers(struct net *net)
4411 {
4412         struct fib6_table *table;
4413         struct hlist_head *head;
4414         unsigned int h;
4415
4416         rcu_read_lock();
4417
4418         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4419                 head = &net->ipv6.fib_table_hash[h];
4420                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4421                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4422                                 __rt6_purge_dflt_routers(net, table);
4423                 }
4424         }
4425
4426         rcu_read_unlock();
4427 }
4428
4429 static void rtmsg_to_fib6_config(struct net *net,
4430                                  struct in6_rtmsg *rtmsg,
4431                                  struct fib6_config *cfg)
4432 {
4433         *cfg = (struct fib6_config){
4434                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4435                          : RT6_TABLE_MAIN,
4436                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4437                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4438                 .fc_expires = rtmsg->rtmsg_info,
4439                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4440                 .fc_src_len = rtmsg->rtmsg_src_len,
4441                 .fc_flags = rtmsg->rtmsg_flags,
4442                 .fc_type = rtmsg->rtmsg_type,
4443
4444                 .fc_nlinfo.nl_net = net,
4445
4446                 .fc_dst = rtmsg->rtmsg_dst,
4447                 .fc_src = rtmsg->rtmsg_src,
4448                 .fc_gateway = rtmsg->rtmsg_gateway,
4449         };
4450 }
4451
4452 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4453 {
4454         struct fib6_config cfg;
4455         int err;
4456
4457         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4458                 return -EINVAL;
4459         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4460                 return -EPERM;
4461
4462         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4463
4464         rtnl_lock();
4465         switch (cmd) {
4466         case SIOCADDRT:
4467                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4468                 break;
4469         case SIOCDELRT:
4470                 err = ip6_route_del(&cfg, NULL);
4471                 break;
4472         }
4473         rtnl_unlock();
4474         return err;
4475 }
4476
4477 /*
4478  *      Drop the packet on the floor
4479  */
4480
4481 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4482 {
4483         struct dst_entry *dst = skb_dst(skb);
4484         struct net *net = dev_net(dst->dev);
4485         struct inet6_dev *idev;
4486         SKB_DR(reason);
4487         int type;
4488
4489         if (netif_is_l3_master(skb->dev) ||
4490             dst->dev == net->loopback_dev)
4491                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4492         else
4493                 idev = ip6_dst_idev(dst);
4494
4495         switch (ipstats_mib_noroutes) {
4496         case IPSTATS_MIB_INNOROUTES:
4497                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4498                 if (type == IPV6_ADDR_ANY) {
4499                         SKB_DR_SET(reason, IP_INADDRERRORS);
4500                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4501                         break;
4502                 }
4503                 SKB_DR_SET(reason, IP_INNOROUTES);
4504                 fallthrough;
4505         case IPSTATS_MIB_OUTNOROUTES:
4506                 SKB_DR_OR(reason, IP_OUTNOROUTES);
4507                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4508                 break;
4509         }
4510
4511         /* Start over by dropping the dst for l3mdev case */
4512         if (netif_is_l3_master(skb->dev))
4513                 skb_dst_drop(skb);
4514
4515         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4516         kfree_skb_reason(skb, reason);
4517         return 0;
4518 }
4519
4520 static int ip6_pkt_discard(struct sk_buff *skb)
4521 {
4522         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4523 }
4524
4525 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4526 {
4527         skb->dev = skb_dst(skb)->dev;
4528         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4529 }
4530
4531 static int ip6_pkt_prohibit(struct sk_buff *skb)
4532 {
4533         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4534 }
4535
4536 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4537 {
4538         skb->dev = skb_dst(skb)->dev;
4539         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4540 }
4541
4542 /*
4543  *      Allocate a dst for local (unicast / anycast) address.
4544  */
4545
4546 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4547                                      struct inet6_dev *idev,
4548                                      const struct in6_addr *addr,
4549                                      bool anycast, gfp_t gfp_flags)
4550 {
4551         struct fib6_config cfg = {
4552                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4553                 .fc_ifindex = idev->dev->ifindex,
4554                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4555                 .fc_dst = *addr,
4556                 .fc_dst_len = 128,
4557                 .fc_protocol = RTPROT_KERNEL,
4558                 .fc_nlinfo.nl_net = net,
4559                 .fc_ignore_dev_down = true,
4560         };
4561         struct fib6_info *f6i;
4562
4563         if (anycast) {
4564                 cfg.fc_type = RTN_ANYCAST;
4565                 cfg.fc_flags |= RTF_ANYCAST;
4566         } else {
4567                 cfg.fc_type = RTN_LOCAL;
4568                 cfg.fc_flags |= RTF_LOCAL;
4569         }
4570
4571         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4572         if (!IS_ERR(f6i))
4573                 f6i->dst_nocount = true;
4574         return f6i;
4575 }
4576
4577 /* remove deleted ip from prefsrc entries */
4578 struct arg_dev_net_ip {
4579         struct net_device *dev;
4580         struct net *net;
4581         struct in6_addr *addr;
4582 };
4583
4584 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4585 {
4586         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4587         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4588         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4589
4590         if (!rt->nh &&
4591             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4592             rt != net->ipv6.fib6_null_entry &&
4593             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4594                 spin_lock_bh(&rt6_exception_lock);
4595                 /* remove prefsrc entry */
4596                 rt->fib6_prefsrc.plen = 0;
4597                 spin_unlock_bh(&rt6_exception_lock);
4598         }
4599         return 0;
4600 }
4601
4602 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4603 {
4604         struct net *net = dev_net(ifp->idev->dev);
4605         struct arg_dev_net_ip adni = {
4606                 .dev = ifp->idev->dev,
4607                 .net = net,
4608                 .addr = &ifp->addr,
4609         };
4610         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4611 }
4612
4613 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4614
4615 /* Remove routers and update dst entries when gateway turn into host. */
4616 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4617 {
4618         struct in6_addr *gateway = (struct in6_addr *)arg;
4619         struct fib6_nh *nh;
4620
4621         /* RA routes do not use nexthops */
4622         if (rt->nh)
4623                 return 0;
4624
4625         nh = rt->fib6_nh;
4626         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4627             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4628                 return -1;
4629
4630         /* Further clean up cached routes in exception table.
4631          * This is needed because cached route may have a different
4632          * gateway than its 'parent' in the case of an ip redirect.
4633          */
4634         fib6_nh_exceptions_clean_tohost(nh, gateway);
4635
4636         return 0;
4637 }
4638
4639 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4640 {
4641         fib6_clean_all(net, fib6_clean_tohost, gateway);
4642 }
4643
4644 struct arg_netdev_event {
4645         const struct net_device *dev;
4646         union {
4647                 unsigned char nh_flags;
4648                 unsigned long event;
4649         };
4650 };
4651
4652 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4653 {
4654         struct fib6_info *iter;
4655         struct fib6_node *fn;
4656
4657         fn = rcu_dereference_protected(rt->fib6_node,
4658                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4659         iter = rcu_dereference_protected(fn->leaf,
4660                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4661         while (iter) {
4662                 if (iter->fib6_metric == rt->fib6_metric &&
4663                     rt6_qualify_for_ecmp(iter))
4664                         return iter;
4665                 iter = rcu_dereference_protected(iter->fib6_next,
4666                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4667         }
4668
4669         return NULL;
4670 }
4671
4672 /* only called for fib entries with builtin fib6_nh */
4673 static bool rt6_is_dead(const struct fib6_info *rt)
4674 {
4675         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4676             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4677              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4678                 return true;
4679
4680         return false;
4681 }
4682
4683 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4684 {
4685         struct fib6_info *iter;
4686         int total = 0;
4687
4688         if (!rt6_is_dead(rt))
4689                 total += rt->fib6_nh->fib_nh_weight;
4690
4691         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4692                 if (!rt6_is_dead(iter))
4693                         total += iter->fib6_nh->fib_nh_weight;
4694         }
4695
4696         return total;
4697 }
4698
4699 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4700 {
4701         int upper_bound = -1;
4702
4703         if (!rt6_is_dead(rt)) {
4704                 *weight += rt->fib6_nh->fib_nh_weight;
4705                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4706                                                     total) - 1;
4707         }
4708         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4709 }
4710
4711 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4712 {
4713         struct fib6_info *iter;
4714         int weight = 0;
4715
4716         rt6_upper_bound_set(rt, &weight, total);
4717
4718         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4719                 rt6_upper_bound_set(iter, &weight, total);
4720 }
4721
4722 void rt6_multipath_rebalance(struct fib6_info *rt)
4723 {
4724         struct fib6_info *first;
4725         int total;
4726
4727         /* In case the entire multipath route was marked for flushing,
4728          * then there is no need to rebalance upon the removal of every
4729          * sibling route.
4730          */
4731         if (!rt->fib6_nsiblings || rt->should_flush)
4732                 return;
4733
4734         /* During lookup routes are evaluated in order, so we need to
4735          * make sure upper bounds are assigned from the first sibling
4736          * onwards.
4737          */
4738         first = rt6_multipath_first_sibling(rt);
4739         if (WARN_ON_ONCE(!first))
4740                 return;
4741
4742         total = rt6_multipath_total_weight(first);
4743         rt6_multipath_upper_bound_set(first, total);
4744 }
4745
4746 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4747 {
4748         const struct arg_netdev_event *arg = p_arg;
4749         struct net *net = dev_net(arg->dev);
4750
4751         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4752             rt->fib6_nh->fib_nh_dev == arg->dev) {
4753                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4754                 fib6_update_sernum_upto_root(net, rt);
4755                 rt6_multipath_rebalance(rt);
4756         }
4757
4758         return 0;
4759 }
4760
4761 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4762 {
4763         struct arg_netdev_event arg = {
4764                 .dev = dev,
4765                 {
4766                         .nh_flags = nh_flags,
4767                 },
4768         };
4769
4770         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4771                 arg.nh_flags |= RTNH_F_LINKDOWN;
4772
4773         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4774 }
4775
4776 /* only called for fib entries with inline fib6_nh */
4777 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4778                                    const struct net_device *dev)
4779 {
4780         struct fib6_info *iter;
4781
4782         if (rt->fib6_nh->fib_nh_dev == dev)
4783                 return true;
4784         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4785                 if (iter->fib6_nh->fib_nh_dev == dev)
4786                         return true;
4787
4788         return false;
4789 }
4790
4791 static void rt6_multipath_flush(struct fib6_info *rt)
4792 {
4793         struct fib6_info *iter;
4794
4795         rt->should_flush = 1;
4796         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4797                 iter->should_flush = 1;
4798 }
4799
4800 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4801                                              const struct net_device *down_dev)
4802 {
4803         struct fib6_info *iter;
4804         unsigned int dead = 0;
4805
4806         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4807             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4808                 dead++;
4809         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4810                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4811                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4812                         dead++;
4813
4814         return dead;
4815 }
4816
4817 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4818                                        const struct net_device *dev,
4819                                        unsigned char nh_flags)
4820 {
4821         struct fib6_info *iter;
4822
4823         if (rt->fib6_nh->fib_nh_dev == dev)
4824                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4825         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4826                 if (iter->fib6_nh->fib_nh_dev == dev)
4827                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4828 }
4829
4830 /* called with write lock held for table with rt */
4831 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4832 {
4833         const struct arg_netdev_event *arg = p_arg;
4834         const struct net_device *dev = arg->dev;
4835         struct net *net = dev_net(dev);
4836
4837         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4838                 return 0;
4839
4840         switch (arg->event) {
4841         case NETDEV_UNREGISTER:
4842                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4843         case NETDEV_DOWN:
4844                 if (rt->should_flush)
4845                         return -1;
4846                 if (!rt->fib6_nsiblings)
4847                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4848                 if (rt6_multipath_uses_dev(rt, dev)) {
4849                         unsigned int count;
4850
4851                         count = rt6_multipath_dead_count(rt, dev);
4852                         if (rt->fib6_nsiblings + 1 == count) {
4853                                 rt6_multipath_flush(rt);
4854                                 return -1;
4855                         }
4856                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4857                                                    RTNH_F_LINKDOWN);
4858                         fib6_update_sernum(net, rt);
4859                         rt6_multipath_rebalance(rt);
4860                 }
4861                 return -2;
4862         case NETDEV_CHANGE:
4863                 if (rt->fib6_nh->fib_nh_dev != dev ||
4864                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4865                         break;
4866                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4867                 rt6_multipath_rebalance(rt);
4868                 break;
4869         }
4870
4871         return 0;
4872 }
4873
4874 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4875 {
4876         struct arg_netdev_event arg = {
4877                 .dev = dev,
4878                 {
4879                         .event = event,
4880                 },
4881         };
4882         struct net *net = dev_net(dev);
4883
4884         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4885                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4886         else
4887                 fib6_clean_all(net, fib6_ifdown, &arg);
4888 }
4889
4890 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4891 {
4892         rt6_sync_down_dev(dev, event);
4893         rt6_uncached_list_flush_dev(dev);
4894         neigh_ifdown(&nd_tbl, dev);
4895 }
4896
4897 struct rt6_mtu_change_arg {
4898         struct net_device *dev;
4899         unsigned int mtu;
4900         struct fib6_info *f6i;
4901 };
4902
4903 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4904 {
4905         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4906         struct fib6_info *f6i = arg->f6i;
4907
4908         /* For administrative MTU increase, there is no way to discover
4909          * IPv6 PMTU increase, so PMTU increase should be updated here.
4910          * Since RFC 1981 doesn't include administrative MTU increase
4911          * update PMTU increase is a MUST. (i.e. jumbo frame)
4912          */
4913         if (nh->fib_nh_dev == arg->dev) {
4914                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4915                 u32 mtu = f6i->fib6_pmtu;
4916
4917                 if (mtu >= arg->mtu ||
4918                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4919                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4920
4921                 spin_lock_bh(&rt6_exception_lock);
4922                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4923                 spin_unlock_bh(&rt6_exception_lock);
4924         }
4925
4926         return 0;
4927 }
4928
4929 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4930 {
4931         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4932         struct inet6_dev *idev;
4933
4934         /* In IPv6 pmtu discovery is not optional,
4935            so that RTAX_MTU lock cannot disable it.
4936            We still use this lock to block changes
4937            caused by addrconf/ndisc.
4938         */
4939
4940         idev = __in6_dev_get(arg->dev);
4941         if (!idev)
4942                 return 0;
4943
4944         if (fib6_metric_locked(f6i, RTAX_MTU))
4945                 return 0;
4946
4947         arg->f6i = f6i;
4948         if (f6i->nh) {
4949                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4950                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4951                                                 arg);
4952         }
4953
4954         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4955 }
4956
4957 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4958 {
4959         struct rt6_mtu_change_arg arg = {
4960                 .dev = dev,
4961                 .mtu = mtu,
4962         };
4963
4964         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4965 }
4966
4967 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4968         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4969         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4970         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4971         [RTA_OIF]               = { .type = NLA_U32 },
4972         [RTA_IIF]               = { .type = NLA_U32 },
4973         [RTA_PRIORITY]          = { .type = NLA_U32 },
4974         [RTA_METRICS]           = { .type = NLA_NESTED },
4975         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4976         [RTA_PREF]              = { .type = NLA_U8 },
4977         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4978         [RTA_ENCAP]             = { .type = NLA_NESTED },
4979         [RTA_EXPIRES]           = { .type = NLA_U32 },
4980         [RTA_UID]               = { .type = NLA_U32 },
4981         [RTA_MARK]              = { .type = NLA_U32 },
4982         [RTA_TABLE]             = { .type = NLA_U32 },
4983         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4984         [RTA_SPORT]             = { .type = NLA_U16 },
4985         [RTA_DPORT]             = { .type = NLA_U16 },
4986         [RTA_NH_ID]             = { .type = NLA_U32 },
4987 };
4988
4989 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4990                               struct fib6_config *cfg,
4991                               struct netlink_ext_ack *extack)
4992 {
4993         struct rtmsg *rtm;
4994         struct nlattr *tb[RTA_MAX+1];
4995         unsigned int pref;
4996         int err;
4997
4998         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4999                                      rtm_ipv6_policy, extack);
5000         if (err < 0)
5001                 goto errout;
5002
5003         err = -EINVAL;
5004         rtm = nlmsg_data(nlh);
5005
5006         if (rtm->rtm_tos) {
5007                 NL_SET_ERR_MSG(extack,
5008                                "Invalid dsfield (tos): option not available for IPv6");
5009                 goto errout;
5010         }
5011
5012         *cfg = (struct fib6_config){
5013                 .fc_table = rtm->rtm_table,
5014                 .fc_dst_len = rtm->rtm_dst_len,
5015                 .fc_src_len = rtm->rtm_src_len,
5016                 .fc_flags = RTF_UP,
5017                 .fc_protocol = rtm->rtm_protocol,
5018                 .fc_type = rtm->rtm_type,
5019
5020                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5021                 .fc_nlinfo.nlh = nlh,
5022                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5023         };
5024
5025         if (rtm->rtm_type == RTN_UNREACHABLE ||
5026             rtm->rtm_type == RTN_BLACKHOLE ||
5027             rtm->rtm_type == RTN_PROHIBIT ||
5028             rtm->rtm_type == RTN_THROW)
5029                 cfg->fc_flags |= RTF_REJECT;
5030
5031         if (rtm->rtm_type == RTN_LOCAL)
5032                 cfg->fc_flags |= RTF_LOCAL;
5033
5034         if (rtm->rtm_flags & RTM_F_CLONED)
5035                 cfg->fc_flags |= RTF_CACHE;
5036
5037         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5038
5039         if (tb[RTA_NH_ID]) {
5040                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5041                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5042                         NL_SET_ERR_MSG(extack,
5043                                        "Nexthop specification and nexthop id are mutually exclusive");
5044                         goto errout;
5045                 }
5046                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5047         }
5048
5049         if (tb[RTA_GATEWAY]) {
5050                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5051                 cfg->fc_flags |= RTF_GATEWAY;
5052         }
5053         if (tb[RTA_VIA]) {
5054                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5055                 goto errout;
5056         }
5057
5058         if (tb[RTA_DST]) {
5059                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5060
5061                 if (nla_len(tb[RTA_DST]) < plen)
5062                         goto errout;
5063
5064                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5065         }
5066
5067         if (tb[RTA_SRC]) {
5068                 int plen = (rtm->rtm_src_len + 7) >> 3;
5069
5070                 if (nla_len(tb[RTA_SRC]) < plen)
5071                         goto errout;
5072
5073                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5074         }
5075
5076         if (tb[RTA_PREFSRC])
5077                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5078
5079         if (tb[RTA_OIF])
5080                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5081
5082         if (tb[RTA_PRIORITY])
5083                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5084
5085         if (tb[RTA_METRICS]) {
5086                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5087                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5088         }
5089
5090         if (tb[RTA_TABLE])
5091                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5092
5093         if (tb[RTA_MULTIPATH]) {
5094                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5095                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5096
5097                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5098                                                      cfg->fc_mp_len, extack);
5099                 if (err < 0)
5100                         goto errout;
5101         }
5102
5103         if (tb[RTA_PREF]) {
5104                 pref = nla_get_u8(tb[RTA_PREF]);
5105                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5106                     pref != ICMPV6_ROUTER_PREF_HIGH)
5107                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5108                 cfg->fc_flags |= RTF_PREF(pref);
5109         }
5110
5111         if (tb[RTA_ENCAP])
5112                 cfg->fc_encap = tb[RTA_ENCAP];
5113
5114         if (tb[RTA_ENCAP_TYPE]) {
5115                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5116
5117                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5118                 if (err < 0)
5119                         goto errout;
5120         }
5121
5122         if (tb[RTA_EXPIRES]) {
5123                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5124
5125                 if (addrconf_finite_timeout(timeout)) {
5126                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5127                         cfg->fc_flags |= RTF_EXPIRES;
5128                 }
5129         }
5130
5131         err = 0;
5132 errout:
5133         return err;
5134 }
5135
5136 struct rt6_nh {
5137         struct fib6_info *fib6_info;
5138         struct fib6_config r_cfg;
5139         struct list_head next;
5140 };
5141
5142 static int ip6_route_info_append(struct net *net,
5143                                  struct list_head *rt6_nh_list,
5144                                  struct fib6_info *rt,
5145                                  struct fib6_config *r_cfg)
5146 {
5147         struct rt6_nh *nh;
5148         int err = -EEXIST;
5149
5150         list_for_each_entry(nh, rt6_nh_list, next) {
5151                 /* check if fib6_info already exists */
5152                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5153                         return err;
5154         }
5155
5156         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5157         if (!nh)
5158                 return -ENOMEM;
5159         nh->fib6_info = rt;
5160         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5161         list_add_tail(&nh->next, rt6_nh_list);
5162
5163         return 0;
5164 }
5165
5166 static void ip6_route_mpath_notify(struct fib6_info *rt,
5167                                    struct fib6_info *rt_last,
5168                                    struct nl_info *info,
5169                                    __u16 nlflags)
5170 {
5171         /* if this is an APPEND route, then rt points to the first route
5172          * inserted and rt_last points to last route inserted. Userspace
5173          * wants a consistent dump of the route which starts at the first
5174          * nexthop. Since sibling routes are always added at the end of
5175          * the list, find the first sibling of the last route appended
5176          */
5177         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5178                 rt = list_first_entry(&rt_last->fib6_siblings,
5179                                       struct fib6_info,
5180                                       fib6_siblings);
5181         }
5182
5183         if (rt)
5184                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5185 }
5186
5187 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5188 {
5189         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5190         bool should_notify = false;
5191         struct fib6_info *leaf;
5192         struct fib6_node *fn;
5193
5194         rcu_read_lock();
5195         fn = rcu_dereference(rt->fib6_node);
5196         if (!fn)
5197                 goto out;
5198
5199         leaf = rcu_dereference(fn->leaf);
5200         if (!leaf)
5201                 goto out;
5202
5203         if (rt == leaf ||
5204             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5205              rt6_qualify_for_ecmp(leaf)))
5206                 should_notify = true;
5207 out:
5208         rcu_read_unlock();
5209
5210         return should_notify;
5211 }
5212
5213 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5214                              struct netlink_ext_ack *extack)
5215 {
5216         if (nla_len(nla) < sizeof(*gw)) {
5217                 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5218                 return -EINVAL;
5219         }
5220
5221         *gw = nla_get_in6_addr(nla);
5222
5223         return 0;
5224 }
5225
5226 static int ip6_route_multipath_add(struct fib6_config *cfg,
5227                                    struct netlink_ext_ack *extack)
5228 {
5229         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5230         struct nl_info *info = &cfg->fc_nlinfo;
5231         struct fib6_config r_cfg;
5232         struct rtnexthop *rtnh;
5233         struct fib6_info *rt;
5234         struct rt6_nh *err_nh;
5235         struct rt6_nh *nh, *nh_safe;
5236         __u16 nlflags;
5237         int remaining;
5238         int attrlen;
5239         int err = 1;
5240         int nhn = 0;
5241         int replace = (cfg->fc_nlinfo.nlh &&
5242                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5243         LIST_HEAD(rt6_nh_list);
5244
5245         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5246         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5247                 nlflags |= NLM_F_APPEND;
5248
5249         remaining = cfg->fc_mp_len;
5250         rtnh = (struct rtnexthop *)cfg->fc_mp;
5251
5252         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5253          * fib6_info structs per nexthop
5254          */
5255         while (rtnh_ok(rtnh, remaining)) {
5256                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5257                 if (rtnh->rtnh_ifindex)
5258                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5259
5260                 attrlen = rtnh_attrlen(rtnh);
5261                 if (attrlen > 0) {
5262                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5263
5264                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5265                         if (nla) {
5266                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5267                                                         extack);
5268                                 if (err)
5269                                         goto cleanup;
5270
5271                                 r_cfg.fc_flags |= RTF_GATEWAY;
5272                         }
5273                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5274
5275                         /* RTA_ENCAP_TYPE length checked in
5276                          * lwtunnel_valid_encap_type_attr
5277                          */
5278                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5279                         if (nla)
5280                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5281                 }
5282
5283                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5284                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5285                 if (IS_ERR(rt)) {
5286                         err = PTR_ERR(rt);
5287                         rt = NULL;
5288                         goto cleanup;
5289                 }
5290                 if (!rt6_qualify_for_ecmp(rt)) {
5291                         err = -EINVAL;
5292                         NL_SET_ERR_MSG(extack,
5293                                        "Device only routes can not be added for IPv6 using the multipath API.");
5294                         fib6_info_release(rt);
5295                         goto cleanup;
5296                 }
5297
5298                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5299
5300                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5301                                             rt, &r_cfg);
5302                 if (err) {
5303                         fib6_info_release(rt);
5304                         goto cleanup;
5305                 }
5306
5307                 rtnh = rtnh_next(rtnh, &remaining);
5308         }
5309
5310         if (list_empty(&rt6_nh_list)) {
5311                 NL_SET_ERR_MSG(extack,
5312                                "Invalid nexthop configuration - no valid nexthops");
5313                 return -EINVAL;
5314         }
5315
5316         /* for add and replace send one notification with all nexthops.
5317          * Skip the notification in fib6_add_rt2node and send one with
5318          * the full route when done
5319          */
5320         info->skip_notify = 1;
5321
5322         /* For add and replace, send one notification with all nexthops. For
5323          * append, send one notification with all appended nexthops.
5324          */
5325         info->skip_notify_kernel = 1;
5326
5327         err_nh = NULL;
5328         list_for_each_entry(nh, &rt6_nh_list, next) {
5329                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5330                 fib6_info_release(nh->fib6_info);
5331
5332                 if (!err) {
5333                         /* save reference to last route successfully inserted */
5334                         rt_last = nh->fib6_info;
5335
5336                         /* save reference to first route for notification */
5337                         if (!rt_notif)
5338                                 rt_notif = nh->fib6_info;
5339                 }
5340
5341                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5342                 nh->fib6_info = NULL;
5343                 if (err) {
5344                         if (replace && nhn)
5345                                 NL_SET_ERR_MSG_MOD(extack,
5346                                                    "multipath route replace failed (check consistency of installed routes)");
5347                         err_nh = nh;
5348                         goto add_errout;
5349                 }
5350
5351                 /* Because each route is added like a single route we remove
5352                  * these flags after the first nexthop: if there is a collision,
5353                  * we have already failed to add the first nexthop:
5354                  * fib6_add_rt2node() has rejected it; when replacing, old
5355                  * nexthops have been replaced by first new, the rest should
5356                  * be added to it.
5357                  */
5358                 if (cfg->fc_nlinfo.nlh) {
5359                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5360                                                              NLM_F_REPLACE);
5361                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5362                 }
5363                 nhn++;
5364         }
5365
5366         /* An in-kernel notification should only be sent in case the new
5367          * multipath route is added as the first route in the node, or if
5368          * it was appended to it. We pass 'rt_notif' since it is the first
5369          * sibling and might allow us to skip some checks in the replace case.
5370          */
5371         if (ip6_route_mpath_should_notify(rt_notif)) {
5372                 enum fib_event_type fib_event;
5373
5374                 if (rt_notif->fib6_nsiblings != nhn - 1)
5375                         fib_event = FIB_EVENT_ENTRY_APPEND;
5376                 else
5377                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5378
5379                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5380                                                           fib_event, rt_notif,
5381                                                           nhn - 1, extack);
5382                 if (err) {
5383                         /* Delete all the siblings that were just added */
5384                         err_nh = NULL;
5385                         goto add_errout;
5386                 }
5387         }
5388
5389         /* success ... tell user about new route */
5390         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5391         goto cleanup;
5392
5393 add_errout:
5394         /* send notification for routes that were added so that
5395          * the delete notifications sent by ip6_route_del are
5396          * coherent
5397          */
5398         if (rt_notif)
5399                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5400
5401         /* Delete routes that were already added */
5402         list_for_each_entry(nh, &rt6_nh_list, next) {
5403                 if (err_nh == nh)
5404                         break;
5405                 ip6_route_del(&nh->r_cfg, extack);
5406         }
5407
5408 cleanup:
5409         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5410                 if (nh->fib6_info)
5411                         fib6_info_release(nh->fib6_info);
5412                 list_del(&nh->next);
5413                 kfree(nh);
5414         }
5415
5416         return err;
5417 }
5418
5419 static int ip6_route_multipath_del(struct fib6_config *cfg,
5420                                    struct netlink_ext_ack *extack)
5421 {
5422         struct fib6_config r_cfg;
5423         struct rtnexthop *rtnh;
5424         int last_err = 0;
5425         int remaining;
5426         int attrlen;
5427         int err;
5428
5429         remaining = cfg->fc_mp_len;
5430         rtnh = (struct rtnexthop *)cfg->fc_mp;
5431
5432         /* Parse a Multipath Entry */
5433         while (rtnh_ok(rtnh, remaining)) {
5434                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5435                 if (rtnh->rtnh_ifindex)
5436                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5437
5438                 attrlen = rtnh_attrlen(rtnh);
5439                 if (attrlen > 0) {
5440                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5441
5442                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5443                         if (nla) {
5444                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5445                                                         extack);
5446                                 if (err) {
5447                                         last_err = err;
5448                                         goto next_rtnh;
5449                                 }
5450
5451                                 r_cfg.fc_flags |= RTF_GATEWAY;
5452                         }
5453                 }
5454                 err = ip6_route_del(&r_cfg, extack);
5455                 if (err)
5456                         last_err = err;
5457
5458 next_rtnh:
5459                 rtnh = rtnh_next(rtnh, &remaining);
5460         }
5461
5462         return last_err;
5463 }
5464
5465 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5466                               struct netlink_ext_ack *extack)
5467 {
5468         struct fib6_config cfg;
5469         int err;
5470
5471         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5472         if (err < 0)
5473                 return err;
5474
5475         if (cfg.fc_nh_id &&
5476             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5477                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5478                 return -EINVAL;
5479         }
5480
5481         if (cfg.fc_mp)
5482                 return ip6_route_multipath_del(&cfg, extack);
5483         else {
5484                 cfg.fc_delete_all_nh = 1;
5485                 return ip6_route_del(&cfg, extack);
5486         }
5487 }
5488
5489 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5490                               struct netlink_ext_ack *extack)
5491 {
5492         struct fib6_config cfg;
5493         int err;
5494
5495         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5496         if (err < 0)
5497                 return err;
5498
5499         if (cfg.fc_metric == 0)
5500                 cfg.fc_metric = IP6_RT_PRIO_USER;
5501
5502         if (cfg.fc_mp)
5503                 return ip6_route_multipath_add(&cfg, extack);
5504         else
5505                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5506 }
5507
5508 /* add the overhead of this fib6_nh to nexthop_len */
5509 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5510 {
5511         int *nexthop_len = arg;
5512
5513         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5514                      + NLA_ALIGN(sizeof(struct rtnexthop))
5515                      + nla_total_size(16); /* RTA_GATEWAY */
5516
5517         if (nh->fib_nh_lws) {
5518                 /* RTA_ENCAP_TYPE */
5519                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5520                 /* RTA_ENCAP */
5521                 *nexthop_len += nla_total_size(2);
5522         }
5523
5524         return 0;
5525 }
5526
5527 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5528 {
5529         int nexthop_len;
5530
5531         if (f6i->nh) {
5532                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5533                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5534                                          &nexthop_len);
5535         } else {
5536                 struct fib6_nh *nh = f6i->fib6_nh;
5537
5538                 nexthop_len = 0;
5539                 if (f6i->fib6_nsiblings) {
5540                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5541                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5542                                     + nla_total_size(16) /* RTA_GATEWAY */
5543                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5544
5545                         nexthop_len *= f6i->fib6_nsiblings;
5546                 }
5547                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5548         }
5549
5550         return NLMSG_ALIGN(sizeof(struct rtmsg))
5551                + nla_total_size(16) /* RTA_SRC */
5552                + nla_total_size(16) /* RTA_DST */
5553                + nla_total_size(16) /* RTA_GATEWAY */
5554                + nla_total_size(16) /* RTA_PREFSRC */
5555                + nla_total_size(4) /* RTA_TABLE */
5556                + nla_total_size(4) /* RTA_IIF */
5557                + nla_total_size(4) /* RTA_OIF */
5558                + nla_total_size(4) /* RTA_PRIORITY */
5559                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5560                + nla_total_size(sizeof(struct rta_cacheinfo))
5561                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5562                + nla_total_size(1) /* RTA_PREF */
5563                + nexthop_len;
5564 }
5565
5566 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5567                                  unsigned char *flags)
5568 {
5569         if (nexthop_is_multipath(nh)) {
5570                 struct nlattr *mp;
5571
5572                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5573                 if (!mp)
5574                         goto nla_put_failure;
5575
5576                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5577                         goto nla_put_failure;
5578
5579                 nla_nest_end(skb, mp);
5580         } else {
5581                 struct fib6_nh *fib6_nh;
5582
5583                 fib6_nh = nexthop_fib6_nh(nh);
5584                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5585                                      flags, false) < 0)
5586                         goto nla_put_failure;
5587         }
5588
5589         return 0;
5590
5591 nla_put_failure:
5592         return -EMSGSIZE;
5593 }
5594
5595 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5596                          struct fib6_info *rt, struct dst_entry *dst,
5597                          struct in6_addr *dest, struct in6_addr *src,
5598                          int iif, int type, u32 portid, u32 seq,
5599                          unsigned int flags)
5600 {
5601         struct rt6_info *rt6 = (struct rt6_info *)dst;
5602         struct rt6key *rt6_dst, *rt6_src;
5603         u32 *pmetrics, table, rt6_flags;
5604         unsigned char nh_flags = 0;
5605         struct nlmsghdr *nlh;
5606         struct rtmsg *rtm;
5607         long expires = 0;
5608
5609         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5610         if (!nlh)
5611                 return -EMSGSIZE;
5612
5613         if (rt6) {
5614                 rt6_dst = &rt6->rt6i_dst;
5615                 rt6_src = &rt6->rt6i_src;
5616                 rt6_flags = rt6->rt6i_flags;
5617         } else {
5618                 rt6_dst = &rt->fib6_dst;
5619                 rt6_src = &rt->fib6_src;
5620                 rt6_flags = rt->fib6_flags;
5621         }
5622
5623         rtm = nlmsg_data(nlh);
5624         rtm->rtm_family = AF_INET6;
5625         rtm->rtm_dst_len = rt6_dst->plen;
5626         rtm->rtm_src_len = rt6_src->plen;
5627         rtm->rtm_tos = 0;
5628         if (rt->fib6_table)
5629                 table = rt->fib6_table->tb6_id;
5630         else
5631                 table = RT6_TABLE_UNSPEC;
5632         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5633         if (nla_put_u32(skb, RTA_TABLE, table))
5634                 goto nla_put_failure;
5635
5636         rtm->rtm_type = rt->fib6_type;
5637         rtm->rtm_flags = 0;
5638         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5639         rtm->rtm_protocol = rt->fib6_protocol;
5640
5641         if (rt6_flags & RTF_CACHE)
5642                 rtm->rtm_flags |= RTM_F_CLONED;
5643
5644         if (dest) {
5645                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5646                         goto nla_put_failure;
5647                 rtm->rtm_dst_len = 128;
5648         } else if (rtm->rtm_dst_len)
5649                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5650                         goto nla_put_failure;
5651 #ifdef CONFIG_IPV6_SUBTREES
5652         if (src) {
5653                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5654                         goto nla_put_failure;
5655                 rtm->rtm_src_len = 128;
5656         } else if (rtm->rtm_src_len &&
5657                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5658                 goto nla_put_failure;
5659 #endif
5660         if (iif) {
5661 #ifdef CONFIG_IPV6_MROUTE
5662                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5663                         int err = ip6mr_get_route(net, skb, rtm, portid);
5664
5665                         if (err == 0)
5666                                 return 0;
5667                         if (err < 0)
5668                                 goto nla_put_failure;
5669                 } else
5670 #endif
5671                         if (nla_put_u32(skb, RTA_IIF, iif))
5672                                 goto nla_put_failure;
5673         } else if (dest) {
5674                 struct in6_addr saddr_buf;
5675                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5676                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5677                         goto nla_put_failure;
5678         }
5679
5680         if (rt->fib6_prefsrc.plen) {
5681                 struct in6_addr saddr_buf;
5682                 saddr_buf = rt->fib6_prefsrc.addr;
5683                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5684                         goto nla_put_failure;
5685         }
5686
5687         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5688         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5689                 goto nla_put_failure;
5690
5691         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5692                 goto nla_put_failure;
5693
5694         /* For multipath routes, walk the siblings list and add
5695          * each as a nexthop within RTA_MULTIPATH.
5696          */
5697         if (rt6) {
5698                 if (rt6_flags & RTF_GATEWAY &&
5699                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5700                         goto nla_put_failure;
5701
5702                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5703                         goto nla_put_failure;
5704
5705                 if (dst->lwtstate &&
5706                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5707                         goto nla_put_failure;
5708         } else if (rt->fib6_nsiblings) {
5709                 struct fib6_info *sibling, *next_sibling;
5710                 struct nlattr *mp;
5711
5712                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5713                 if (!mp)
5714                         goto nla_put_failure;
5715
5716                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5717                                     rt->fib6_nh->fib_nh_weight, AF_INET6,
5718                                     0) < 0)
5719                         goto nla_put_failure;
5720
5721                 list_for_each_entry_safe(sibling, next_sibling,
5722                                          &rt->fib6_siblings, fib6_siblings) {
5723                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5724                                             sibling->fib6_nh->fib_nh_weight,
5725                                             AF_INET6, 0) < 0)
5726                                 goto nla_put_failure;
5727                 }
5728
5729                 nla_nest_end(skb, mp);
5730         } else if (rt->nh) {
5731                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5732                         goto nla_put_failure;
5733
5734                 if (nexthop_is_blackhole(rt->nh))
5735                         rtm->rtm_type = RTN_BLACKHOLE;
5736
5737                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5738                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5739                         goto nla_put_failure;
5740
5741                 rtm->rtm_flags |= nh_flags;
5742         } else {
5743                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5744                                      &nh_flags, false) < 0)
5745                         goto nla_put_failure;
5746
5747                 rtm->rtm_flags |= nh_flags;
5748         }
5749
5750         if (rt6_flags & RTF_EXPIRES) {
5751                 expires = dst ? dst->expires : rt->expires;
5752                 expires -= jiffies;
5753         }
5754
5755         if (!dst) {
5756                 if (READ_ONCE(rt->offload))
5757                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5758                 if (READ_ONCE(rt->trap))
5759                         rtm->rtm_flags |= RTM_F_TRAP;
5760                 if (READ_ONCE(rt->offload_failed))
5761                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5762         }
5763
5764         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5765                 goto nla_put_failure;
5766
5767         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5768                 goto nla_put_failure;
5769
5770
5771         nlmsg_end(skb, nlh);
5772         return 0;
5773
5774 nla_put_failure:
5775         nlmsg_cancel(skb, nlh);
5776         return -EMSGSIZE;
5777 }
5778
5779 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5780 {
5781         const struct net_device *dev = arg;
5782
5783         if (nh->fib_nh_dev == dev)
5784                 return 1;
5785
5786         return 0;
5787 }
5788
5789 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5790                                const struct net_device *dev)
5791 {
5792         if (f6i->nh) {
5793                 struct net_device *_dev = (struct net_device *)dev;
5794
5795                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5796                                                   fib6_info_nh_uses_dev,
5797                                                   _dev);
5798         }
5799
5800         if (f6i->fib6_nh->fib_nh_dev == dev)
5801                 return true;
5802
5803         if (f6i->fib6_nsiblings) {
5804                 struct fib6_info *sibling, *next_sibling;
5805
5806                 list_for_each_entry_safe(sibling, next_sibling,
5807                                          &f6i->fib6_siblings, fib6_siblings) {
5808                         if (sibling->fib6_nh->fib_nh_dev == dev)
5809                                 return true;
5810                 }
5811         }
5812
5813         return false;
5814 }
5815
5816 struct fib6_nh_exception_dump_walker {
5817         struct rt6_rtnl_dump_arg *dump;
5818         struct fib6_info *rt;
5819         unsigned int flags;
5820         unsigned int skip;
5821         unsigned int count;
5822 };
5823
5824 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5825 {
5826         struct fib6_nh_exception_dump_walker *w = arg;
5827         struct rt6_rtnl_dump_arg *dump = w->dump;
5828         struct rt6_exception_bucket *bucket;
5829         struct rt6_exception *rt6_ex;
5830         int i, err;
5831
5832         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5833         if (!bucket)
5834                 return 0;
5835
5836         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5837                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5838                         if (w->skip) {
5839                                 w->skip--;
5840                                 continue;
5841                         }
5842
5843                         /* Expiration of entries doesn't bump sernum, insertion
5844                          * does. Removal is triggered by insertion, so we can
5845                          * rely on the fact that if entries change between two
5846                          * partial dumps, this node is scanned again completely,
5847                          * see rt6_insert_exception() and fib6_dump_table().
5848                          *
5849                          * Count expired entries we go through as handled
5850                          * entries that we'll skip next time, in case of partial
5851                          * node dump. Otherwise, if entries expire meanwhile,
5852                          * we'll skip the wrong amount.
5853                          */
5854                         if (rt6_check_expired(rt6_ex->rt6i)) {
5855                                 w->count++;
5856                                 continue;
5857                         }
5858
5859                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5860                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5861                                             RTM_NEWROUTE,
5862                                             NETLINK_CB(dump->cb->skb).portid,
5863                                             dump->cb->nlh->nlmsg_seq, w->flags);
5864                         if (err)
5865                                 return err;
5866
5867                         w->count++;
5868                 }
5869                 bucket++;
5870         }
5871
5872         return 0;
5873 }
5874
5875 /* Return -1 if done with node, number of handled routes on partial dump */
5876 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5877 {
5878         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5879         struct fib_dump_filter *filter = &arg->filter;
5880         unsigned int flags = NLM_F_MULTI;
5881         struct net *net = arg->net;
5882         int count = 0;
5883
5884         if (rt == net->ipv6.fib6_null_entry)
5885                 return -1;
5886
5887         if ((filter->flags & RTM_F_PREFIX) &&
5888             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5889                 /* success since this is not a prefix route */
5890                 return -1;
5891         }
5892         if (filter->filter_set &&
5893             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5894              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5895              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5896                 return -1;
5897         }
5898
5899         if (filter->filter_set ||
5900             !filter->dump_routes || !filter->dump_exceptions) {
5901                 flags |= NLM_F_DUMP_FILTERED;
5902         }
5903
5904         if (filter->dump_routes) {
5905                 if (skip) {
5906                         skip--;
5907                 } else {
5908                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5909                                           0, RTM_NEWROUTE,
5910                                           NETLINK_CB(arg->cb->skb).portid,
5911                                           arg->cb->nlh->nlmsg_seq, flags)) {
5912                                 return 0;
5913                         }
5914                         count++;
5915                 }
5916         }
5917
5918         if (filter->dump_exceptions) {
5919                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5920                                                            .rt = rt,
5921                                                            .flags = flags,
5922                                                            .skip = skip,
5923                                                            .count = 0 };
5924                 int err;
5925
5926                 rcu_read_lock();
5927                 if (rt->nh) {
5928                         err = nexthop_for_each_fib6_nh(rt->nh,
5929                                                        rt6_nh_dump_exceptions,
5930                                                        &w);
5931                 } else {
5932                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5933                 }
5934                 rcu_read_unlock();
5935
5936                 if (err)
5937                         return count += w.count;
5938         }
5939
5940         return -1;
5941 }
5942
5943 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5944                                         const struct nlmsghdr *nlh,
5945                                         struct nlattr **tb,
5946                                         struct netlink_ext_ack *extack)
5947 {
5948         struct rtmsg *rtm;
5949         int i, err;
5950
5951         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5952                 NL_SET_ERR_MSG_MOD(extack,
5953                                    "Invalid header for get route request");
5954                 return -EINVAL;
5955         }
5956
5957         if (!netlink_strict_get_check(skb))
5958                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5959                                               rtm_ipv6_policy, extack);
5960
5961         rtm = nlmsg_data(nlh);
5962         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5963             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5964             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5965             rtm->rtm_type) {
5966                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5967                 return -EINVAL;
5968         }
5969         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5970                 NL_SET_ERR_MSG_MOD(extack,
5971                                    "Invalid flags for get route request");
5972                 return -EINVAL;
5973         }
5974
5975         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5976                                             rtm_ipv6_policy, extack);
5977         if (err)
5978                 return err;
5979
5980         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5981             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5982                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5983                 return -EINVAL;
5984         }
5985
5986         for (i = 0; i <= RTA_MAX; i++) {
5987                 if (!tb[i])
5988                         continue;
5989
5990                 switch (i) {
5991                 case RTA_SRC:
5992                 case RTA_DST:
5993                 case RTA_IIF:
5994                 case RTA_OIF:
5995                 case RTA_MARK:
5996                 case RTA_UID:
5997                 case RTA_SPORT:
5998                 case RTA_DPORT:
5999                 case RTA_IP_PROTO:
6000                         break;
6001                 default:
6002                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6003                         return -EINVAL;
6004                 }
6005         }
6006
6007         return 0;
6008 }
6009
6010 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6011                               struct netlink_ext_ack *extack)
6012 {
6013         struct net *net = sock_net(in_skb->sk);
6014         struct nlattr *tb[RTA_MAX+1];
6015         int err, iif = 0, oif = 0;
6016         struct fib6_info *from;
6017         struct dst_entry *dst;
6018         struct rt6_info *rt;
6019         struct sk_buff *skb;
6020         struct rtmsg *rtm;
6021         struct flowi6 fl6 = {};
6022         bool fibmatch;
6023
6024         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6025         if (err < 0)
6026                 goto errout;
6027
6028         err = -EINVAL;
6029         rtm = nlmsg_data(nlh);
6030         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6031         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6032
6033         if (tb[RTA_SRC]) {
6034                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6035                         goto errout;
6036
6037                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6038         }
6039
6040         if (tb[RTA_DST]) {
6041                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6042                         goto errout;
6043
6044                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6045         }
6046
6047         if (tb[RTA_IIF])
6048                 iif = nla_get_u32(tb[RTA_IIF]);
6049
6050         if (tb[RTA_OIF])
6051                 oif = nla_get_u32(tb[RTA_OIF]);
6052
6053         if (tb[RTA_MARK])
6054                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6055
6056         if (tb[RTA_UID])
6057                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6058                                            nla_get_u32(tb[RTA_UID]));
6059         else
6060                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6061
6062         if (tb[RTA_SPORT])
6063                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6064
6065         if (tb[RTA_DPORT])
6066                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6067
6068         if (tb[RTA_IP_PROTO]) {
6069                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6070                                                   &fl6.flowi6_proto, AF_INET6,
6071                                                   extack);
6072                 if (err)
6073                         goto errout;
6074         }
6075
6076         if (iif) {
6077                 struct net_device *dev;
6078                 int flags = 0;
6079
6080                 rcu_read_lock();
6081
6082                 dev = dev_get_by_index_rcu(net, iif);
6083                 if (!dev) {
6084                         rcu_read_unlock();
6085                         err = -ENODEV;
6086                         goto errout;
6087                 }
6088
6089                 fl6.flowi6_iif = iif;
6090
6091                 if (!ipv6_addr_any(&fl6.saddr))
6092                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6093
6094                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6095
6096                 rcu_read_unlock();
6097         } else {
6098                 fl6.flowi6_oif = oif;
6099
6100                 dst = ip6_route_output(net, NULL, &fl6);
6101         }
6102
6103
6104         rt = container_of(dst, struct rt6_info, dst);
6105         if (rt->dst.error) {
6106                 err = rt->dst.error;
6107                 ip6_rt_put(rt);
6108                 goto errout;
6109         }
6110
6111         if (rt == net->ipv6.ip6_null_entry) {
6112                 err = rt->dst.error;
6113                 ip6_rt_put(rt);
6114                 goto errout;
6115         }
6116
6117         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6118         if (!skb) {
6119                 ip6_rt_put(rt);
6120                 err = -ENOBUFS;
6121                 goto errout;
6122         }
6123
6124         skb_dst_set(skb, &rt->dst);
6125
6126         rcu_read_lock();
6127         from = rcu_dereference(rt->from);
6128         if (from) {
6129                 if (fibmatch)
6130                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6131                                             iif, RTM_NEWROUTE,
6132                                             NETLINK_CB(in_skb).portid,
6133                                             nlh->nlmsg_seq, 0);
6134                 else
6135                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6136                                             &fl6.saddr, iif, RTM_NEWROUTE,
6137                                             NETLINK_CB(in_skb).portid,
6138                                             nlh->nlmsg_seq, 0);
6139         } else {
6140                 err = -ENETUNREACH;
6141         }
6142         rcu_read_unlock();
6143
6144         if (err < 0) {
6145                 kfree_skb(skb);
6146                 goto errout;
6147         }
6148
6149         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6150 errout:
6151         return err;
6152 }
6153
6154 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6155                      unsigned int nlm_flags)
6156 {
6157         struct sk_buff *skb;
6158         struct net *net = info->nl_net;
6159         u32 seq;
6160         int err;
6161
6162         err = -ENOBUFS;
6163         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6164
6165         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6166         if (!skb)
6167                 goto errout;
6168
6169         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6170                             event, info->portid, seq, nlm_flags);
6171         if (err < 0) {
6172                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6173                 WARN_ON(err == -EMSGSIZE);
6174                 kfree_skb(skb);
6175                 goto errout;
6176         }
6177         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6178                     info->nlh, gfp_any());
6179         return;
6180 errout:
6181         if (err < 0)
6182                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6183 }
6184
6185 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6186                     struct nl_info *info)
6187 {
6188         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6189         struct sk_buff *skb;
6190         int err = -ENOBUFS;
6191
6192         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6193         if (!skb)
6194                 goto errout;
6195
6196         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6197                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6198         if (err < 0) {
6199                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6200                 WARN_ON(err == -EMSGSIZE);
6201                 kfree_skb(skb);
6202                 goto errout;
6203         }
6204         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6205                     info->nlh, gfp_any());
6206         return;
6207 errout:
6208         if (err < 0)
6209                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6210 }
6211
6212 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6213                             bool offload, bool trap, bool offload_failed)
6214 {
6215         struct sk_buff *skb;
6216         int err;
6217
6218         if (READ_ONCE(f6i->offload) == offload &&
6219             READ_ONCE(f6i->trap) == trap &&
6220             READ_ONCE(f6i->offload_failed) == offload_failed)
6221                 return;
6222
6223         WRITE_ONCE(f6i->offload, offload);
6224         WRITE_ONCE(f6i->trap, trap);
6225
6226         /* 2 means send notifications only if offload_failed was changed. */
6227         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6228             READ_ONCE(f6i->offload_failed) == offload_failed)
6229                 return;
6230
6231         WRITE_ONCE(f6i->offload_failed, offload_failed);
6232
6233         if (!rcu_access_pointer(f6i->fib6_node))
6234                 /* The route was removed from the tree, do not send
6235                  * notification.
6236                  */
6237                 return;
6238
6239         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6240                 return;
6241
6242         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6243         if (!skb) {
6244                 err = -ENOBUFS;
6245                 goto errout;
6246         }
6247
6248         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6249                             0, 0);
6250         if (err < 0) {
6251                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6252                 WARN_ON(err == -EMSGSIZE);
6253                 kfree_skb(skb);
6254                 goto errout;
6255         }
6256
6257         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6258         return;
6259
6260 errout:
6261         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6262 }
6263 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6264
6265 static int ip6_route_dev_notify(struct notifier_block *this,
6266                                 unsigned long event, void *ptr)
6267 {
6268         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6269         struct net *net = dev_net(dev);
6270
6271         if (!(dev->flags & IFF_LOOPBACK))
6272                 return NOTIFY_OK;
6273
6274         if (event == NETDEV_REGISTER) {
6275                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6276                 net->ipv6.ip6_null_entry->dst.dev = dev;
6277                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6278 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6279                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6280                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6281                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6282                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6283 #endif
6284          } else if (event == NETDEV_UNREGISTER &&
6285                     dev->reg_state != NETREG_UNREGISTERED) {
6286                 /* NETDEV_UNREGISTER could be fired for multiple times by
6287                  * netdev_wait_allrefs(). Make sure we only call this once.
6288                  */
6289                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6290 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6291                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6292                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6293 #endif
6294         }
6295
6296         return NOTIFY_OK;
6297 }
6298
6299 /*
6300  *      /proc
6301  */
6302
6303 #ifdef CONFIG_PROC_FS
6304 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6305 {
6306         struct net *net = (struct net *)seq->private;
6307         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6308                    net->ipv6.rt6_stats->fib_nodes,
6309                    net->ipv6.rt6_stats->fib_route_nodes,
6310                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6311                    net->ipv6.rt6_stats->fib_rt_entries,
6312                    net->ipv6.rt6_stats->fib_rt_cache,
6313                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6314                    net->ipv6.rt6_stats->fib_discarded_routes);
6315
6316         return 0;
6317 }
6318 #endif  /* CONFIG_PROC_FS */
6319
6320 #ifdef CONFIG_SYSCTL
6321
6322 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6323                               void *buffer, size_t *lenp, loff_t *ppos)
6324 {
6325         struct net *net;
6326         int delay;
6327         int ret;
6328         if (!write)
6329                 return -EINVAL;
6330
6331         net = (struct net *)ctl->extra1;
6332         delay = net->ipv6.sysctl.flush_delay;
6333         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6334         if (ret)
6335                 return ret;
6336
6337         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6338         return 0;
6339 }
6340
6341 static struct ctl_table ipv6_route_table_template[] = {
6342         {
6343                 .procname       =       "max_size",
6344                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6345                 .maxlen         =       sizeof(int),
6346                 .mode           =       0644,
6347                 .proc_handler   =       proc_dointvec,
6348         },
6349         {
6350                 .procname       =       "gc_thresh",
6351                 .data           =       &ip6_dst_ops_template.gc_thresh,
6352                 .maxlen         =       sizeof(int),
6353                 .mode           =       0644,
6354                 .proc_handler   =       proc_dointvec,
6355         },
6356         {
6357                 .procname       =       "flush",
6358                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6359                 .maxlen         =       sizeof(int),
6360                 .mode           =       0200,
6361                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6362         },
6363         {
6364                 .procname       =       "gc_min_interval",
6365                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6366                 .maxlen         =       sizeof(int),
6367                 .mode           =       0644,
6368                 .proc_handler   =       proc_dointvec_jiffies,
6369         },
6370         {
6371                 .procname       =       "gc_timeout",
6372                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6373                 .maxlen         =       sizeof(int),
6374                 .mode           =       0644,
6375                 .proc_handler   =       proc_dointvec_jiffies,
6376         },
6377         {
6378                 .procname       =       "gc_interval",
6379                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6380                 .maxlen         =       sizeof(int),
6381                 .mode           =       0644,
6382                 .proc_handler   =       proc_dointvec_jiffies,
6383         },
6384         {
6385                 .procname       =       "gc_elasticity",
6386                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6387                 .maxlen         =       sizeof(int),
6388                 .mode           =       0644,
6389                 .proc_handler   =       proc_dointvec,
6390         },
6391         {
6392                 .procname       =       "mtu_expires",
6393                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6394                 .maxlen         =       sizeof(int),
6395                 .mode           =       0644,
6396                 .proc_handler   =       proc_dointvec_jiffies,
6397         },
6398         {
6399                 .procname       =       "min_adv_mss",
6400                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6401                 .maxlen         =       sizeof(int),
6402                 .mode           =       0644,
6403                 .proc_handler   =       proc_dointvec,
6404         },
6405         {
6406                 .procname       =       "gc_min_interval_ms",
6407                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6408                 .maxlen         =       sizeof(int),
6409                 .mode           =       0644,
6410                 .proc_handler   =       proc_dointvec_ms_jiffies,
6411         },
6412         {
6413                 .procname       =       "skip_notify_on_dev_down",
6414                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6415                 .maxlen         =       sizeof(int),
6416                 .mode           =       0644,
6417                 .proc_handler   =       proc_dointvec_minmax,
6418                 .extra1         =       SYSCTL_ZERO,
6419                 .extra2         =       SYSCTL_ONE,
6420         },
6421         { }
6422 };
6423
6424 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6425 {
6426         struct ctl_table *table;
6427
6428         table = kmemdup(ipv6_route_table_template,
6429                         sizeof(ipv6_route_table_template),
6430                         GFP_KERNEL);
6431
6432         if (table) {
6433                 table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6434                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6435                 table[2].data = &net->ipv6.sysctl.flush_delay;
6436                 table[2].extra1 = net;
6437                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6438                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6439                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6440                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6441                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6442                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6443                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6444                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6445
6446                 /* Don't export sysctls to unprivileged users */
6447                 if (net->user_ns != &init_user_ns)
6448                         table[1].procname = NULL;
6449         }
6450
6451         return table;
6452 }
6453 #endif
6454
6455 static int __net_init ip6_route_net_init(struct net *net)
6456 {
6457         int ret = -ENOMEM;
6458
6459         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6460                sizeof(net->ipv6.ip6_dst_ops));
6461
6462         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6463                 goto out_ip6_dst_ops;
6464
6465         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6466         if (!net->ipv6.fib6_null_entry)
6467                 goto out_ip6_dst_entries;
6468         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6469                sizeof(*net->ipv6.fib6_null_entry));
6470
6471         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6472                                            sizeof(*net->ipv6.ip6_null_entry),
6473                                            GFP_KERNEL);
6474         if (!net->ipv6.ip6_null_entry)
6475                 goto out_fib6_null_entry;
6476         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6477         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6478                          ip6_template_metrics, true);
6479         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6480
6481 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6482         net->ipv6.fib6_has_custom_rules = false;
6483         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6484                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6485                                                GFP_KERNEL);
6486         if (!net->ipv6.ip6_prohibit_entry)
6487                 goto out_ip6_null_entry;
6488         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6489         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6490                          ip6_template_metrics, true);
6491         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6492
6493         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6494                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6495                                                GFP_KERNEL);
6496         if (!net->ipv6.ip6_blk_hole_entry)
6497                 goto out_ip6_prohibit_entry;
6498         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6499         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6500                          ip6_template_metrics, true);
6501         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6502 #ifdef CONFIG_IPV6_SUBTREES
6503         net->ipv6.fib6_routes_require_src = 0;
6504 #endif
6505 #endif
6506
6507         net->ipv6.sysctl.flush_delay = 0;
6508         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6509         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6510         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6511         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6512         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6513         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6514         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6515         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6516
6517         atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6518
6519         ret = 0;
6520 out:
6521         return ret;
6522
6523 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6524 out_ip6_prohibit_entry:
6525         kfree(net->ipv6.ip6_prohibit_entry);
6526 out_ip6_null_entry:
6527         kfree(net->ipv6.ip6_null_entry);
6528 #endif
6529 out_fib6_null_entry:
6530         kfree(net->ipv6.fib6_null_entry);
6531 out_ip6_dst_entries:
6532         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6533 out_ip6_dst_ops:
6534         goto out;
6535 }
6536
6537 static void __net_exit ip6_route_net_exit(struct net *net)
6538 {
6539         kfree(net->ipv6.fib6_null_entry);
6540         kfree(net->ipv6.ip6_null_entry);
6541 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6542         kfree(net->ipv6.ip6_prohibit_entry);
6543         kfree(net->ipv6.ip6_blk_hole_entry);
6544 #endif
6545         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6546 }
6547
6548 static int __net_init ip6_route_net_init_late(struct net *net)
6549 {
6550 #ifdef CONFIG_PROC_FS
6551         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6552                         sizeof(struct ipv6_route_iter));
6553         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6554                         rt6_stats_seq_show, NULL);
6555 #endif
6556         return 0;
6557 }
6558
6559 static void __net_exit ip6_route_net_exit_late(struct net *net)
6560 {
6561 #ifdef CONFIG_PROC_FS
6562         remove_proc_entry("ipv6_route", net->proc_net);
6563         remove_proc_entry("rt6_stats", net->proc_net);
6564 #endif
6565 }
6566
6567 static struct pernet_operations ip6_route_net_ops = {
6568         .init = ip6_route_net_init,
6569         .exit = ip6_route_net_exit,
6570 };
6571
6572 static int __net_init ipv6_inetpeer_init(struct net *net)
6573 {
6574         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6575
6576         if (!bp)
6577                 return -ENOMEM;
6578         inet_peer_base_init(bp);
6579         net->ipv6.peers = bp;
6580         return 0;
6581 }
6582
6583 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6584 {
6585         struct inet_peer_base *bp = net->ipv6.peers;
6586
6587         net->ipv6.peers = NULL;
6588         inetpeer_invalidate_tree(bp);
6589         kfree(bp);
6590 }
6591
6592 static struct pernet_operations ipv6_inetpeer_ops = {
6593         .init   =       ipv6_inetpeer_init,
6594         .exit   =       ipv6_inetpeer_exit,
6595 };
6596
6597 static struct pernet_operations ip6_route_net_late_ops = {
6598         .init = ip6_route_net_init_late,
6599         .exit = ip6_route_net_exit_late,
6600 };
6601
6602 static struct notifier_block ip6_route_dev_notifier = {
6603         .notifier_call = ip6_route_dev_notify,
6604         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6605 };
6606
6607 void __init ip6_route_init_special_entries(void)
6608 {
6609         /* Registering of the loopback is done before this portion of code,
6610          * the loopback reference in rt6_info will not be taken, do it
6611          * manually for init_net */
6612         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6613         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6614         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6615   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6616         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6617         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6618         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6619         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6620   #endif
6621 }
6622
6623 #if IS_BUILTIN(CONFIG_IPV6)
6624 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6625 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6626
6627 BTF_ID_LIST(btf_fib6_info_id)
6628 BTF_ID(struct, fib6_info)
6629
6630 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6631         .seq_ops                = &ipv6_route_seq_ops,
6632         .init_seq_private       = bpf_iter_init_seq_net,
6633         .fini_seq_private       = bpf_iter_fini_seq_net,
6634         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6635 };
6636
6637 static struct bpf_iter_reg ipv6_route_reg_info = {
6638         .target                 = "ipv6_route",
6639         .ctx_arg_info_size      = 1,
6640         .ctx_arg_info           = {
6641                 { offsetof(struct bpf_iter__ipv6_route, rt),
6642                   PTR_TO_BTF_ID_OR_NULL },
6643         },
6644         .seq_info               = &ipv6_route_seq_info,
6645 };
6646
6647 static int __init bpf_iter_register(void)
6648 {
6649         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6650         return bpf_iter_reg_target(&ipv6_route_reg_info);
6651 }
6652
6653 static void bpf_iter_unregister(void)
6654 {
6655         bpf_iter_unreg_target(&ipv6_route_reg_info);
6656 }
6657 #endif
6658 #endif
6659
6660 int __init ip6_route_init(void)
6661 {
6662         int ret;
6663         int cpu;
6664
6665         ret = -ENOMEM;
6666         ip6_dst_ops_template.kmem_cachep =
6667                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6668                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6669         if (!ip6_dst_ops_template.kmem_cachep)
6670                 goto out;
6671
6672         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6673         if (ret)
6674                 goto out_kmem_cache;
6675
6676         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6677         if (ret)
6678                 goto out_dst_entries;
6679
6680         ret = register_pernet_subsys(&ip6_route_net_ops);
6681         if (ret)
6682                 goto out_register_inetpeer;
6683
6684         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6685
6686         ret = fib6_init();
6687         if (ret)
6688                 goto out_register_subsys;
6689
6690         ret = xfrm6_init();
6691         if (ret)
6692                 goto out_fib6_init;
6693
6694         ret = fib6_rules_init();
6695         if (ret)
6696                 goto xfrm6_init;
6697
6698         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6699         if (ret)
6700                 goto fib6_rules_init;
6701
6702         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6703                                    inet6_rtm_newroute, NULL, 0);
6704         if (ret < 0)
6705                 goto out_register_late_subsys;
6706
6707         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6708                                    inet6_rtm_delroute, NULL, 0);
6709         if (ret < 0)
6710                 goto out_register_late_subsys;
6711
6712         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6713                                    inet6_rtm_getroute, NULL,
6714                                    RTNL_FLAG_DOIT_UNLOCKED);
6715         if (ret < 0)
6716                 goto out_register_late_subsys;
6717
6718         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6719         if (ret)
6720                 goto out_register_late_subsys;
6721
6722 #if IS_BUILTIN(CONFIG_IPV6)
6723 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6724         ret = bpf_iter_register();
6725         if (ret)
6726                 goto out_register_late_subsys;
6727 #endif
6728 #endif
6729
6730         for_each_possible_cpu(cpu) {
6731                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6732
6733                 INIT_LIST_HEAD(&ul->head);
6734                 INIT_LIST_HEAD(&ul->quarantine);
6735                 spin_lock_init(&ul->lock);
6736         }
6737
6738 out:
6739         return ret;
6740
6741 out_register_late_subsys:
6742         rtnl_unregister_all(PF_INET6);
6743         unregister_pernet_subsys(&ip6_route_net_late_ops);
6744 fib6_rules_init:
6745         fib6_rules_cleanup();
6746 xfrm6_init:
6747         xfrm6_fini();
6748 out_fib6_init:
6749         fib6_gc_cleanup();
6750 out_register_subsys:
6751         unregister_pernet_subsys(&ip6_route_net_ops);
6752 out_register_inetpeer:
6753         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6754 out_dst_entries:
6755         dst_entries_destroy(&ip6_dst_blackhole_ops);
6756 out_kmem_cache:
6757         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6758         goto out;
6759 }
6760
6761 void ip6_route_cleanup(void)
6762 {
6763 #if IS_BUILTIN(CONFIG_IPV6)
6764 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6765         bpf_iter_unregister();
6766 #endif
6767 #endif
6768         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6769         unregister_pernet_subsys(&ip6_route_net_late_ops);
6770         fib6_rules_cleanup();
6771         xfrm6_fini();
6772         fib6_gc_cleanup();
6773         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6774         unregister_pernet_subsys(&ip6_route_net_ops);
6775         dst_entries_destroy(&ip6_dst_blackhole_ops);
6776         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6777 }