ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <linux/siphash.h>
  45 #include <net/net_namespace.h>
  46 #include <net/snmp.h>
  47 #include <net/ipv6.h>
  48 #include <net/ip6_fib.h>
  49 #include <net/ip6_route.h>
  50 #include <net/ndisc.h>
  51 #include <net/addrconf.h>
  52 #include <net/tcp.h>
  53 #include <linux/rtnetlink.h>
  54 #include <net/dst.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/xfrm.h>
  57 #include <net/netevent.h>
  58 #include <net/netlink.h>
  59 #include <net/rtnh.h>
  60 #include <net/lwtunnel.h>
  61 #include <net/ip_tunnels.h>
  62 #include <net/l3mdev.h>
  63 #include <net/ip.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/btf_ids.h>
  66
  67 #ifdef CONFIG_SYSCTL
  68 #include <linux/sysctl.h>
  69 #endif
  70
  71 static int ip6_rt_type_to_error(u8 fib6_type);
  72
  73 #define CREATE_TRACE_POINTS
  74 #include <trace/events/fib6.h>
  75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  76 #undef CREATE_TRACE_POINTS
  77
  78 enum rt6_nud_state {
  79         RT6_NUD_FAIL_HARD = -3,
  80         RT6_NUD_FAIL_PROBE = -2,
  81         RT6_NUD_FAIL_DO_RR = -1,
  82         RT6_NUD_SUCCEED = 1
  83 };
  84
  85 INDIRECT_CALLABLE_SCOPE
  86 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  87 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  88 INDIRECT_CALLABLE_SCOPE
  89 unsigned int            ip6_mtu(const struct dst_entry *dst);
  90 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  91 static void             ip6_dst_destroy(struct dst_entry *);
  92 static void             ip6_dst_ifdown(struct dst_entry *,
  93                                        struct net_device *dev, int how);
  94 static void              ip6_dst_gc(struct dst_ops *ops);
  95
  96 static int              ip6_pkt_discard(struct sk_buff *skb);
  97 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  98 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  99 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
 100 static void             ip6_link_failure(struct sk_buff *skb);
 101 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 102                                            struct sk_buff *skb, u32 mtu,
 103                                            bool confirm_neigh);
 104 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 105                                         struct sk_buff *skb);
 106 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 107                            int strict);
 108 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 109 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 110                          struct fib6_info *rt, struct dst_entry *dst,
 111                          struct in6_addr *dest, struct in6_addr *src,
 112                          int iif, int type, u32 portid, u32 seq,
 113                          unsigned int flags);
 114 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 115                                            const struct in6_addr *daddr,
 116                                            const struct in6_addr *saddr);
 117
 118 #ifdef CONFIG_IPV6_ROUTE_INFO
 119 static struct fib6_info *rt6_add_route_info(struct net *net,
 120                                            const struct in6_addr *prefix, int prefixlen,
 121                                            const struct in6_addr *gwaddr,
 122                                            struct net_device *dev,
 123                                            unsigned int pref);
 124 static struct fib6_info *rt6_get_route_info(struct net *net,
 125                                            const struct in6_addr *prefix, int prefixlen,
 126                                            const struct in6_addr *gwaddr,
 127                                            struct net_device *dev);
 128 #endif
 129
 130 struct uncached_list {
 131         spinlock_t              lock;
 132         struct list_head        head;
 133         struct list_head        quarantine;
 134 };
 135
 136 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 137
 138 void rt6_uncached_list_add(struct rt6_info *rt)
 139 {
 140         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 141
 142         rt->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                                 netdev_ref_replace(rt_dev, blackhole_netdev,
 186                                                    &rt->dst.dev_tracker,
 187                                                    GFP_ATOMIC);
 188                                 handled = true;
 189                         }
 190                         if (handled)
 191                                 list_move(&rt->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         if (skb)
 429                 IP6CB(skb)->flags |= IP6SKB_MULTIPATH;
 430
 431         /* We might have already computed the hash for ICMPv6 errors. In such
 432          * case it will always be non-zero. Otherwise now is the time to do it.
 433          */
 434         if (!fl6->mp_hash &&
 435             (!match->nh || nexthop_is_multipath(match->nh)))
 436                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 437
 438         if (unlikely(match->nh)) {
 439                 nexthop_path_fib6_result(res, fl6->mp_hash);
 440                 return;
 441         }
 442
 443         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 444                 goto out;
 445
 446         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 447                                  fib6_siblings) {
 448                 const struct fib6_nh *nh = sibling->fib6_nh;
 449                 int nh_upper_bound;
 450
 451                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 452                 if (fl6->mp_hash > nh_upper_bound)
 453                         continue;
 454                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 455                         break;
 456                 match = sibling;
 457                 break;
 458         }
 459
 460 out:
 461         res->f6i = match;
 462         res->nh = match->fib6_nh;
 463 }
 464
 465 /*
 466  *      Route lookup. rcu_read_lock() should be held.
 467  */
 468
 469 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 470                                const struct in6_addr *saddr, int oif, int flags)
 471 {
 472         const struct net_device *dev;
 473
 474         if (nh->fib_nh_flags & RTNH_F_DEAD)
 475                 return false;
 476
 477         dev = nh->fib_nh_dev;
 478         if (oif) {
 479                 if (dev->ifindex == oif)
 480                         return true;
 481         } else {
 482                 if (ipv6_chk_addr(net, saddr, dev,
 483                                   flags & RT6_LOOKUP_F_IFACE))
 484                         return true;
 485         }
 486
 487         return false;
 488 }
 489
 490 struct fib6_nh_dm_arg {
 491         struct net              *net;
 492         const struct in6_addr   *saddr;
 493         int                     oif;
 494         int                     flags;
 495         struct fib6_nh          *nh;
 496 };
 497
 498 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 499 {
 500         struct fib6_nh_dm_arg *arg = _arg;
 501
 502         arg->nh = nh;
 503         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 504                                   arg->flags);
 505 }
 506
 507 /* returns fib6_nh from nexthop or NULL */
 508 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 509                                         struct fib6_result *res,
 510                                         const struct in6_addr *saddr,
 511                                         int oif, int flags)
 512 {
 513         struct fib6_nh_dm_arg arg = {
 514                 .net   = net,
 515                 .saddr = saddr,
 516                 .oif   = oif,
 517                 .flags = flags,
 518         };
 519
 520         if (nexthop_is_blackhole(nh))
 521                 return NULL;
 522
 523         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 524                 return arg.nh;
 525
 526         return NULL;
 527 }
 528
 529 static void rt6_device_match(struct net *net, struct fib6_result *res,
 530                              const struct in6_addr *saddr, int oif, int flags)
 531 {
 532         struct fib6_info *f6i = res->f6i;
 533         struct fib6_info *spf6i;
 534         struct fib6_nh *nh;
 535
 536         if (!oif && ipv6_addr_any(saddr)) {
 537                 if (unlikely(f6i->nh)) {
 538                         nh = nexthop_fib6_nh(f6i->nh);
 539                         if (nexthop_is_blackhole(f6i->nh))
 540                                 goto out_blackhole;
 541                 } else {
 542                         nh = f6i->fib6_nh;
 543                 }
 544                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 545                         goto out;
 546         }
 547
 548         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 549                 bool matched = false;
 550
 551                 if (unlikely(spf6i->nh)) {
 552                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 553                                               oif, flags);
 554                         if (nh)
 555                                 matched = true;
 556                 } else {
 557                         nh = spf6i->fib6_nh;
 558                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 559                                 matched = true;
 560                 }
 561                 if (matched) {
 562                         res->f6i = spf6i;
 563                         goto out;
 564                 }
 565         }
 566
 567         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 568                 res->f6i = net->ipv6.fib6_null_entry;
 569                 nh = res->f6i->fib6_nh;
 570                 goto out;
 571         }
 572
 573         if (unlikely(f6i->nh)) {
 574                 nh = nexthop_fib6_nh(f6i->nh);
 575                 if (nexthop_is_blackhole(f6i->nh))
 576                         goto out_blackhole;
 577         } else {
 578                 nh = f6i->fib6_nh;
 579         }
 580
 581         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 582                 res->f6i = net->ipv6.fib6_null_entry;
 583                 nh = res->f6i->fib6_nh;
 584         }
 585 out:
 586         res->nh = nh;
 587         res->fib6_type = res->f6i->fib6_type;
 588         res->fib6_flags = res->f6i->fib6_flags;
 589         return;
 590
 591 out_blackhole:
 592         res->fib6_flags |= RTF_REJECT;
 593         res->fib6_type = RTN_BLACKHOLE;
 594         res->nh = nh;
 595 }
 596
 597 #ifdef CONFIG_IPV6_ROUTER_PREF
 598 struct __rt6_probe_work {
 599         struct work_struct work;
 600         struct in6_addr target;
 601         struct net_device *dev;
 602         netdevice_tracker dev_tracker;
 603 };
 604
 605 static void rt6_probe_deferred(struct work_struct *w)
 606 {
 607         struct in6_addr mcaddr;
 608         struct __rt6_probe_work *work =
 609                 container_of(w, struct __rt6_probe_work, work);
 610
 611         addrconf_addr_solict_mult(&work->target, &mcaddr);
 612         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 613         netdev_put(work->dev, &work->dev_tracker);
 614         kfree(work);
 615 }
 616
 617 static void rt6_probe(struct fib6_nh *fib6_nh)
 618 {
 619         struct __rt6_probe_work *work = NULL;
 620         const struct in6_addr *nh_gw;
 621         unsigned long last_probe;
 622         struct neighbour *neigh;
 623         struct net_device *dev;
 624         struct inet6_dev *idev;
 625
 626         /*
 627          * Okay, this does not seem to be appropriate
 628          * for now, however, we need to check if it
 629          * is really so; aka Router Reachability Probing.
 630          *
 631          * Router Reachability Probe MUST be rate-limited
 632          * to no more than one per minute.
 633          */
 634         if (!fib6_nh->fib_nh_gw_family)
 635                 return;
 636
 637         nh_gw = &fib6_nh->fib_nh_gw6;
 638         dev = fib6_nh->fib_nh_dev;
 639         rcu_read_lock();
 640         last_probe = READ_ONCE(fib6_nh->last_probe);
 641         idev = __in6_dev_get(dev);
 642         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 643         if (neigh) {
 644                 if (READ_ONCE(neigh->nud_state) & NUD_VALID)
 645                         goto out;
 646
 647                 write_lock_bh(&neigh->lock);
 648                 if (!(neigh->nud_state & NUD_VALID) &&
 649                     time_after(jiffies,
 650                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 651                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 652                         if (work)
 653                                 __neigh_set_probe_once(neigh);
 654                 }
 655                 write_unlock_bh(&neigh->lock);
 656         } else if (time_after(jiffies, last_probe +
 657                                        idev->cnf.rtr_probe_interval)) {
 658                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 659         }
 660
 661         if (!work || cmpxchg(&fib6_nh->last_probe,
 662                              last_probe, jiffies) != last_probe) {
 663                 kfree(work);
 664         } else {
 665                 INIT_WORK(&work->work, rt6_probe_deferred);
 666                 work->target = *nh_gw;
 667                 netdev_hold(dev, &work->dev_tracker, GFP_ATOMIC);
 668                 work->dev = dev;
 669                 schedule_work(&work->work);
 670         }
 671
 672 out:
 673         rcu_read_unlock();
 674 }
 675 #else
 676 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 677 {
 678 }
 679 #endif
 680
 681 /*
 682  * Default Router Selection (RFC 2461 6.3.6)
 683  */
 684 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 685 {
 686         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 687         struct neighbour *neigh;
 688
 689         rcu_read_lock();
 690         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 691                                           &fib6_nh->fib_nh_gw6);
 692         if (neigh) {
 693                 u8 nud_state = READ_ONCE(neigh->nud_state);
 694
 695                 if (nud_state & NUD_VALID)
 696                         ret = RT6_NUD_SUCCEED;
 697 #ifdef CONFIG_IPV6_ROUTER_PREF
 698                 else if (!(nud_state & NUD_FAILED))
 699                         ret = RT6_NUD_SUCCEED;
 700                 else
 701                         ret = RT6_NUD_FAIL_PROBE;
 702 #endif
 703         } else {
 704                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 705                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 706         }
 707         rcu_read_unlock();
 708
 709         return ret;
 710 }
 711
 712 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 713                            int strict)
 714 {
 715         int m = 0;
 716
 717         if (!oif || nh->fib_nh_dev->ifindex == oif)
 718                 m = 2;
 719
 720         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 721                 return RT6_NUD_FAIL_HARD;
 722 #ifdef CONFIG_IPV6_ROUTER_PREF
 723         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 724 #endif
 725         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 726             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 727                 int n = rt6_check_neigh(nh);
 728                 if (n < 0)
 729                         return n;
 730         }
 731         return m;
 732 }
 733
 734 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 735                        int oif, int strict, int *mpri, bool *do_rr)
 736 {
 737         bool match_do_rr = false;
 738         bool rc = false;
 739         int m;
 740
 741         if (nh->fib_nh_flags & RTNH_F_DEAD)
 742                 goto out;
 743
 744         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 745             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 746             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 747                 goto out;
 748
 749         m = rt6_score_route(nh, fib6_flags, oif, strict);
 750         if (m == RT6_NUD_FAIL_DO_RR) {
 751                 match_do_rr = true;
 752                 m = 0; /* lowest valid score */
 753         } else if (m == RT6_NUD_FAIL_HARD) {
 754                 goto out;
 755         }
 756
 757         if (strict & RT6_LOOKUP_F_REACHABLE)
 758                 rt6_probe(nh);
 759
 760         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 761         if (m > *mpri) {
 762                 *do_rr = match_do_rr;
 763                 *mpri = m;
 764                 rc = true;
 765         }
 766 out:
 767         return rc;
 768 }
 769
 770 struct fib6_nh_frl_arg {
 771         u32             flags;
 772         int             oif;
 773         int             strict;
 774         int             *mpri;
 775         bool            *do_rr;
 776         struct fib6_nh  *nh;
 777 };
 778
 779 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 780 {
 781         struct fib6_nh_frl_arg *arg = _arg;
 782
 783         arg->nh = nh;
 784         return find_match(nh, arg->flags, arg->oif, arg->strict,
 785                           arg->mpri, arg->do_rr);
 786 }
 787
 788 static void __find_rr_leaf(struct fib6_info *f6i_start,
 789                            struct fib6_info *nomatch, u32 metric,
 790                            struct fib6_result *res, struct fib6_info **cont,
 791                            int oif, int strict, bool *do_rr, int *mpri)
 792 {
 793         struct fib6_info *f6i;
 794
 795         for (f6i = f6i_start;
 796              f6i && f6i != nomatch;
 797              f6i = rcu_dereference(f6i->fib6_next)) {
 798                 bool matched = false;
 799                 struct fib6_nh *nh;
 800
 801                 if (cont && f6i->fib6_metric != metric) {
 802                         *cont = f6i;
 803                         return;
 804                 }
 805
 806                 if (fib6_check_expired(f6i))
 807                         continue;
 808
 809                 if (unlikely(f6i->nh)) {
 810                         struct fib6_nh_frl_arg arg = {
 811                                 .flags  = f6i->fib6_flags,
 812                                 .oif    = oif,
 813                                 .strict = strict,
 814                                 .mpri   = mpri,
 815                                 .do_rr  = do_rr
 816                         };
 817
 818                         if (nexthop_is_blackhole(f6i->nh)) {
 819                                 res->fib6_flags = RTF_REJECT;
 820                                 res->fib6_type = RTN_BLACKHOLE;
 821                                 res->f6i = f6i;
 822                                 res->nh = nexthop_fib6_nh(f6i->nh);
 823                                 return;
 824                         }
 825                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 826                                                      &arg)) {
 827                                 matched = true;
 828                                 nh = arg.nh;
 829                         }
 830                 } else {
 831                         nh = f6i->fib6_nh;
 832                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 833                                        mpri, do_rr))
 834                                 matched = true;
 835                 }
 836                 if (matched) {
 837                         res->f6i = f6i;
 838                         res->nh = nh;
 839                         res->fib6_flags = f6i->fib6_flags;
 840                         res->fib6_type = f6i->fib6_type;
 841                 }
 842         }
 843 }
 844
 845 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 846                          struct fib6_info *rr_head, int oif, int strict,
 847                          bool *do_rr, struct fib6_result *res)
 848 {
 849         u32 metric = rr_head->fib6_metric;
 850         struct fib6_info *cont = NULL;
 851         int mpri = -1;
 852
 853         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 854                        oif, strict, do_rr, &mpri);
 855
 856         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 857                        oif, strict, do_rr, &mpri);
 858
 859         if (res->f6i || !cont)
 860                 return;
 861
 862         __find_rr_leaf(cont, NULL, metric, res, NULL,
 863                        oif, strict, do_rr, &mpri);
 864 }
 865
 866 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 867                        struct fib6_result *res, int strict)
 868 {
 869         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 870         struct fib6_info *rt0;
 871         bool do_rr = false;
 872         int key_plen;
 873
 874         /* make sure this function or its helpers sets f6i */
 875         res->f6i = NULL;
 876
 877         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 878                 goto out;
 879
 880         rt0 = rcu_dereference(fn->rr_ptr);
 881         if (!rt0)
 882                 rt0 = leaf;
 883
 884         /* Double check to make sure fn is not an intermediate node
 885          * and fn->leaf does not points to its child's leaf
 886          * (This might happen if all routes under fn are deleted from
 887          * the tree and fib6_repair_tree() is called on the node.)
 888          */
 889         key_plen = rt0->fib6_dst.plen;
 890 #ifdef CONFIG_IPV6_SUBTREES
 891         if (rt0->fib6_src.plen)
 892                 key_plen = rt0->fib6_src.plen;
 893 #endif
 894         if (fn->fn_bit != key_plen)
 895                 goto out;
 896
 897         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 898         if (do_rr) {
 899                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 900
 901                 /* no entries matched; do round-robin */
 902                 if (!next || next->fib6_metric != rt0->fib6_metric)
 903                         next = leaf;
 904
 905                 if (next != rt0) {
 906                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 907                         /* make sure next is not being deleted from the tree */
 908                         if (next->fib6_node)
 909                                 rcu_assign_pointer(fn->rr_ptr, next);
 910                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 911                 }
 912         }
 913
 914 out:
 915         if (!res->f6i) {
 916                 res->f6i = net->ipv6.fib6_null_entry;
 917                 res->nh = res->f6i->fib6_nh;
 918                 res->fib6_flags = res->f6i->fib6_flags;
 919                 res->fib6_type = res->f6i->fib6_type;
 920         }
 921 }
 922
 923 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 924 {
 925         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 926                res->nh->fib_nh_gw_family;
 927 }
 928
 929 #ifdef CONFIG_IPV6_ROUTE_INFO
 930 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 931                   const struct in6_addr *gwaddr)
 932 {
 933         struct net *net = dev_net(dev);
 934         struct route_info *rinfo = (struct route_info *) opt;
 935         struct in6_addr prefix_buf, *prefix;
 936         unsigned int pref;
 937         unsigned long lifetime;
 938         struct fib6_info *rt;
 939
 940         if (len < sizeof(struct route_info)) {
 941                 return -EINVAL;
 942         }
 943
 944         /* Sanity check for prefix_len and length */
 945         if (rinfo->length > 3) {
 946                 return -EINVAL;
 947         } else if (rinfo->prefix_len > 128) {
 948                 return -EINVAL;
 949         } else if (rinfo->prefix_len > 64) {
 950                 if (rinfo->length < 2) {
 951                         return -EINVAL;
 952                 }
 953         } else if (rinfo->prefix_len > 0) {
 954                 if (rinfo->length < 1) {
 955                         return -EINVAL;
 956                 }
 957         }
 958
 959         pref = rinfo->route_pref;
 960         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 961                 return -EINVAL;
 962
 963         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 964
 965         if (rinfo->length == 3)
 966                 prefix = (struct in6_addr *)rinfo->prefix;
 967         else {
 968                 /* this function is safe */
 969                 ipv6_addr_prefix(&prefix_buf,
 970                                  (struct in6_addr *)rinfo->prefix,
 971                                  rinfo->prefix_len);
 972                 prefix = &prefix_buf;
 973         }
 974
 975         if (rinfo->prefix_len == 0)
 976                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 977         else
 978                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 979                                         gwaddr, dev);
 980
 981         if (rt && !lifetime) {
 982                 ip6_del_rt(net, rt, false);
 983                 rt = NULL;
 984         }
 985
 986         if (!rt && lifetime)
 987                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 988                                         dev, pref);
 989         else if (rt)
 990                 rt->fib6_flags = RTF_ROUTEINFO |
 991                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 992
 993         if (rt) {
 994                 if (!addrconf_finite_timeout(lifetime))
 995                         fib6_clean_expires(rt);
 996                 else
 997                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 998
 999                 fib6_info_release(rt);
1000         }
1001         return 0;
1002 }
1003 #endif
1004
1005 /*
1006  *      Misc support functions
1007  */
1008
1009 /* called with rcu_lock held */
1010 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1011 {
1012         struct net_device *dev = res->nh->fib_nh_dev;
1013
1014         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1015                 /* for copies of local routes, dst->dev needs to be the
1016                  * device if it is a master device, the master device if
1017                  * device is enslaved, and the loopback as the default
1018                  */
1019                 if (netif_is_l3_slave(dev) &&
1020                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1021                         dev = l3mdev_master_dev_rcu(dev);
1022                 else if (!netif_is_l3_master(dev))
1023                         dev = dev_net(dev)->loopback_dev;
1024                 /* last case is netif_is_l3_master(dev) is true in which
1025                  * case we want dev returned to be dev
1026                  */
1027         }
1028
1029         return dev;
1030 }
1031
1032 static const int fib6_prop[RTN_MAX + 1] = {
1033         [RTN_UNSPEC]    = 0,
1034         [RTN_UNICAST]   = 0,
1035         [RTN_LOCAL]     = 0,
1036         [RTN_BROADCAST] = 0,
1037         [RTN_ANYCAST]   = 0,
1038         [RTN_MULTICAST] = 0,
1039         [RTN_BLACKHOLE] = -EINVAL,
1040         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1041         [RTN_PROHIBIT]  = -EACCES,
1042         [RTN_THROW]     = -EAGAIN,
1043         [RTN_NAT]       = -EINVAL,
1044         [RTN_XRESOLVE]  = -EINVAL,
1045 };
1046
1047 static int ip6_rt_type_to_error(u8 fib6_type)
1048 {
1049         return fib6_prop[fib6_type];
1050 }
1051
1052 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1053 {
1054         unsigned short flags = 0;
1055
1056         if (rt->dst_nocount)
1057                 flags |= DST_NOCOUNT;
1058         if (rt->dst_nopolicy)
1059                 flags |= DST_NOPOLICY;
1060
1061         return flags;
1062 }
1063
1064 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1065 {
1066         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1067
1068         switch (fib6_type) {
1069         case RTN_BLACKHOLE:
1070                 rt->dst.output = dst_discard_out;
1071                 rt->dst.input = dst_discard;
1072                 break;
1073         case RTN_PROHIBIT:
1074                 rt->dst.output = ip6_pkt_prohibit_out;
1075                 rt->dst.input = ip6_pkt_prohibit;
1076                 break;
1077         case RTN_THROW:
1078         case RTN_UNREACHABLE:
1079         default:
1080                 rt->dst.output = ip6_pkt_discard_out;
1081                 rt->dst.input = ip6_pkt_discard;
1082                 break;
1083         }
1084 }
1085
1086 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1087 {
1088         struct fib6_info *f6i = res->f6i;
1089
1090         if (res->fib6_flags & RTF_REJECT) {
1091                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1092                 return;
1093         }
1094
1095         rt->dst.error = 0;
1096         rt->dst.output = ip6_output;
1097
1098         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1099                 rt->dst.input = ip6_input;
1100         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1101                 rt->dst.input = ip6_mc_input;
1102         } else {
1103                 rt->dst.input = ip6_forward;
1104         }
1105
1106         if (res->nh->fib_nh_lws) {
1107                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1108                 lwtunnel_set_redirect(&rt->dst);
1109         }
1110
1111         rt->dst.lastuse = jiffies;
1112 }
1113
1114 /* Caller must already hold reference to @from */
1115 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1116 {
1117         rt->rt6i_flags &= ~RTF_EXPIRES;
1118         rcu_assign_pointer(rt->from, from);
1119         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1120 }
1121
1122 /* Caller must already hold reference to f6i in result */
1123 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1124 {
1125         const struct fib6_nh *nh = res->nh;
1126         const struct net_device *dev = nh->fib_nh_dev;
1127         struct fib6_info *f6i = res->f6i;
1128
1129         ip6_rt_init_dst(rt, res);
1130
1131         rt->rt6i_dst = f6i->fib6_dst;
1132         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1133         rt->rt6i_flags = res->fib6_flags;
1134         if (nh->fib_nh_gw_family) {
1135                 rt->rt6i_gateway = nh->fib_nh_gw6;
1136                 rt->rt6i_flags |= RTF_GATEWAY;
1137         }
1138         rt6_set_from(rt, f6i);
1139 #ifdef CONFIG_IPV6_SUBTREES
1140         rt->rt6i_src = f6i->fib6_src;
1141 #endif
1142 }
1143
1144 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1145                                         struct in6_addr *saddr)
1146 {
1147         struct fib6_node *pn, *sn;
1148         while (1) {
1149                 if (fn->fn_flags & RTN_TL_ROOT)
1150                         return NULL;
1151                 pn = rcu_dereference(fn->parent);
1152                 sn = FIB6_SUBTREE(pn);
1153                 if (sn && sn != fn)
1154                         fn = fib6_node_lookup(sn, NULL, saddr);
1155                 else
1156                         fn = pn;
1157                 if (fn->fn_flags & RTN_RTINFO)
1158                         return fn;
1159         }
1160 }
1161
1162 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1163 {
1164         struct rt6_info *rt = *prt;
1165
1166         if (dst_hold_safe(&rt->dst))
1167                 return true;
1168         if (net) {
1169                 rt = net->ipv6.ip6_null_entry;
1170                 dst_hold(&rt->dst);
1171         } else {
1172                 rt = NULL;
1173         }
1174         *prt = rt;
1175         return false;
1176 }
1177
1178 /* called with rcu_lock held */
1179 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1180 {
1181         struct net_device *dev = res->nh->fib_nh_dev;
1182         struct fib6_info *f6i = res->f6i;
1183         unsigned short flags;
1184         struct rt6_info *nrt;
1185
1186         if (!fib6_info_hold_safe(f6i))
1187                 goto fallback;
1188
1189         flags = fib6_info_dst_flags(f6i);
1190         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1191         if (!nrt) {
1192                 fib6_info_release(f6i);
1193                 goto fallback;
1194         }
1195
1196         ip6_rt_copy_init(nrt, res);
1197         return nrt;
1198
1199 fallback:
1200         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1201         dst_hold(&nrt->dst);
1202         return nrt;
1203 }
1204
1205 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1206                                              struct fib6_table *table,
1207                                              struct flowi6 *fl6,
1208                                              const struct sk_buff *skb,
1209                                              int flags)
1210 {
1211         struct fib6_result res = {};
1212         struct fib6_node *fn;
1213         struct rt6_info *rt;
1214
1215         rcu_read_lock();
1216         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1217 restart:
1218         res.f6i = rcu_dereference(fn->leaf);
1219         if (!res.f6i)
1220                 res.f6i = net->ipv6.fib6_null_entry;
1221         else
1222                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1223                                  flags);
1224
1225         if (res.f6i == net->ipv6.fib6_null_entry) {
1226                 fn = fib6_backtrack(fn, &fl6->saddr);
1227                 if (fn)
1228                         goto restart;
1229
1230                 rt = net->ipv6.ip6_null_entry;
1231                 dst_hold(&rt->dst);
1232                 goto out;
1233         } else if (res.fib6_flags & RTF_REJECT) {
1234                 goto do_create;
1235         }
1236
1237         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1238                          fl6->flowi6_oif != 0, skb, flags);
1239
1240         /* Search through exception table */
1241         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1242         if (rt) {
1243                 if (ip6_hold_safe(net, &rt))
1244                         dst_use_noref(&rt->dst, jiffies);
1245         } else {
1246 do_create:
1247                 rt = ip6_create_rt_rcu(&res);
1248         }
1249
1250 out:
1251         trace_fib6_table_lookup(net, &res, table, fl6);
1252
1253         rcu_read_unlock();
1254
1255         return rt;
1256 }
1257
1258 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1259                                    const struct sk_buff *skb, int flags)
1260 {
1261         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1262 }
1263 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1264
1265 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1266                             const struct in6_addr *saddr, int oif,
1267                             const struct sk_buff *skb, int strict)
1268 {
1269         struct flowi6 fl6 = {
1270                 .flowi6_oif = oif,
1271                 .daddr = *daddr,
1272         };
1273         struct dst_entry *dst;
1274         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1275
1276         if (saddr) {
1277                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1278                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1279         }
1280
1281         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1282         if (dst->error == 0)
1283                 return (struct rt6_info *) dst;
1284
1285         dst_release(dst);
1286
1287         return NULL;
1288 }
1289 EXPORT_SYMBOL(rt6_lookup);
1290
1291 /* ip6_ins_rt is called with FREE table->tb6_lock.
1292  * It takes new route entry, the addition fails by any reason the
1293  * route is released.
1294  * Caller must hold dst before calling it.
1295  */
1296
1297 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1298                         struct netlink_ext_ack *extack)
1299 {
1300         int err;
1301         struct fib6_table *table;
1302
1303         table = rt->fib6_table;
1304         spin_lock_bh(&table->tb6_lock);
1305         err = fib6_add(&table->tb6_root, rt, info, extack);
1306         spin_unlock_bh(&table->tb6_lock);
1307
1308         return err;
1309 }
1310
1311 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1312 {
1313         struct nl_info info = { .nl_net = net, };
1314
1315         return __ip6_ins_rt(rt, &info, NULL);
1316 }
1317
1318 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1319                                            const struct in6_addr *daddr,
1320                                            const struct in6_addr *saddr)
1321 {
1322         struct fib6_info *f6i = res->f6i;
1323         struct net_device *dev;
1324         struct rt6_info *rt;
1325
1326         /*
1327          *      Clone the route.
1328          */
1329
1330         if (!fib6_info_hold_safe(f6i))
1331                 return NULL;
1332
1333         dev = ip6_rt_get_dev_rcu(res);
1334         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1335         if (!rt) {
1336                 fib6_info_release(f6i);
1337                 return NULL;
1338         }
1339
1340         ip6_rt_copy_init(rt, res);
1341         rt->rt6i_flags |= RTF_CACHE;
1342         rt->rt6i_dst.addr = *daddr;
1343         rt->rt6i_dst.plen = 128;
1344
1345         if (!rt6_is_gw_or_nonexthop(res)) {
1346                 if (f6i->fib6_dst.plen != 128 &&
1347                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1348                         rt->rt6i_flags |= RTF_ANYCAST;
1349 #ifdef CONFIG_IPV6_SUBTREES
1350                 if (rt->rt6i_src.plen && saddr) {
1351                         rt->rt6i_src.addr = *saddr;
1352                         rt->rt6i_src.plen = 128;
1353                 }
1354 #endif
1355         }
1356
1357         return rt;
1358 }
1359
1360 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1361 {
1362         struct fib6_info *f6i = res->f6i;
1363         unsigned short flags = fib6_info_dst_flags(f6i);
1364         struct net_device *dev;
1365         struct rt6_info *pcpu_rt;
1366
1367         if (!fib6_info_hold_safe(f6i))
1368                 return NULL;
1369
1370         rcu_read_lock();
1371         dev = ip6_rt_get_dev_rcu(res);
1372         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1373         rcu_read_unlock();
1374         if (!pcpu_rt) {
1375                 fib6_info_release(f6i);
1376                 return NULL;
1377         }
1378         ip6_rt_copy_init(pcpu_rt, res);
1379         pcpu_rt->rt6i_flags |= RTF_PCPU;
1380
1381         if (f6i->nh)
1382                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1383
1384         return pcpu_rt;
1385 }
1386
1387 static bool rt6_is_valid(const struct rt6_info *rt6)
1388 {
1389         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1390 }
1391
1392 /* It should be called with rcu_read_lock() acquired */
1393 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1394 {
1395         struct rt6_info *pcpu_rt;
1396
1397         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1398
1399         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1400                 struct rt6_info *prev, **p;
1401
1402                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1403                 prev = xchg(p, NULL);
1404                 if (prev) {
1405                         dst_dev_put(&prev->dst);
1406                         dst_release(&prev->dst);
1407                 }
1408
1409                 pcpu_rt = NULL;
1410         }
1411
1412         return pcpu_rt;
1413 }
1414
1415 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1416                                             const struct fib6_result *res)
1417 {
1418         struct rt6_info *pcpu_rt, *prev, **p;
1419
1420         pcpu_rt = ip6_rt_pcpu_alloc(res);
1421         if (!pcpu_rt)
1422                 return NULL;
1423
1424         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1425         prev = cmpxchg(p, NULL, pcpu_rt);
1426         BUG_ON(prev);
1427
1428         if (res->f6i->fib6_destroying) {
1429                 struct fib6_info *from;
1430
1431                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1432                 fib6_info_release(from);
1433         }
1434
1435         return pcpu_rt;
1436 }
1437
1438 /* exception hash table implementation
1439  */
1440 static DEFINE_SPINLOCK(rt6_exception_lock);
1441
1442 /* Remove rt6_ex from hash table and free the memory
1443  * Caller must hold rt6_exception_lock
1444  */
1445 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1446                                  struct rt6_exception *rt6_ex)
1447 {
1448         struct fib6_info *from;
1449         struct net *net;
1450
1451         if (!bucket || !rt6_ex)
1452                 return;
1453
1454         net = dev_net(rt6_ex->rt6i->dst.dev);
1455         net->ipv6.rt6_stats->fib_rt_cache--;
1456
1457         /* purge completely the exception to allow releasing the held resources:
1458          * some [sk] cache may keep the dst around for unlimited time
1459          */
1460         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1461         fib6_info_release(from);
1462         dst_dev_put(&rt6_ex->rt6i->dst);
1463
1464         hlist_del_rcu(&rt6_ex->hlist);
1465         dst_release(&rt6_ex->rt6i->dst);
1466         kfree_rcu(rt6_ex, rcu);
1467         WARN_ON_ONCE(!bucket->depth);
1468         bucket->depth--;
1469 }
1470
1471 /* Remove oldest rt6_ex in bucket and free the memory
1472  * Caller must hold rt6_exception_lock
1473  */
1474 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1475 {
1476         struct rt6_exception *rt6_ex, *oldest = NULL;
1477
1478         if (!bucket)
1479                 return;
1480
1481         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1482                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1483                         oldest = rt6_ex;
1484         }
1485         rt6_remove_exception(bucket, oldest);
1486 }
1487
1488 static u32 rt6_exception_hash(const struct in6_addr *dst,
1489                               const struct in6_addr *src)
1490 {
1491         static siphash_aligned_key_t rt6_exception_key;
1492         struct {
1493                 struct in6_addr dst;
1494                 struct in6_addr src;
1495         } __aligned(SIPHASH_ALIGNMENT) combined = {
1496                 .dst = *dst,
1497         };
1498         u64 val;
1499
1500         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1501
1502 #ifdef CONFIG_IPV6_SUBTREES
1503         if (src)
1504                 combined.src = *src;
1505 #endif
1506         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1507
1508         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1509 }
1510
1511 /* Helper function to find the cached rt in the hash table
1512  * and update bucket pointer to point to the bucket for this
1513  * (daddr, saddr) pair
1514  * Caller must hold rt6_exception_lock
1515  */
1516 static struct rt6_exception *
1517 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1518                               const struct in6_addr *daddr,
1519                               const struct in6_addr *saddr)
1520 {
1521         struct rt6_exception *rt6_ex;
1522         u32 hval;
1523
1524         if (!(*bucket) || !daddr)
1525                 return NULL;
1526
1527         hval = rt6_exception_hash(daddr, saddr);
1528         *bucket += hval;
1529
1530         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1531                 struct rt6_info *rt6 = rt6_ex->rt6i;
1532                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1533
1534 #ifdef CONFIG_IPV6_SUBTREES
1535                 if (matched && saddr)
1536                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1537 #endif
1538                 if (matched)
1539                         return rt6_ex;
1540         }
1541         return NULL;
1542 }
1543
1544 /* Helper function to find the cached rt in the hash table
1545  * and update bucket pointer to point to the bucket for this
1546  * (daddr, saddr) pair
1547  * Caller must hold rcu_read_lock()
1548  */
1549 static struct rt6_exception *
1550 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1551                          const struct in6_addr *daddr,
1552                          const struct in6_addr *saddr)
1553 {
1554         struct rt6_exception *rt6_ex;
1555         u32 hval;
1556
1557         WARN_ON_ONCE(!rcu_read_lock_held());
1558
1559         if (!(*bucket) || !daddr)
1560                 return NULL;
1561
1562         hval = rt6_exception_hash(daddr, saddr);
1563         *bucket += hval;
1564
1565         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1566                 struct rt6_info *rt6 = rt6_ex->rt6i;
1567                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1568
1569 #ifdef CONFIG_IPV6_SUBTREES
1570                 if (matched && saddr)
1571                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1572 #endif
1573                 if (matched)
1574                         return rt6_ex;
1575         }
1576         return NULL;
1577 }
1578
1579 static unsigned int fib6_mtu(const struct fib6_result *res)
1580 {
1581         const struct fib6_nh *nh = res->nh;
1582         unsigned int mtu;
1583
1584         if (res->f6i->fib6_pmtu) {
1585                 mtu = res->f6i->fib6_pmtu;
1586         } else {
1587                 struct net_device *dev = nh->fib_nh_dev;
1588                 struct inet6_dev *idev;
1589
1590                 rcu_read_lock();
1591                 idev = __in6_dev_get(dev);
1592                 mtu = idev->cnf.mtu6;
1593                 rcu_read_unlock();
1594         }
1595
1596         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1597
1598         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1599 }
1600
1601 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1602
1603 /* used when the flushed bit is not relevant, only access to the bucket
1604  * (ie., all bucket users except rt6_insert_exception);
1605  *
1606  * called under rcu lock; sometimes called with rt6_exception_lock held
1607  */
1608 static
1609 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1610                                                        spinlock_t *lock)
1611 {
1612         struct rt6_exception_bucket *bucket;
1613
1614         if (lock)
1615                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1616                                                    lockdep_is_held(lock));
1617         else
1618                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1619
1620         /* remove bucket flushed bit if set */
1621         if (bucket) {
1622                 unsigned long p = (unsigned long)bucket;
1623
1624                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1625                 bucket = (struct rt6_exception_bucket *)p;
1626         }
1627
1628         return bucket;
1629 }
1630
1631 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1632 {
1633         unsigned long p = (unsigned long)bucket;
1634
1635         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1636 }
1637
1638 /* called with rt6_exception_lock held */
1639 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1640                                               spinlock_t *lock)
1641 {
1642         struct rt6_exception_bucket *bucket;
1643         unsigned long p;
1644
1645         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1646                                            lockdep_is_held(lock));
1647
1648         p = (unsigned long)bucket;
1649         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1650         bucket = (struct rt6_exception_bucket *)p;
1651         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1652 }
1653
1654 static int rt6_insert_exception(struct rt6_info *nrt,
1655                                 const struct fib6_result *res)
1656 {
1657         struct net *net = dev_net(nrt->dst.dev);
1658         struct rt6_exception_bucket *bucket;
1659         struct fib6_info *f6i = res->f6i;
1660         struct in6_addr *src_key = NULL;
1661         struct rt6_exception *rt6_ex;
1662         struct fib6_nh *nh = res->nh;
1663         int max_depth;
1664         int err = 0;
1665
1666         spin_lock_bh(&rt6_exception_lock);
1667
1668         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1669                                           lockdep_is_held(&rt6_exception_lock));
1670         if (!bucket) {
1671                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1672                                  GFP_ATOMIC);
1673                 if (!bucket) {
1674                         err = -ENOMEM;
1675                         goto out;
1676                 }
1677                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1678         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1679                 err = -EINVAL;
1680                 goto out;
1681         }
1682
1683 #ifdef CONFIG_IPV6_SUBTREES
1684         /* fib6_src.plen != 0 indicates f6i is in subtree
1685          * and exception table is indexed by a hash of
1686          * both fib6_dst and fib6_src.
1687          * Otherwise, the exception table is indexed by
1688          * a hash of only fib6_dst.
1689          */
1690         if (f6i->fib6_src.plen)
1691                 src_key = &nrt->rt6i_src.addr;
1692 #endif
1693         /* rt6_mtu_change() might lower mtu on f6i.
1694          * Only insert this exception route if its mtu
1695          * is less than f6i's mtu value.
1696          */
1697         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1698                 err = -EINVAL;
1699                 goto out;
1700         }
1701
1702         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1703                                                src_key);
1704         if (rt6_ex)
1705                 rt6_remove_exception(bucket, rt6_ex);
1706
1707         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1708         if (!rt6_ex) {
1709                 err = -ENOMEM;
1710                 goto out;
1711         }
1712         rt6_ex->rt6i = nrt;
1713         rt6_ex->stamp = jiffies;
1714         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1715         bucket->depth++;
1716         net->ipv6.rt6_stats->fib_rt_cache++;
1717
1718         /* Randomize max depth to avoid some side channels attacks. */
1719         max_depth = FIB6_MAX_DEPTH + prandom_u32_max(FIB6_MAX_DEPTH);
1720         while (bucket->depth > max_depth)
1721                 rt6_exception_remove_oldest(bucket);
1722
1723 out:
1724         spin_unlock_bh(&rt6_exception_lock);
1725
1726         /* Update fn->fn_sernum to invalidate all cached dst */
1727         if (!err) {
1728                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1729                 fib6_update_sernum(net, f6i);
1730                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1731                 fib6_force_start_gc(net);
1732         }
1733
1734         return err;
1735 }
1736
1737 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1738 {
1739         struct rt6_exception_bucket *bucket;
1740         struct rt6_exception *rt6_ex;
1741         struct hlist_node *tmp;
1742         int i;
1743
1744         spin_lock_bh(&rt6_exception_lock);
1745
1746         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1747         if (!bucket)
1748                 goto out;
1749
1750         /* Prevent rt6_insert_exception() to recreate the bucket list */
1751         if (!from)
1752                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1753
1754         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1755                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1756                         if (!from ||
1757                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1758                                 rt6_remove_exception(bucket, rt6_ex);
1759                 }
1760                 WARN_ON_ONCE(!from && bucket->depth);
1761                 bucket++;
1762         }
1763 out:
1764         spin_unlock_bh(&rt6_exception_lock);
1765 }
1766
1767 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1768 {
1769         struct fib6_info *f6i = arg;
1770
1771         fib6_nh_flush_exceptions(nh, f6i);
1772
1773         return 0;
1774 }
1775
1776 void rt6_flush_exceptions(struct fib6_info *f6i)
1777 {
1778         if (f6i->nh)
1779                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1780                                          f6i);
1781         else
1782                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1783 }
1784
1785 /* Find cached rt in the hash table inside passed in rt
1786  * Caller has to hold rcu_read_lock()
1787  */
1788 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1789                                            const struct in6_addr *daddr,
1790                                            const struct in6_addr *saddr)
1791 {
1792         const struct in6_addr *src_key = NULL;
1793         struct rt6_exception_bucket *bucket;
1794         struct rt6_exception *rt6_ex;
1795         struct rt6_info *ret = NULL;
1796
1797 #ifdef CONFIG_IPV6_SUBTREES
1798         /* fib6i_src.plen != 0 indicates f6i is in subtree
1799          * and exception table is indexed by a hash of
1800          * both fib6_dst and fib6_src.
1801          * However, the src addr used to create the hash
1802          * might not be exactly the passed in saddr which
1803          * is a /128 addr from the flow.
1804          * So we need to use f6i->fib6_src to redo lookup
1805          * if the passed in saddr does not find anything.
1806          * (See the logic in ip6_rt_cache_alloc() on how
1807          * rt->rt6i_src is updated.)
1808          */
1809         if (res->f6i->fib6_src.plen)
1810                 src_key = saddr;
1811 find_ex:
1812 #endif
1813         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1814         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1815
1816         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1817                 ret = rt6_ex->rt6i;
1818
1819 #ifdef CONFIG_IPV6_SUBTREES
1820         /* Use fib6_src as src_key and redo lookup */
1821         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1822                 src_key = &res->f6i->fib6_src.addr;
1823                 goto find_ex;
1824         }
1825 #endif
1826
1827         return ret;
1828 }
1829
1830 /* Remove the passed in cached rt from the hash table that contains it */
1831 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1832                                     const struct rt6_info *rt)
1833 {
1834         const struct in6_addr *src_key = NULL;
1835         struct rt6_exception_bucket *bucket;
1836         struct rt6_exception *rt6_ex;
1837         int err;
1838
1839         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1840                 return -ENOENT;
1841
1842         spin_lock_bh(&rt6_exception_lock);
1843         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1844
1845 #ifdef CONFIG_IPV6_SUBTREES
1846         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1847          * and exception table is indexed by a hash of
1848          * both rt6i_dst and rt6i_src.
1849          * Otherwise, the exception table is indexed by
1850          * a hash of only rt6i_dst.
1851          */
1852         if (plen)
1853                 src_key = &rt->rt6i_src.addr;
1854 #endif
1855         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1856                                                &rt->rt6i_dst.addr,
1857                                                src_key);
1858         if (rt6_ex) {
1859                 rt6_remove_exception(bucket, rt6_ex);
1860                 err = 0;
1861         } else {
1862                 err = -ENOENT;
1863         }
1864
1865         spin_unlock_bh(&rt6_exception_lock);
1866         return err;
1867 }
1868
1869 struct fib6_nh_excptn_arg {
1870         struct rt6_info *rt;
1871         int             plen;
1872 };
1873
1874 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1875 {
1876         struct fib6_nh_excptn_arg *arg = _arg;
1877         int err;
1878
1879         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1880         if (err == 0)
1881                 return 1;
1882
1883         return 0;
1884 }
1885
1886 static int rt6_remove_exception_rt(struct rt6_info *rt)
1887 {
1888         struct fib6_info *from;
1889
1890         from = rcu_dereference(rt->from);
1891         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1892                 return -EINVAL;
1893
1894         if (from->nh) {
1895                 struct fib6_nh_excptn_arg arg = {
1896                         .rt = rt,
1897                         .plen = from->fib6_src.plen
1898                 };
1899                 int rc;
1900
1901                 /* rc = 1 means an entry was found */
1902                 rc = nexthop_for_each_fib6_nh(from->nh,
1903                                               rt6_nh_remove_exception_rt,
1904                                               &arg);
1905                 return rc ? 0 : -ENOENT;
1906         }
1907
1908         return fib6_nh_remove_exception(from->fib6_nh,
1909                                         from->fib6_src.plen, rt);
1910 }
1911
1912 /* Find rt6_ex which contains the passed in rt cache and
1913  * refresh its stamp
1914  */
1915 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1916                                      const struct rt6_info *rt)
1917 {
1918         const struct in6_addr *src_key = NULL;
1919         struct rt6_exception_bucket *bucket;
1920         struct rt6_exception *rt6_ex;
1921
1922         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1923 #ifdef CONFIG_IPV6_SUBTREES
1924         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1925          * and exception table is indexed by a hash of
1926          * both rt6i_dst and rt6i_src.
1927          * Otherwise, the exception table is indexed by
1928          * a hash of only rt6i_dst.
1929          */
1930         if (plen)
1931                 src_key = &rt->rt6i_src.addr;
1932 #endif
1933         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1934         if (rt6_ex)
1935                 rt6_ex->stamp = jiffies;
1936 }
1937
1938 struct fib6_nh_match_arg {
1939         const struct net_device *dev;
1940         const struct in6_addr   *gw;
1941         struct fib6_nh          *match;
1942 };
1943
1944 /* determine if fib6_nh has given device and gateway */
1945 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1946 {
1947         struct fib6_nh_match_arg *arg = _arg;
1948
1949         if (arg->dev != nh->fib_nh_dev ||
1950             (arg->gw && !nh->fib_nh_gw_family) ||
1951             (!arg->gw && nh->fib_nh_gw_family) ||
1952             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1953                 return 0;
1954
1955         arg->match = nh;
1956
1957         /* found a match, break the loop */
1958         return 1;
1959 }
1960
1961 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1962 {
1963         struct fib6_info *from;
1964         struct fib6_nh *fib6_nh;
1965
1966         rcu_read_lock();
1967
1968         from = rcu_dereference(rt->from);
1969         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1970                 goto unlock;
1971
1972         if (from->nh) {
1973                 struct fib6_nh_match_arg arg = {
1974                         .dev = rt->dst.dev,
1975                         .gw = &rt->rt6i_gateway,
1976                 };
1977
1978                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1979
1980                 if (!arg.match)
1981                         goto unlock;
1982                 fib6_nh = arg.match;
1983         } else {
1984                 fib6_nh = from->fib6_nh;
1985         }
1986         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1987 unlock:
1988         rcu_read_unlock();
1989 }
1990
1991 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1992                                          struct rt6_info *rt, int mtu)
1993 {
1994         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1995          * lowest MTU in the path: always allow updating the route PMTU to
1996          * reflect PMTU decreases.
1997          *
1998          * If the new MTU is higher, and the route PMTU is equal to the local
1999          * MTU, this means the old MTU is the lowest in the path, so allow
2000          * updating it: if other nodes now have lower MTUs, PMTU discovery will
2001          * handle this.
2002          */
2003
2004         if (dst_mtu(&rt->dst) >= mtu)
2005                 return true;
2006
2007         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2008                 return true;
2009
2010         return false;
2011 }
2012
2013 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2014                                        const struct fib6_nh *nh, int mtu)
2015 {
2016         struct rt6_exception_bucket *bucket;
2017         struct rt6_exception *rt6_ex;
2018         int i;
2019
2020         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2021         if (!bucket)
2022                 return;
2023
2024         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2025                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2026                         struct rt6_info *entry = rt6_ex->rt6i;
2027
2028                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2029                          * route), the metrics of its rt->from have already
2030                          * been updated.
2031                          */
2032                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2033                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2034                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2035                 }
2036                 bucket++;
2037         }
2038 }
2039
2040 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2041
2042 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2043                                             const struct in6_addr *gateway)
2044 {
2045         struct rt6_exception_bucket *bucket;
2046         struct rt6_exception *rt6_ex;
2047         struct hlist_node *tmp;
2048         int i;
2049
2050         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2051                 return;
2052
2053         spin_lock_bh(&rt6_exception_lock);
2054         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2055         if (bucket) {
2056                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2057                         hlist_for_each_entry_safe(rt6_ex, tmp,
2058                                                   &bucket->chain, hlist) {
2059                                 struct rt6_info *entry = rt6_ex->rt6i;
2060
2061                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2062                                     RTF_CACHE_GATEWAY &&
2063                                     ipv6_addr_equal(gateway,
2064                                                     &entry->rt6i_gateway)) {
2065                                         rt6_remove_exception(bucket, rt6_ex);
2066                                 }
2067                         }
2068                         bucket++;
2069                 }
2070         }
2071
2072         spin_unlock_bh(&rt6_exception_lock);
2073 }
2074
2075 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2076                                       struct rt6_exception *rt6_ex,
2077                                       struct fib6_gc_args *gc_args,
2078                                       unsigned long now)
2079 {
2080         struct rt6_info *rt = rt6_ex->rt6i;
2081
2082         /* we are pruning and obsoleting aged-out and non gateway exceptions
2083          * even if others have still references to them, so that on next
2084          * dst_check() such references can be dropped.
2085          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2086          * expired, independently from their aging, as per RFC 8201 section 4
2087          */
2088         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2089                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2090                         RT6_TRACE("aging clone %p\n", rt);
2091                         rt6_remove_exception(bucket, rt6_ex);
2092                         return;
2093                 }
2094         } else if (time_after(jiffies, rt->dst.expires)) {
2095                 RT6_TRACE("purging expired route %p\n", rt);
2096                 rt6_remove_exception(bucket, rt6_ex);
2097                 return;
2098         }
2099
2100         if (rt->rt6i_flags & RTF_GATEWAY) {
2101                 struct neighbour *neigh;
2102
2103                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2104
2105                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2106                         RT6_TRACE("purging route %p via non-router but gateway\n",
2107                                   rt);
2108                         rt6_remove_exception(bucket, rt6_ex);
2109                         return;
2110                 }
2111         }
2112
2113         gc_args->more++;
2114 }
2115
2116 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2117                                    struct fib6_gc_args *gc_args,
2118                                    unsigned long now)
2119 {
2120         struct rt6_exception_bucket *bucket;
2121         struct rt6_exception *rt6_ex;
2122         struct hlist_node *tmp;
2123         int i;
2124
2125         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2126                 return;
2127
2128         rcu_read_lock_bh();
2129         spin_lock(&rt6_exception_lock);
2130         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2131         if (bucket) {
2132                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2133                         hlist_for_each_entry_safe(rt6_ex, tmp,
2134                                                   &bucket->chain, hlist) {
2135                                 rt6_age_examine_exception(bucket, rt6_ex,
2136                                                           gc_args, now);
2137                         }
2138                         bucket++;
2139                 }
2140         }
2141         spin_unlock(&rt6_exception_lock);
2142         rcu_read_unlock_bh();
2143 }
2144
2145 struct fib6_nh_age_excptn_arg {
2146         struct fib6_gc_args     *gc_args;
2147         unsigned long           now;
2148 };
2149
2150 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2151 {
2152         struct fib6_nh_age_excptn_arg *arg = _arg;
2153
2154         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2155         return 0;
2156 }
2157
2158 void rt6_age_exceptions(struct fib6_info *f6i,
2159                         struct fib6_gc_args *gc_args,
2160                         unsigned long now)
2161 {
2162         if (f6i->nh) {
2163                 struct fib6_nh_age_excptn_arg arg = {
2164                         .gc_args = gc_args,
2165                         .now = now
2166                 };
2167
2168                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2169                                          &arg);
2170         } else {
2171                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2172         }
2173 }
2174
2175 /* must be called with rcu lock held */
2176 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2177                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2178 {
2179         struct fib6_node *fn, *saved_fn;
2180
2181         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2182         saved_fn = fn;
2183
2184 redo_rt6_select:
2185         rt6_select(net, fn, oif, res, strict);
2186         if (res->f6i == net->ipv6.fib6_null_entry) {
2187                 fn = fib6_backtrack(fn, &fl6->saddr);
2188                 if (fn)
2189                         goto redo_rt6_select;
2190                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2191                         /* also consider unreachable route */
2192                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2193                         fn = saved_fn;
2194                         goto redo_rt6_select;
2195                 }
2196         }
2197
2198         trace_fib6_table_lookup(net, res, table, fl6);
2199
2200         return 0;
2201 }
2202
2203 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2204                                int oif, struct flowi6 *fl6,
2205                                const struct sk_buff *skb, int flags)
2206 {
2207         struct fib6_result res = {};
2208         struct rt6_info *rt = NULL;
2209         int strict = 0;
2210
2211         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2212                      !rcu_read_lock_held());
2213
2214         strict |= flags & RT6_LOOKUP_F_IFACE;
2215         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2216         if (net->ipv6.devconf_all->forwarding == 0)
2217                 strict |= RT6_LOOKUP_F_REACHABLE;
2218
2219         rcu_read_lock();
2220
2221         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2222         if (res.f6i == net->ipv6.fib6_null_entry)
2223                 goto out;
2224
2225         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2226
2227         /*Search through exception table */
2228         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2229         if (rt) {
2230                 goto out;
2231         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2232                             !res.nh->fib_nh_gw_family)) {
2233                 /* Create a RTF_CACHE clone which will not be
2234                  * owned by the fib6 tree.  It is for the special case where
2235                  * the daddr in the skb during the neighbor look-up is different
2236                  * from the fl6->daddr used to look-up route here.
2237                  */
2238                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2239
2240                 if (rt) {
2241                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2242                          * As rt6_uncached_list_add() does not consume refcnt,
2243                          * this refcnt is always returned to the caller even
2244                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2245                          */
2246                         rt6_uncached_list_add(rt);
2247                         rcu_read_unlock();
2248
2249                         return rt;
2250                 }
2251         } else {
2252                 /* Get a percpu copy */
2253                 local_bh_disable();
2254                 rt = rt6_get_pcpu_route(&res);
2255
2256                 if (!rt)
2257                         rt = rt6_make_pcpu_route(net, &res);
2258
2259                 local_bh_enable();
2260         }
2261 out:
2262         if (!rt)
2263                 rt = net->ipv6.ip6_null_entry;
2264         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2265                 ip6_hold_safe(net, &rt);
2266         rcu_read_unlock();
2267
2268         return rt;
2269 }
2270 EXPORT_SYMBOL_GPL(ip6_pol_route);
2271
2272 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2273                                             struct fib6_table *table,
2274                                             struct flowi6 *fl6,
2275                                             const struct sk_buff *skb,
2276                                             int flags)
2277 {
2278         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2279 }
2280
2281 struct dst_entry *ip6_route_input_lookup(struct net *net,
2282                                          struct net_device *dev,
2283                                          struct flowi6 *fl6,
2284                                          const struct sk_buff *skb,
2285                                          int flags)
2286 {
2287         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2288                 flags |= RT6_LOOKUP_F_IFACE;
2289
2290         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2291 }
2292 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2293
2294 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2295                                   struct flow_keys *keys,
2296                                   struct flow_keys *flkeys)
2297 {
2298         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2299         const struct ipv6hdr *key_iph = outer_iph;
2300         struct flow_keys *_flkeys = flkeys;
2301         const struct ipv6hdr *inner_iph;
2302         const struct icmp6hdr *icmph;
2303         struct ipv6hdr _inner_iph;
2304         struct icmp6hdr _icmph;
2305
2306         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2307                 goto out;
2308
2309         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2310                                    sizeof(_icmph), &_icmph);
2311         if (!icmph)
2312                 goto out;
2313
2314         if (!icmpv6_is_err(icmph->icmp6_type))
2315                 goto out;
2316
2317         inner_iph = skb_header_pointer(skb,
2318                                        skb_transport_offset(skb) + sizeof(*icmph),
2319                                        sizeof(_inner_iph), &_inner_iph);
2320         if (!inner_iph)
2321                 goto out;
2322
2323         key_iph = inner_iph;
2324         _flkeys = NULL;
2325 out:
2326         if (_flkeys) {
2327                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2328                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2329                 keys->tags.flow_label = _flkeys->tags.flow_label;
2330                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2331         } else {
2332                 keys->addrs.v6addrs.src = key_iph->saddr;
2333                 keys->addrs.v6addrs.dst = key_iph->daddr;
2334                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2335                 keys->basic.ip_proto = key_iph->nexthdr;
2336         }
2337 }
2338
2339 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2340                                            const struct sk_buff *skb,
2341                                            bool *p_has_inner)
2342 {
2343         u32 hash_fields = ip6_multipath_hash_fields(net);
2344         struct flow_keys keys, hash_keys;
2345
2346         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2347                 return 0;
2348
2349         memset(&hash_keys, 0, sizeof(hash_keys));
2350         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2351
2352         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2353         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2354                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2355         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2356                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2357         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2358                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2359         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2360                 hash_keys.tags.flow_label = keys.tags.flow_label;
2361         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2362                 hash_keys.ports.src = keys.ports.src;
2363         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2364                 hash_keys.ports.dst = keys.ports.dst;
2365
2366         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2367         return flow_hash_from_keys(&hash_keys);
2368 }
2369
2370 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2371                                            const struct sk_buff *skb,
2372                                            bool has_inner)
2373 {
2374         u32 hash_fields = ip6_multipath_hash_fields(net);
2375         struct flow_keys keys, hash_keys;
2376
2377         /* We assume the packet carries an encapsulation, but if none was
2378          * encountered during dissection of the outer flow, then there is no
2379          * point in calling the flow dissector again.
2380          */
2381         if (!has_inner)
2382                 return 0;
2383
2384         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2385                 return 0;
2386
2387         memset(&hash_keys, 0, sizeof(hash_keys));
2388         skb_flow_dissect_flow_keys(skb, &keys, 0);
2389
2390         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2391                 return 0;
2392
2393         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2394                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2395                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2396                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2397                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2398                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2399         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2400                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2401                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2402                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2403                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2404                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2405                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2406                         hash_keys.tags.flow_label = keys.tags.flow_label;
2407         }
2408
2409         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2410                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2411         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2412                 hash_keys.ports.src = keys.ports.src;
2413         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2414                 hash_keys.ports.dst = keys.ports.dst;
2415
2416         return flow_hash_from_keys(&hash_keys);
2417 }
2418
2419 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2420                                          const struct sk_buff *skb)
2421 {
2422         u32 mhash, mhash_inner;
2423         bool has_inner = true;
2424
2425         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2426         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2427
2428         return jhash_2words(mhash, mhash_inner, 0);
2429 }
2430
2431 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2432                                          const struct flowi6 *fl6)
2433 {
2434         u32 hash_fields = ip6_multipath_hash_fields(net);
2435         struct flow_keys hash_keys;
2436
2437         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2438                 return 0;
2439
2440         memset(&hash_keys, 0, sizeof(hash_keys));
2441         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2442         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2443                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2444         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2445                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2446         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2447                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2448         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2449                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2450         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2451                 hash_keys.ports.src = fl6->fl6_sport;
2452         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2453                 hash_keys.ports.dst = fl6->fl6_dport;
2454
2455         return flow_hash_from_keys(&hash_keys);
2456 }
2457
2458 /* if skb is set it will be used and fl6 can be NULL */
2459 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2460                        const struct sk_buff *skb, struct flow_keys *flkeys)
2461 {
2462         struct flow_keys hash_keys;
2463         u32 mhash = 0;
2464
2465         switch (ip6_multipath_hash_policy(net)) {
2466         case 0:
2467                 memset(&hash_keys, 0, sizeof(hash_keys));
2468                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2469                 if (skb) {
2470                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2471                 } else {
2472                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2473                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2474                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2475                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2476                 }
2477                 mhash = flow_hash_from_keys(&hash_keys);
2478                 break;
2479         case 1:
2480                 if (skb) {
2481                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2482                         struct flow_keys keys;
2483
2484                         /* short-circuit if we already have L4 hash present */
2485                         if (skb->l4_hash)
2486                                 return skb_get_hash_raw(skb) >> 1;
2487
2488                         memset(&hash_keys, 0, sizeof(hash_keys));
2489
2490                         if (!flkeys) {
2491                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2492                                 flkeys = &keys;
2493                         }
2494                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2495                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2496                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2497                         hash_keys.ports.src = flkeys->ports.src;
2498                         hash_keys.ports.dst = flkeys->ports.dst;
2499                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2500                 } else {
2501                         memset(&hash_keys, 0, sizeof(hash_keys));
2502                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2503                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2504                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2505                         hash_keys.ports.src = fl6->fl6_sport;
2506                         hash_keys.ports.dst = fl6->fl6_dport;
2507                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2508                 }
2509                 mhash = flow_hash_from_keys(&hash_keys);
2510                 break;
2511         case 2:
2512                 memset(&hash_keys, 0, sizeof(hash_keys));
2513                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2514                 if (skb) {
2515                         struct flow_keys keys;
2516
2517                         if (!flkeys) {
2518                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2519                                 flkeys = &keys;
2520                         }
2521
2522                         /* Inner can be v4 or v6 */
2523                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2524                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2525                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2526                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2527                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2528                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2529                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2530                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2531                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2532                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2533                         } else {
2534                                 /* Same as case 0 */
2535                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2536                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2537                         }
2538                 } else {
2539                         /* Same as case 0 */
2540                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2541                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2542                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2543                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2544                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2545                 }
2546                 mhash = flow_hash_from_keys(&hash_keys);
2547                 break;
2548         case 3:
2549                 if (skb)
2550                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2551                 else
2552                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2553                 break;
2554         }
2555
2556         return mhash >> 1;
2557 }
2558
2559 /* Called with rcu held */
2560 void ip6_route_input(struct sk_buff *skb)
2561 {
2562         const struct ipv6hdr *iph = ipv6_hdr(skb);
2563         struct net *net = dev_net(skb->dev);
2564         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2565         struct ip_tunnel_info *tun_info;
2566         struct flowi6 fl6 = {
2567                 .flowi6_iif = skb->dev->ifindex,
2568                 .daddr = iph->daddr,
2569                 .saddr = iph->saddr,
2570                 .flowlabel = ip6_flowinfo(iph),
2571                 .flowi6_mark = skb->mark,
2572                 .flowi6_proto = iph->nexthdr,
2573         };
2574         struct flow_keys *flkeys = NULL, _flkeys;
2575
2576         tun_info = skb_tunnel_info(skb);
2577         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2578                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2579
2580         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2581                 flkeys = &_flkeys;
2582
2583         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2584                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2585         skb_dst_drop(skb);
2586         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2587                                                       &fl6, skb, flags));
2588 }
2589
2590 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2591                                              struct fib6_table *table,
2592                                              struct flowi6 *fl6,
2593                                              const struct sk_buff *skb,
2594                                              int flags)
2595 {
2596         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2597 }
2598
2599 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2600                                                const struct sock *sk,
2601                                                struct flowi6 *fl6, int flags)
2602 {
2603         bool any_src;
2604
2605         if (ipv6_addr_type(&fl6->daddr) &
2606             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2607                 struct dst_entry *dst;
2608
2609                 /* This function does not take refcnt on the dst */
2610                 dst = l3mdev_link_scope_lookup(net, fl6);
2611                 if (dst)
2612                         return dst;
2613         }
2614
2615         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2616
2617         flags |= RT6_LOOKUP_F_DST_NOREF;
2618         any_src = ipv6_addr_any(&fl6->saddr);
2619         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2620             (fl6->flowi6_oif && any_src))
2621                 flags |= RT6_LOOKUP_F_IFACE;
2622
2623         if (!any_src)
2624                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2625         else if (sk)
2626                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2627
2628         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2629 }
2630 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2631
2632 struct dst_entry *ip6_route_output_flags(struct net *net,
2633                                          const struct sock *sk,
2634                                          struct flowi6 *fl6,
2635                                          int flags)
2636 {
2637         struct dst_entry *dst;
2638         struct rt6_info *rt6;
2639
2640         rcu_read_lock();
2641         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2642         rt6 = (struct rt6_info *)dst;
2643         /* For dst cached in uncached_list, refcnt is already taken. */
2644         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2645                 dst = &net->ipv6.ip6_null_entry->dst;
2646                 dst_hold(dst);
2647         }
2648         rcu_read_unlock();
2649
2650         return dst;
2651 }
2652 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2653
2654 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2655 {
2656         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2657         struct net_device *loopback_dev = net->loopback_dev;
2658         struct dst_entry *new = NULL;
2659
2660         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2661                        DST_OBSOLETE_DEAD, 0);
2662         if (rt) {
2663                 rt6_info_init(rt);
2664                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2665
2666                 new = &rt->dst;
2667                 new->__use = 1;
2668                 new->input = dst_discard;
2669                 new->output = dst_discard_out;
2670
2671                 dst_copy_metrics(new, &ort->dst);
2672
2673                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2674                 rt->rt6i_gateway = ort->rt6i_gateway;
2675                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2676
2677                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2678 #ifdef CONFIG_IPV6_SUBTREES
2679                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2680 #endif
2681         }
2682
2683         dst_release(dst_orig);
2684         return new ? new : ERR_PTR(-ENOMEM);
2685 }
2686
2687 /*
2688  *      Destination cache support functions
2689  */
2690
2691 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2692 {
2693         u32 rt_cookie = 0;
2694
2695         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2696                 return false;
2697
2698         if (fib6_check_expired(f6i))
2699                 return false;
2700
2701         return true;
2702 }
2703
2704 static struct dst_entry *rt6_check(struct rt6_info *rt,
2705                                    struct fib6_info *from,
2706                                    u32 cookie)
2707 {
2708         u32 rt_cookie = 0;
2709
2710         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2711             rt_cookie != cookie)
2712                 return NULL;
2713
2714         if (rt6_check_expired(rt))
2715                 return NULL;
2716
2717         return &rt->dst;
2718 }
2719
2720 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2721                                             struct fib6_info *from,
2722                                             u32 cookie)
2723 {
2724         if (!__rt6_check_expired(rt) &&
2725             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2726             fib6_check(from, cookie))
2727                 return &rt->dst;
2728         else
2729                 return NULL;
2730 }
2731
2732 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2733                                                         u32 cookie)
2734 {
2735         struct dst_entry *dst_ret;
2736         struct fib6_info *from;
2737         struct rt6_info *rt;
2738
2739         rt = container_of(dst, struct rt6_info, dst);
2740
2741         if (rt->sernum)
2742                 return rt6_is_valid(rt) ? dst : NULL;
2743
2744         rcu_read_lock();
2745
2746         /* All IPV6 dsts are created with ->obsolete set to the value
2747          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2748          * into this function always.
2749          */
2750
2751         from = rcu_dereference(rt->from);
2752
2753         if (from && (rt->rt6i_flags & RTF_PCPU ||
2754             unlikely(!list_empty(&rt->rt6i_uncached))))
2755                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2756         else
2757                 dst_ret = rt6_check(rt, from, cookie);
2758
2759         rcu_read_unlock();
2760
2761         return dst_ret;
2762 }
2763 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2764
2765 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2766 {
2767         struct rt6_info *rt = (struct rt6_info *) dst;
2768
2769         if (rt) {
2770                 if (rt->rt6i_flags & RTF_CACHE) {
2771                         rcu_read_lock();
2772                         if (rt6_check_expired(rt)) {
2773                                 rt6_remove_exception_rt(rt);
2774                                 dst = NULL;
2775                         }
2776                         rcu_read_unlock();
2777                 } else {
2778                         dst_release(dst);
2779                         dst = NULL;
2780                 }
2781         }
2782         return dst;
2783 }
2784
2785 static void ip6_link_failure(struct sk_buff *skb)
2786 {
2787         struct rt6_info *rt;
2788
2789         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2790
2791         rt = (struct rt6_info *) skb_dst(skb);
2792         if (rt) {
2793                 rcu_read_lock();
2794                 if (rt->rt6i_flags & RTF_CACHE) {
2795                         rt6_remove_exception_rt(rt);
2796                 } else {
2797                         struct fib6_info *from;
2798                         struct fib6_node *fn;
2799
2800                         from = rcu_dereference(rt->from);
2801                         if (from) {
2802                                 fn = rcu_dereference(from->fib6_node);
2803                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2804                                         WRITE_ONCE(fn->fn_sernum, -1);
2805                         }
2806                 }
2807                 rcu_read_unlock();
2808         }
2809 }
2810
2811 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2812 {
2813         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2814                 struct fib6_info *from;
2815
2816                 rcu_read_lock();
2817                 from = rcu_dereference(rt0->from);
2818                 if (from)
2819                         rt0->dst.expires = from->expires;
2820                 rcu_read_unlock();
2821         }
2822
2823         dst_set_expires(&rt0->dst, timeout);
2824         rt0->rt6i_flags |= RTF_EXPIRES;
2825 }
2826
2827 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2828 {
2829         struct net *net = dev_net(rt->dst.dev);
2830
2831         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2832         rt->rt6i_flags |= RTF_MODIFIED;
2833         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2834 }
2835
2836 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2837 {
2838         return !(rt->rt6i_flags & RTF_CACHE) &&
2839                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2840 }
2841
2842 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2843                                  const struct ipv6hdr *iph, u32 mtu,
2844                                  bool confirm_neigh)
2845 {
2846         const struct in6_addr *daddr, *saddr;
2847         struct rt6_info *rt6 = (struct rt6_info *)dst;
2848
2849         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2850          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2851          * [see also comment in rt6_mtu_change_route()]
2852          */
2853
2854         if (iph) {
2855                 daddr = &iph->daddr;
2856                 saddr = &iph->saddr;
2857         } else if (sk) {
2858                 daddr = &sk->sk_v6_daddr;
2859                 saddr = &inet6_sk(sk)->saddr;
2860         } else {
2861                 daddr = NULL;
2862                 saddr = NULL;
2863         }
2864
2865         if (confirm_neigh)
2866                 dst_confirm_neigh(dst, daddr);
2867
2868         if (mtu < IPV6_MIN_MTU)
2869                 return;
2870         if (mtu >= dst_mtu(dst))
2871                 return;
2872
2873         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2874                 rt6_do_update_pmtu(rt6, mtu);
2875                 /* update rt6_ex->stamp for cache */
2876                 if (rt6->rt6i_flags & RTF_CACHE)
2877                         rt6_update_exception_stamp_rt(rt6);
2878         } else if (daddr) {
2879                 struct fib6_result res = {};
2880                 struct rt6_info *nrt6;
2881
2882                 rcu_read_lock();
2883                 res.f6i = rcu_dereference(rt6->from);
2884                 if (!res.f6i)
2885                         goto out_unlock;
2886
2887                 res.fib6_flags = res.f6i->fib6_flags;
2888                 res.fib6_type = res.f6i->fib6_type;
2889
2890                 if (res.f6i->nh) {
2891                         struct fib6_nh_match_arg arg = {
2892                                 .dev = dst->dev,
2893                                 .gw = &rt6->rt6i_gateway,
2894                         };
2895
2896                         nexthop_for_each_fib6_nh(res.f6i->nh,
2897                                                  fib6_nh_find_match, &arg);
2898
2899                         /* fib6_info uses a nexthop that does not have fib6_nh
2900                          * using the dst->dev + gw. Should be impossible.
2901                          */
2902                         if (!arg.match)
2903                                 goto out_unlock;
2904
2905                         res.nh = arg.match;
2906                 } else {
2907                         res.nh = res.f6i->fib6_nh;
2908                 }
2909
2910                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2911                 if (nrt6) {
2912                         rt6_do_update_pmtu(nrt6, mtu);
2913                         if (rt6_insert_exception(nrt6, &res))
2914                                 dst_release_immediate(&nrt6->dst);
2915                 }
2916 out_unlock:
2917                 rcu_read_unlock();
2918         }
2919 }
2920
2921 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2922                                struct sk_buff *skb, u32 mtu,
2923                                bool confirm_neigh)
2924 {
2925         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2926                              confirm_neigh);
2927 }
2928
2929 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2930                      int oif, u32 mark, kuid_t uid)
2931 {
2932         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2933         struct dst_entry *dst;
2934         struct flowi6 fl6 = {
2935                 .flowi6_oif = oif,
2936                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2937                 .daddr = iph->daddr,
2938                 .saddr = iph->saddr,
2939                 .flowlabel = ip6_flowinfo(iph),
2940                 .flowi6_uid = uid,
2941         };
2942
2943         dst = ip6_route_output(net, NULL, &fl6);
2944         if (!dst->error)
2945                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2946         dst_release(dst);
2947 }
2948 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2949
2950 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2951 {
2952         int oif = sk->sk_bound_dev_if;
2953         struct dst_entry *dst;
2954
2955         if (!oif && skb->dev)
2956                 oif = l3mdev_master_ifindex(skb->dev);
2957
2958         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, READ_ONCE(sk->sk_mark),
2959                         sk->sk_uid);
2960
2961         dst = __sk_dst_get(sk);
2962         if (!dst || !dst->obsolete ||
2963             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2964                 return;
2965
2966         bh_lock_sock(sk);
2967         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2968                 ip6_datagram_dst_update(sk, false);
2969         bh_unlock_sock(sk);
2970 }
2971 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2972
2973 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2974                            const struct flowi6 *fl6)
2975 {
2976 #ifdef CONFIG_IPV6_SUBTREES
2977         struct ipv6_pinfo *np = inet6_sk(sk);
2978 #endif
2979
2980         ip6_dst_store(sk, dst,
2981                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2982                       &sk->sk_v6_daddr : NULL,
2983 #ifdef CONFIG_IPV6_SUBTREES
2984                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2985                       &np->saddr :
2986 #endif
2987                       NULL);
2988 }
2989
2990 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2991                                   struct flowi6 *fl6,
2992                                   const struct in6_addr *gw,
2993                                   struct rt6_info **ret)
2994 {
2995         const struct fib6_nh *nh = res->nh;
2996
2997         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2998             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2999                 return false;
3000
3001         /* rt_cache's gateway might be different from its 'parent'
3002          * in the case of an ip redirect.
3003          * So we keep searching in the exception table if the gateway
3004          * is different.
3005          */
3006         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3007                 struct rt6_info *rt_cache;
3008
3009                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3010                 if (rt_cache &&
3011                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3012                         *ret = rt_cache;
3013                         return true;
3014                 }
3015                 return false;
3016         }
3017         return true;
3018 }
3019
3020 struct fib6_nh_rd_arg {
3021         struct fib6_result      *res;
3022         struct flowi6           *fl6;
3023         const struct in6_addr   *gw;
3024         struct rt6_info         **ret;
3025 };
3026
3027 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3028 {
3029         struct fib6_nh_rd_arg *arg = _arg;
3030
3031         arg->res->nh = nh;
3032         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3033 }
3034
3035 /* Handle redirects */
3036 struct ip6rd_flowi {
3037         struct flowi6 fl6;
3038         struct in6_addr gateway;
3039 };
3040
3041 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3042                                              struct fib6_table *table,
3043                                              struct flowi6 *fl6,
3044                                              const struct sk_buff *skb,
3045                                              int flags)
3046 {
3047         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3048         struct rt6_info *ret = NULL;
3049         struct fib6_result res = {};
3050         struct fib6_nh_rd_arg arg = {
3051                 .res = &res,
3052                 .fl6 = fl6,
3053                 .gw  = &rdfl->gateway,
3054                 .ret = &ret
3055         };
3056         struct fib6_info *rt;
3057         struct fib6_node *fn;
3058
3059         /* Get the "current" route for this destination and
3060          * check if the redirect has come from appropriate router.
3061          *
3062          * RFC 4861 specifies that redirects should only be
3063          * accepted if they come from the nexthop to the target.
3064          * Due to the way the routes are chosen, this notion
3065          * is a bit fuzzy and one might need to check all possible
3066          * routes.
3067          */
3068
3069         rcu_read_lock();
3070         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3071 restart:
3072         for_each_fib6_node_rt_rcu(fn) {
3073                 res.f6i = rt;
3074                 if (fib6_check_expired(rt))
3075                         continue;
3076                 if (rt->fib6_flags & RTF_REJECT)
3077                         break;
3078                 if (unlikely(rt->nh)) {
3079                         if (nexthop_is_blackhole(rt->nh))
3080                                 continue;
3081                         /* on match, res->nh is filled in and potentially ret */
3082                         if (nexthop_for_each_fib6_nh(rt->nh,
3083                                                      fib6_nh_redirect_match,
3084                                                      &arg))
3085                                 goto out;
3086                 } else {
3087                         res.nh = rt->fib6_nh;
3088                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3089                                                   &ret))
3090                                 goto out;
3091                 }
3092         }
3093
3094         if (!rt)
3095                 rt = net->ipv6.fib6_null_entry;
3096         else if (rt->fib6_flags & RTF_REJECT) {
3097                 ret = net->ipv6.ip6_null_entry;
3098                 goto out;
3099         }
3100
3101         if (rt == net->ipv6.fib6_null_entry) {
3102                 fn = fib6_backtrack(fn, &fl6->saddr);
3103                 if (fn)
3104                         goto restart;
3105         }
3106
3107         res.f6i = rt;
3108         res.nh = rt->fib6_nh;
3109 out:
3110         if (ret) {
3111                 ip6_hold_safe(net, &ret);
3112         } else {
3113                 res.fib6_flags = res.f6i->fib6_flags;
3114                 res.fib6_type = res.f6i->fib6_type;
3115                 ret = ip6_create_rt_rcu(&res);
3116         }
3117
3118         rcu_read_unlock();
3119
3120         trace_fib6_table_lookup(net, &res, table, fl6);
3121         return ret;
3122 };
3123
3124 static struct dst_entry *ip6_route_redirect(struct net *net,
3125                                             const struct flowi6 *fl6,
3126                                             const struct sk_buff *skb,
3127                                             const struct in6_addr *gateway)
3128 {
3129         int flags = RT6_LOOKUP_F_HAS_SADDR;
3130         struct ip6rd_flowi rdfl;
3131
3132         rdfl.fl6 = *fl6;
3133         rdfl.gateway = *gateway;
3134
3135         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3136                                 flags, __ip6_route_redirect);
3137 }
3138
3139 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3140                   kuid_t uid)
3141 {
3142         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3143         struct dst_entry *dst;
3144         struct flowi6 fl6 = {
3145                 .flowi6_iif = LOOPBACK_IFINDEX,
3146                 .flowi6_oif = oif,
3147                 .flowi6_mark = mark,
3148                 .daddr = iph->daddr,
3149                 .saddr = iph->saddr,
3150                 .flowlabel = ip6_flowinfo(iph),
3151                 .flowi6_uid = uid,
3152         };
3153
3154         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3155         rt6_do_redirect(dst, NULL, skb);
3156         dst_release(dst);
3157 }
3158 EXPORT_SYMBOL_GPL(ip6_redirect);
3159
3160 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3161 {
3162         const struct ipv6hdr *iph = ipv6_hdr(skb);
3163         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3164         struct dst_entry *dst;
3165         struct flowi6 fl6 = {
3166                 .flowi6_iif = LOOPBACK_IFINDEX,
3167                 .flowi6_oif = oif,
3168                 .daddr = msg->dest,
3169                 .saddr = iph->daddr,
3170                 .flowi6_uid = sock_net_uid(net, NULL),
3171         };
3172
3173         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3174         rt6_do_redirect(dst, NULL, skb);
3175         dst_release(dst);
3176 }
3177
3178 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3179 {
3180         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if,
3181                      READ_ONCE(sk->sk_mark), sk->sk_uid);
3182 }
3183 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3184
3185 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3186 {
3187         struct net_device *dev = dst->dev;
3188         unsigned int mtu = dst_mtu(dst);
3189         struct net *net = dev_net(dev);
3190
3191         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3192
3193         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3194                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3195
3196         /*
3197          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3198          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3199          * IPV6_MAXPLEN is also valid and means: "any MSS,
3200          * rely only on pmtu discovery"
3201          */
3202         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3203                 mtu = IPV6_MAXPLEN;
3204         return mtu;
3205 }
3206
3207 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3208 {
3209         return ip6_dst_mtu_maybe_forward(dst, false);
3210 }
3211 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3212
3213 /* MTU selection:
3214  * 1. mtu on route is locked - use it
3215  * 2. mtu from nexthop exception
3216  * 3. mtu from egress device
3217  *
3218  * based on ip6_dst_mtu_forward and exception logic of
3219  * rt6_find_cached_rt; called with rcu_read_lock
3220  */
3221 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3222                       const struct in6_addr *daddr,
3223                       const struct in6_addr *saddr)
3224 {
3225         const struct fib6_nh *nh = res->nh;
3226         struct fib6_info *f6i = res->f6i;
3227         struct inet6_dev *idev;
3228         struct rt6_info *rt;
3229         u32 mtu = 0;
3230
3231         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3232                 mtu = f6i->fib6_pmtu;
3233                 if (mtu)
3234                         goto out;
3235         }
3236
3237         rt = rt6_find_cached_rt(res, daddr, saddr);
3238         if (unlikely(rt)) {
3239                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3240         } else {
3241                 struct net_device *dev = nh->fib_nh_dev;
3242
3243                 mtu = IPV6_MIN_MTU;
3244                 idev = __in6_dev_get(dev);
3245                 if (idev && idev->cnf.mtu6 > mtu)
3246                         mtu = idev->cnf.mtu6;
3247         }
3248
3249         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3250 out:
3251         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3252 }
3253
3254 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3255                                   struct flowi6 *fl6)
3256 {
3257         struct dst_entry *dst;
3258         struct rt6_info *rt;
3259         struct inet6_dev *idev = in6_dev_get(dev);
3260         struct net *net = dev_net(dev);
3261
3262         if (unlikely(!idev))
3263                 return ERR_PTR(-ENODEV);
3264
3265         rt = ip6_dst_alloc(net, dev, 0);
3266         if (unlikely(!rt)) {
3267                 in6_dev_put(idev);
3268                 dst = ERR_PTR(-ENOMEM);
3269                 goto out;
3270         }
3271
3272         rt->dst.input = ip6_input;
3273         rt->dst.output  = ip6_output;
3274         rt->rt6i_gateway  = fl6->daddr;
3275         rt->rt6i_dst.addr = fl6->daddr;
3276         rt->rt6i_dst.plen = 128;
3277         rt->rt6i_idev     = idev;
3278         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3279
3280         /* Add this dst into uncached_list so that rt6_disable_ip() can
3281          * do proper release of the net_device
3282          */
3283         rt6_uncached_list_add(rt);
3284
3285         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3286
3287 out:
3288         return dst;
3289 }
3290
3291 static void ip6_dst_gc(struct dst_ops *ops)
3292 {
3293         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3294         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3295         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3296         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3297         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3298         unsigned int val;
3299         int entries;
3300
3301         entries = dst_entries_get_fast(ops);
3302         if (entries > ops->gc_thresh)
3303                 entries = dst_entries_get_slow(ops);
3304
3305         if (time_after(rt_last_gc + rt_min_interval, jiffies))
3306                 goto out;
3307
3308         fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3309         entries = dst_entries_get_slow(ops);
3310         if (entries < ops->gc_thresh)
3311                 atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3312 out:
3313         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3314         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3315 }
3316
3317 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3318                                const struct in6_addr *gw_addr, u32 tbid,
3319                                int flags, struct fib6_result *res)
3320 {
3321         struct flowi6 fl6 = {
3322                 .flowi6_oif = cfg->fc_ifindex,
3323                 .daddr = *gw_addr,
3324                 .saddr = cfg->fc_prefsrc,
3325         };
3326         struct fib6_table *table;
3327         int err;
3328
3329         table = fib6_get_table(net, tbid);
3330         if (!table)
3331                 return -EINVAL;
3332
3333         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3334                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3335
3336         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3337
3338         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3339         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3340                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3341                                  cfg->fc_ifindex != 0, NULL, flags);
3342
3343         return err;
3344 }
3345
3346 static int ip6_route_check_nh_onlink(struct net *net,
3347                                      struct fib6_config *cfg,
3348                                      const struct net_device *dev,
3349                                      struct netlink_ext_ack *extack)
3350 {
3351         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3352         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3353         struct fib6_result res = {};
3354         int err;
3355
3356         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3357         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3358             /* ignore match if it is the default route */
3359             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3360             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3361                 NL_SET_ERR_MSG(extack,
3362                                "Nexthop has invalid gateway or device mismatch");
3363                 err = -EINVAL;
3364         }
3365
3366         return err;
3367 }
3368
3369 static int ip6_route_check_nh(struct net *net,
3370                               struct fib6_config *cfg,
3371                               struct net_device **_dev,
3372                               struct inet6_dev **idev)
3373 {
3374         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3375         struct net_device *dev = _dev ? *_dev : NULL;
3376         int flags = RT6_LOOKUP_F_IFACE;
3377         struct fib6_result res = {};
3378         int err = -EHOSTUNREACH;
3379
3380         if (cfg->fc_table) {
3381                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3382                                           cfg->fc_table, flags, &res);
3383                 /* gw_addr can not require a gateway or resolve to a reject
3384                  * route. If a device is given, it must match the result.
3385                  */
3386                 if (err || res.fib6_flags & RTF_REJECT ||
3387                     res.nh->fib_nh_gw_family ||
3388                     (dev && dev != res.nh->fib_nh_dev))
3389                         err = -EHOSTUNREACH;
3390         }
3391
3392         if (err < 0) {
3393                 struct flowi6 fl6 = {
3394                         .flowi6_oif = cfg->fc_ifindex,
3395                         .daddr = *gw_addr,
3396                 };
3397
3398                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3399                 if (err || res.fib6_flags & RTF_REJECT ||
3400                     res.nh->fib_nh_gw_family)
3401                         err = -EHOSTUNREACH;
3402
3403                 if (err)
3404                         return err;
3405
3406                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3407                                  cfg->fc_ifindex != 0, NULL, flags);
3408         }
3409
3410         err = 0;
3411         if (dev) {
3412                 if (dev != res.nh->fib_nh_dev)
3413                         err = -EHOSTUNREACH;
3414         } else {
3415                 *_dev = dev = res.nh->fib_nh_dev;
3416                 dev_hold(dev);
3417                 *idev = in6_dev_get(dev);
3418         }
3419
3420         return err;
3421 }
3422
3423 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3424                            struct net_device **_dev, struct inet6_dev **idev,
3425                            struct netlink_ext_ack *extack)
3426 {
3427         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3428         int gwa_type = ipv6_addr_type(gw_addr);
3429         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3430         const struct net_device *dev = *_dev;
3431         bool need_addr_check = !dev;
3432         int err = -EINVAL;
3433
3434         /* if gw_addr is local we will fail to detect this in case
3435          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3436          * will return already-added prefix route via interface that
3437          * prefix route was assigned to, which might be non-loopback.
3438          */
3439         if (dev &&
3440             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3441                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3442                 goto out;
3443         }
3444
3445         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3446                 /* IPv6 strictly inhibits using not link-local
3447                  * addresses as nexthop address.
3448                  * Otherwise, router will not able to send redirects.
3449                  * It is very good, but in some (rare!) circumstances
3450                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3451                  * some exceptions. --ANK
3452                  * We allow IPv4-mapped nexthops to support RFC4798-type
3453                  * addressing
3454                  */
3455                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3456                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3457                         goto out;
3458                 }
3459
3460                 rcu_read_lock();
3461
3462                 if (cfg->fc_flags & RTNH_F_ONLINK)
3463                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3464                 else
3465                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3466
3467                 rcu_read_unlock();
3468
3469                 if (err)
3470                         goto out;
3471         }
3472
3473         /* reload in case device was changed */
3474         dev = *_dev;
3475
3476         err = -EINVAL;
3477         if (!dev) {
3478                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3479                 goto out;
3480         } else if (dev->flags & IFF_LOOPBACK) {
3481                 NL_SET_ERR_MSG(extack,
3482                                "Egress device can not be loopback device for this route");
3483                 goto out;
3484         }
3485
3486         /* if we did not check gw_addr above, do so now that the
3487          * egress device has been resolved.
3488          */
3489         if (need_addr_check &&
3490             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3491                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3492                 goto out;
3493         }
3494
3495         err = 0;
3496 out:
3497         return err;
3498 }
3499
3500 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3501 {
3502         if ((flags & RTF_REJECT) ||
3503             (dev && (dev->flags & IFF_LOOPBACK) &&
3504              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3505              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3506                 return true;
3507
3508         return false;
3509 }
3510
3511 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3512                  struct fib6_config *cfg, gfp_t gfp_flags,
3513                  struct netlink_ext_ack *extack)
3514 {
3515         struct net_device *dev = NULL;
3516         struct inet6_dev *idev = NULL;
3517         int addr_type;
3518         int err;
3519
3520         fib6_nh->fib_nh_family = AF_INET6;
3521 #ifdef CONFIG_IPV6_ROUTER_PREF
3522         fib6_nh->last_probe = jiffies;
3523 #endif
3524         if (cfg->fc_is_fdb) {
3525                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3526                 fib6_nh->fib_nh_gw_family = AF_INET6;
3527                 return 0;
3528         }
3529
3530         err = -ENODEV;
3531         if (cfg->fc_ifindex) {
3532                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3533                 if (!dev)
3534                         goto out;
3535                 idev = in6_dev_get(dev);
3536                 if (!idev)
3537                         goto out;
3538         }
3539
3540         if (cfg->fc_flags & RTNH_F_ONLINK) {
3541                 if (!dev) {
3542                         NL_SET_ERR_MSG(extack,
3543                                        "Nexthop device required for onlink");
3544                         goto out;
3545                 }
3546
3547                 if (!(dev->flags & IFF_UP)) {
3548                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3549                         err = -ENETDOWN;
3550                         goto out;
3551                 }
3552
3553                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3554         }
3555
3556         fib6_nh->fib_nh_weight = 1;
3557
3558         /* We cannot add true routes via loopback here,
3559          * they would result in kernel looping; promote them to reject routes
3560          */
3561         addr_type = ipv6_addr_type(&cfg->fc_dst);
3562         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3563                 /* hold loopback dev/idev if we haven't done so. */
3564                 if (dev != net->loopback_dev) {
3565                         if (dev) {
3566                                 dev_put(dev);
3567                                 in6_dev_put(idev);
3568                         }
3569                         dev = net->loopback_dev;
3570                         dev_hold(dev);
3571                         idev = in6_dev_get(dev);
3572                         if (!idev) {
3573                                 err = -ENODEV;
3574                                 goto out;
3575                         }
3576                 }
3577                 goto pcpu_alloc;
3578         }
3579
3580         if (cfg->fc_flags & RTF_GATEWAY) {
3581                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3582                 if (err)
3583                         goto out;
3584
3585                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3586                 fib6_nh->fib_nh_gw_family = AF_INET6;
3587         }
3588
3589         err = -ENODEV;
3590         if (!dev)
3591                 goto out;
3592
3593         if (idev->cnf.disable_ipv6) {
3594                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3595                 err = -EACCES;
3596                 goto out;
3597         }
3598
3599         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3600                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3601                 err = -ENETDOWN;
3602                 goto out;
3603         }
3604
3605         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3606             !netif_carrier_ok(dev))
3607                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3608
3609         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3610                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3611         if (err)
3612                 goto out;
3613
3614 pcpu_alloc:
3615         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3616         if (!fib6_nh->rt6i_pcpu) {
3617                 err = -ENOMEM;
3618                 goto out;
3619         }
3620
3621         fib6_nh->fib_nh_dev = dev;
3622         netdev_tracker_alloc(dev, &fib6_nh->fib_nh_dev_tracker, gfp_flags);
3623
3624         fib6_nh->fib_nh_oif = dev->ifindex;
3625         err = 0;
3626 out:
3627         if (idev)
3628                 in6_dev_put(idev);
3629
3630         if (err) {
3631                 lwtstate_put(fib6_nh->fib_nh_lws);
3632                 fib6_nh->fib_nh_lws = NULL;
3633                 dev_put(dev);
3634         }
3635
3636         return err;
3637 }
3638
3639 void fib6_nh_release(struct fib6_nh *fib6_nh)
3640 {
3641         struct rt6_exception_bucket *bucket;
3642
3643         rcu_read_lock();
3644
3645         fib6_nh_flush_exceptions(fib6_nh, NULL);
3646         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3647         if (bucket) {
3648                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3649                 kfree(bucket);
3650         }
3651
3652         rcu_read_unlock();
3653
3654         fib6_nh_release_dsts(fib6_nh);
3655         free_percpu(fib6_nh->rt6i_pcpu);
3656
3657         fib_nh_common_release(&fib6_nh->nh_common);
3658 }
3659
3660 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3661 {
3662         int cpu;
3663
3664         if (!fib6_nh->rt6i_pcpu)
3665                 return;
3666
3667         for_each_possible_cpu(cpu) {
3668                 struct rt6_info *pcpu_rt, **ppcpu_rt;
3669
3670                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3671                 pcpu_rt = xchg(ppcpu_rt, NULL);
3672                 if (pcpu_rt) {
3673                         dst_dev_put(&pcpu_rt->dst);
3674                         dst_release(&pcpu_rt->dst);
3675                 }
3676         }
3677 }
3678
3679 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3680                                               gfp_t gfp_flags,
3681                                               struct netlink_ext_ack *extack)
3682 {
3683         struct net *net = cfg->fc_nlinfo.nl_net;
3684         struct fib6_info *rt = NULL;
3685         struct nexthop *nh = NULL;
3686         struct fib6_table *table;
3687         struct fib6_nh *fib6_nh;
3688         int err = -EINVAL;
3689         int addr_type;
3690
3691         /* RTF_PCPU is an internal flag; can not be set by userspace */
3692         if (cfg->fc_flags & RTF_PCPU) {
3693                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3694                 goto out;
3695         }
3696
3697         /* RTF_CACHE is an internal flag; can not be set by userspace */
3698         if (cfg->fc_flags & RTF_CACHE) {
3699                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3700                 goto out;
3701         }
3702
3703         if (cfg->fc_type > RTN_MAX) {
3704                 NL_SET_ERR_MSG(extack, "Invalid route type");
3705                 goto out;
3706         }
3707
3708         if (cfg->fc_dst_len > 128) {
3709                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3710                 goto out;
3711         }
3712         if (cfg->fc_src_len > 128) {
3713                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3714                 goto out;
3715         }
3716 #ifndef CONFIG_IPV6_SUBTREES
3717         if (cfg->fc_src_len) {
3718                 NL_SET_ERR_MSG(extack,
3719                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3720                 goto out;
3721         }
3722 #endif
3723         if (cfg->fc_nh_id) {
3724                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3725                 if (!nh) {
3726                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3727                         goto out;
3728                 }
3729                 err = fib6_check_nexthop(nh, cfg, extack);
3730                 if (err)
3731                         goto out;
3732         }
3733
3734         err = -ENOBUFS;
3735         if (cfg->fc_nlinfo.nlh &&
3736             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3737                 table = fib6_get_table(net, cfg->fc_table);
3738                 if (!table) {
3739                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3740                         table = fib6_new_table(net, cfg->fc_table);
3741                 }
3742         } else {
3743                 table = fib6_new_table(net, cfg->fc_table);
3744         }
3745
3746         if (!table)
3747                 goto out;
3748
3749         err = -ENOMEM;
3750         rt = fib6_info_alloc(gfp_flags, !nh);
3751         if (!rt)
3752                 goto out;
3753
3754         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3755                                                extack);
3756         if (IS_ERR(rt->fib6_metrics)) {
3757                 err = PTR_ERR(rt->fib6_metrics);
3758                 /* Do not leave garbage there. */
3759                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3760                 goto out_free;
3761         }
3762
3763         if (cfg->fc_flags & RTF_ADDRCONF)
3764                 rt->dst_nocount = true;
3765
3766         if (cfg->fc_flags & RTF_EXPIRES)
3767                 fib6_set_expires(rt, jiffies +
3768                                 clock_t_to_jiffies(cfg->fc_expires));
3769         else
3770                 fib6_clean_expires(rt);
3771
3772         if (cfg->fc_protocol == RTPROT_UNSPEC)
3773                 cfg->fc_protocol = RTPROT_BOOT;
3774         rt->fib6_protocol = cfg->fc_protocol;
3775
3776         rt->fib6_table = table;
3777         rt->fib6_metric = cfg->fc_metric;
3778         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3779         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3780
3781         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3782         rt->fib6_dst.plen = cfg->fc_dst_len;
3783
3784 #ifdef CONFIG_IPV6_SUBTREES
3785         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3786         rt->fib6_src.plen = cfg->fc_src_len;
3787 #endif
3788         if (nh) {
3789                 if (rt->fib6_src.plen) {
3790                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3791                         goto out_free;
3792                 }
3793                 if (!nexthop_get(nh)) {
3794                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3795                         goto out_free;
3796                 }
3797                 rt->nh = nh;
3798                 fib6_nh = nexthop_fib6_nh(rt->nh);
3799         } else {
3800                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3801                 if (err)
3802                         goto out;
3803
3804                 fib6_nh = rt->fib6_nh;
3805
3806                 /* We cannot add true routes via loopback here, they would
3807                  * result in kernel looping; promote them to reject routes
3808                  */
3809                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3810                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3811                                    addr_type))
3812                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3813         }
3814
3815         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3816                 struct net_device *dev = fib6_nh->fib_nh_dev;
3817
3818                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3819                         NL_SET_ERR_MSG(extack, "Invalid source address");
3820                         err = -EINVAL;
3821                         goto out;
3822                 }
3823                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3824                 rt->fib6_prefsrc.plen = 128;
3825         } else
3826                 rt->fib6_prefsrc.plen = 0;
3827
3828         return rt;
3829 out:
3830         fib6_info_release(rt);
3831         return ERR_PTR(err);
3832 out_free:
3833         ip_fib_metrics_put(rt->fib6_metrics);
3834         kfree(rt);
3835         return ERR_PTR(err);
3836 }
3837
3838 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3839                   struct netlink_ext_ack *extack)
3840 {
3841         struct fib6_info *rt;
3842         int err;
3843
3844         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3845         if (IS_ERR(rt))
3846                 return PTR_ERR(rt);
3847
3848         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3849         fib6_info_release(rt);
3850
3851         return err;
3852 }
3853
3854 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3855 {
3856         struct net *net = info->nl_net;
3857         struct fib6_table *table;
3858         int err;
3859
3860         if (rt == net->ipv6.fib6_null_entry) {
3861                 err = -ENOENT;
3862                 goto out;
3863         }
3864
3865         table = rt->fib6_table;
3866         spin_lock_bh(&table->tb6_lock);
3867         err = fib6_del(rt, info);
3868         spin_unlock_bh(&table->tb6_lock);
3869
3870 out:
3871         fib6_info_release(rt);
3872         return err;
3873 }
3874
3875 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3876 {
3877         struct nl_info info = {
3878                 .nl_net = net,
3879                 .skip_notify = skip_notify
3880         };
3881
3882         return __ip6_del_rt(rt, &info);
3883 }
3884
3885 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3886 {
3887         struct nl_info *info = &cfg->fc_nlinfo;
3888         struct net *net = info->nl_net;
3889         struct sk_buff *skb = NULL;
3890         struct fib6_table *table;
3891         int err = -ENOENT;
3892
3893         if (rt == net->ipv6.fib6_null_entry)
3894                 goto out_put;
3895         table = rt->fib6_table;
3896         spin_lock_bh(&table->tb6_lock);
3897
3898         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3899                 struct fib6_info *sibling, *next_sibling;
3900                 struct fib6_node *fn;
3901
3902                 /* prefer to send a single notification with all hops */
3903                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3904                 if (skb) {
3905                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3906
3907                         if (rt6_fill_node(net, skb, rt, NULL,
3908                                           NULL, NULL, 0, RTM_DELROUTE,
3909                                           info->portid, seq, 0) < 0) {
3910                                 kfree_skb(skb);
3911                                 skb = NULL;
3912                         } else
3913                                 info->skip_notify = 1;
3914                 }
3915
3916                 /* 'rt' points to the first sibling route. If it is not the
3917                  * leaf, then we do not need to send a notification. Otherwise,
3918                  * we need to check if the last sibling has a next route or not
3919                  * and emit a replace or delete notification, respectively.
3920                  */
3921                 info->skip_notify_kernel = 1;
3922                 fn = rcu_dereference_protected(rt->fib6_node,
3923                                             lockdep_is_held(&table->tb6_lock));
3924                 if (rcu_access_pointer(fn->leaf) == rt) {
3925                         struct fib6_info *last_sibling, *replace_rt;
3926
3927                         last_sibling = list_last_entry(&rt->fib6_siblings,
3928                                                        struct fib6_info,
3929                                                        fib6_siblings);
3930                         replace_rt = rcu_dereference_protected(
3931                                             last_sibling->fib6_next,
3932                                             lockdep_is_held(&table->tb6_lock));
3933                         if (replace_rt)
3934                                 call_fib6_entry_notifiers_replace(net,
3935                                                                   replace_rt);
3936                         else
3937                                 call_fib6_multipath_entry_notifiers(net,
3938                                                        FIB_EVENT_ENTRY_DEL,
3939                                                        rt, rt->fib6_nsiblings,
3940                                                        NULL);
3941                 }
3942                 list_for_each_entry_safe(sibling, next_sibling,
3943                                          &rt->fib6_siblings,
3944                                          fib6_siblings) {
3945                         err = fib6_del(sibling, info);
3946                         if (err)
3947                                 goto out_unlock;
3948                 }
3949         }
3950
3951         err = fib6_del(rt, info);
3952 out_unlock:
3953         spin_unlock_bh(&table->tb6_lock);
3954 out_put:
3955         fib6_info_release(rt);
3956
3957         if (skb) {
3958                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3959                             info->nlh, gfp_any());
3960         }
3961         return err;
3962 }
3963
3964 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3965 {
3966         int rc = -ESRCH;
3967
3968         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3969                 goto out;
3970
3971         if (cfg->fc_flags & RTF_GATEWAY &&
3972             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3973                 goto out;
3974
3975         rc = rt6_remove_exception_rt(rt);
3976 out:
3977         return rc;
3978 }
3979
3980 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3981                              struct fib6_nh *nh)
3982 {
3983         struct fib6_result res = {
3984                 .f6i = rt,
3985                 .nh = nh,
3986         };
3987         struct rt6_info *rt_cache;
3988
3989         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3990         if (rt_cache)
3991                 return __ip6_del_cached_rt(rt_cache, cfg);
3992
3993         return 0;
3994 }
3995
3996 struct fib6_nh_del_cached_rt_arg {
3997         struct fib6_config *cfg;
3998         struct fib6_info *f6i;
3999 };
4000
4001 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4002 {
4003         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4004         int rc;
4005
4006         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4007         return rc != -ESRCH ? rc : 0;
4008 }
4009
4010 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4011 {
4012         struct fib6_nh_del_cached_rt_arg arg = {
4013                 .cfg = cfg,
4014                 .f6i = f6i
4015         };
4016
4017         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4018 }
4019
4020 static int ip6_route_del(struct fib6_config *cfg,
4021                          struct netlink_ext_ack *extack)
4022 {
4023         struct fib6_table *table;
4024         struct fib6_info *rt;
4025         struct fib6_node *fn;
4026         int err = -ESRCH;
4027
4028         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4029         if (!table) {
4030                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4031                 return err;
4032         }
4033
4034         rcu_read_lock();
4035
4036         fn = fib6_locate(&table->tb6_root,
4037                          &cfg->fc_dst, cfg->fc_dst_len,
4038                          &cfg->fc_src, cfg->fc_src_len,
4039                          !(cfg->fc_flags & RTF_CACHE));
4040
4041         if (fn) {
4042                 for_each_fib6_node_rt_rcu(fn) {
4043                         struct fib6_nh *nh;
4044
4045                         if (rt->nh && cfg->fc_nh_id &&
4046                             rt->nh->id != cfg->fc_nh_id)
4047                                 continue;
4048
4049                         if (cfg->fc_flags & RTF_CACHE) {
4050                                 int rc = 0;
4051
4052                                 if (rt->nh) {
4053                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4054                                 } else if (cfg->fc_nh_id) {
4055                                         continue;
4056                                 } else {
4057                                         nh = rt->fib6_nh;
4058                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4059                                 }
4060                                 if (rc != -ESRCH) {
4061                                         rcu_read_unlock();
4062                                         return rc;
4063                                 }
4064                                 continue;
4065                         }
4066
4067                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4068                                 continue;
4069                         if (cfg->fc_protocol &&
4070                             cfg->fc_protocol != rt->fib6_protocol)
4071                                 continue;
4072
4073                         if (rt->nh) {
4074                                 if (!fib6_info_hold_safe(rt))
4075                                         continue;
4076                                 rcu_read_unlock();
4077
4078                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4079                         }
4080                         if (cfg->fc_nh_id)
4081                                 continue;
4082
4083                         nh = rt->fib6_nh;
4084                         if (cfg->fc_ifindex &&
4085                             (!nh->fib_nh_dev ||
4086                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4087                                 continue;
4088                         if (cfg->fc_flags & RTF_GATEWAY &&
4089                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4090                                 continue;
4091                         if (!fib6_info_hold_safe(rt))
4092                                 continue;
4093                         rcu_read_unlock();
4094
4095                         /* if gateway was specified only delete the one hop */
4096                         if (cfg->fc_flags & RTF_GATEWAY)
4097                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4098
4099                         return __ip6_del_rt_siblings(rt, cfg);
4100                 }
4101         }
4102         rcu_read_unlock();
4103
4104         return err;
4105 }
4106
4107 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4108 {
4109         struct netevent_redirect netevent;
4110         struct rt6_info *rt, *nrt = NULL;
4111         struct fib6_result res = {};
4112         struct ndisc_options ndopts;
4113         struct inet6_dev *in6_dev;
4114         struct neighbour *neigh;
4115         struct rd_msg *msg;
4116         int optlen, on_link;
4117         u8 *lladdr;
4118
4119         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4120         optlen -= sizeof(*msg);
4121
4122         if (optlen < 0) {
4123                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4124                 return;
4125         }
4126
4127         msg = (struct rd_msg *)icmp6_hdr(skb);
4128
4129         if (ipv6_addr_is_multicast(&msg->dest)) {
4130                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4131                 return;
4132         }
4133
4134         on_link = 0;
4135         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4136                 on_link = 1;
4137         } else if (ipv6_addr_type(&msg->target) !=
4138                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4139                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4140                 return;
4141         }
4142
4143         in6_dev = __in6_dev_get(skb->dev);
4144         if (!in6_dev)
4145                 return;
4146         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4147                 return;
4148
4149         /* RFC2461 8.1:
4150          *      The IP source address of the Redirect MUST be the same as the current
4151          *      first-hop router for the specified ICMP Destination Address.
4152          */
4153
4154         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4155                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4156                 return;
4157         }
4158
4159         lladdr = NULL;
4160         if (ndopts.nd_opts_tgt_lladdr) {
4161                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4162                                              skb->dev);
4163                 if (!lladdr) {
4164                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4165                         return;
4166                 }
4167         }
4168
4169         rt = (struct rt6_info *) dst;
4170         if (rt->rt6i_flags & RTF_REJECT) {
4171                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4172                 return;
4173         }
4174
4175         /* Redirect received -> path was valid.
4176          * Look, redirects are sent only in response to data packets,
4177          * so that this nexthop apparently is reachable. --ANK
4178          */
4179         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4180
4181         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4182         if (!neigh)
4183                 return;
4184
4185         /*
4186          *      We have finally decided to accept it.
4187          */
4188
4189         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4190                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4191                      NEIGH_UPDATE_F_OVERRIDE|
4192                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4193                                      NEIGH_UPDATE_F_ISROUTER)),
4194                      NDISC_REDIRECT, &ndopts);
4195
4196         rcu_read_lock();
4197         res.f6i = rcu_dereference(rt->from);
4198         if (!res.f6i)
4199                 goto out;
4200
4201         if (res.f6i->nh) {
4202                 struct fib6_nh_match_arg arg = {
4203                         .dev = dst->dev,
4204                         .gw = &rt->rt6i_gateway,
4205                 };
4206
4207                 nexthop_for_each_fib6_nh(res.f6i->nh,
4208                                          fib6_nh_find_match, &arg);
4209
4210                 /* fib6_info uses a nexthop that does not have fib6_nh
4211                  * using the dst->dev. Should be impossible
4212                  */
4213                 if (!arg.match)
4214                         goto out;
4215                 res.nh = arg.match;
4216         } else {
4217                 res.nh = res.f6i->fib6_nh;
4218         }
4219
4220         res.fib6_flags = res.f6i->fib6_flags;
4221         res.fib6_type = res.f6i->fib6_type;
4222         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4223         if (!nrt)
4224                 goto out;
4225
4226         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4227         if (on_link)
4228                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4229
4230         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4231
4232         /* rt6_insert_exception() will take care of duplicated exceptions */
4233         if (rt6_insert_exception(nrt, &res)) {
4234                 dst_release_immediate(&nrt->dst);
4235                 goto out;
4236         }
4237
4238         netevent.old = &rt->dst;
4239         netevent.new = &nrt->dst;
4240         netevent.daddr = &msg->dest;
4241         netevent.neigh = neigh;
4242         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4243
4244 out:
4245         rcu_read_unlock();
4246         neigh_release(neigh);
4247 }
4248
4249 #ifdef CONFIG_IPV6_ROUTE_INFO
4250 static struct fib6_info *rt6_get_route_info(struct net *net,
4251                                            const struct in6_addr *prefix, int prefixlen,
4252                                            const struct in6_addr *gwaddr,
4253                                            struct net_device *dev)
4254 {
4255         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4256         int ifindex = dev->ifindex;
4257         struct fib6_node *fn;
4258         struct fib6_info *rt = NULL;
4259         struct fib6_table *table;
4260
4261         table = fib6_get_table(net, tb_id);
4262         if (!table)
4263                 return NULL;
4264
4265         rcu_read_lock();
4266         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4267         if (!fn)
4268                 goto out;
4269
4270         for_each_fib6_node_rt_rcu(fn) {
4271                 /* these routes do not use nexthops */
4272                 if (rt->nh)
4273                         continue;
4274                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4275                         continue;
4276                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4277                     !rt->fib6_nh->fib_nh_gw_family)
4278                         continue;
4279                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4280                         continue;
4281                 if (!fib6_info_hold_safe(rt))
4282                         continue;
4283                 break;
4284         }
4285 out:
4286         rcu_read_unlock();
4287         return rt;
4288 }
4289
4290 static struct fib6_info *rt6_add_route_info(struct net *net,
4291                                            const struct in6_addr *prefix, int prefixlen,
4292                                            const struct in6_addr *gwaddr,
4293                                            struct net_device *dev,
4294                                            unsigned int pref)
4295 {
4296         struct fib6_config cfg = {
4297                 .fc_metric      = IP6_RT_PRIO_USER,
4298                 .fc_ifindex     = dev->ifindex,
4299                 .fc_dst_len     = prefixlen,
4300                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4301                                   RTF_UP | RTF_PREF(pref),
4302                 .fc_protocol = RTPROT_RA,
4303                 .fc_type = RTN_UNICAST,
4304                 .fc_nlinfo.portid = 0,
4305                 .fc_nlinfo.nlh = NULL,
4306                 .fc_nlinfo.nl_net = net,
4307         };
4308
4309         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4310         cfg.fc_dst = *prefix;
4311         cfg.fc_gateway = *gwaddr;
4312
4313         /* We should treat it as a default route if prefix length is 0. */
4314         if (!prefixlen)
4315                 cfg.fc_flags |= RTF_DEFAULT;
4316
4317         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4318
4319         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4320 }
4321 #endif
4322
4323 struct fib6_info *rt6_get_dflt_router(struct net *net,
4324                                      const struct in6_addr *addr,
4325                                      struct net_device *dev)
4326 {
4327         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4328         struct fib6_info *rt;
4329         struct fib6_table *table;
4330
4331         table = fib6_get_table(net, tb_id);
4332         if (!table)
4333                 return NULL;
4334
4335         rcu_read_lock();
4336         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4337                 struct fib6_nh *nh;
4338
4339                 /* RA routes do not use nexthops */
4340                 if (rt->nh)
4341                         continue;
4342
4343                 nh = rt->fib6_nh;
4344                 if (dev == nh->fib_nh_dev &&
4345                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4346                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4347                         break;
4348         }
4349         if (rt && !fib6_info_hold_safe(rt))
4350                 rt = NULL;
4351         rcu_read_unlock();
4352         return rt;
4353 }
4354
4355 struct fib6_info *rt6_add_dflt_router(struct net *net,
4356                                      const struct in6_addr *gwaddr,
4357                                      struct net_device *dev,
4358                                      unsigned int pref,
4359                                      u32 defrtr_usr_metric)
4360 {
4361         struct fib6_config cfg = {
4362                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4363                 .fc_metric      = defrtr_usr_metric,
4364                 .fc_ifindex     = dev->ifindex,
4365                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4366                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4367                 .fc_protocol = RTPROT_RA,
4368                 .fc_type = RTN_UNICAST,
4369                 .fc_nlinfo.portid = 0,
4370                 .fc_nlinfo.nlh = NULL,
4371                 .fc_nlinfo.nl_net = net,
4372         };
4373
4374         cfg.fc_gateway = *gwaddr;
4375
4376         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4377                 struct fib6_table *table;
4378
4379                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4380                 if (table)
4381                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4382         }
4383
4384         return rt6_get_dflt_router(net, gwaddr, dev);
4385 }
4386
4387 static void __rt6_purge_dflt_routers(struct net *net,
4388                                      struct fib6_table *table)
4389 {
4390         struct fib6_info *rt;
4391
4392 restart:
4393         rcu_read_lock();
4394         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4395                 struct net_device *dev = fib6_info_nh_dev(rt);
4396                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4397
4398                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4399                     (!idev || idev->cnf.accept_ra != 2) &&
4400                     fib6_info_hold_safe(rt)) {
4401                         rcu_read_unlock();
4402                         ip6_del_rt(net, rt, false);
4403                         goto restart;
4404                 }
4405         }
4406         rcu_read_unlock();
4407
4408         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4409 }
4410
4411 void rt6_purge_dflt_routers(struct net *net)
4412 {
4413         struct fib6_table *table;
4414         struct hlist_head *head;
4415         unsigned int h;
4416
4417         rcu_read_lock();
4418
4419         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4420                 head = &net->ipv6.fib_table_hash[h];
4421                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4422                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4423                                 __rt6_purge_dflt_routers(net, table);
4424                 }
4425         }
4426
4427         rcu_read_unlock();
4428 }
4429
4430 static void rtmsg_to_fib6_config(struct net *net,
4431                                  struct in6_rtmsg *rtmsg,
4432                                  struct fib6_config *cfg)
4433 {
4434         *cfg = (struct fib6_config){
4435                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4436                          : RT6_TABLE_MAIN,
4437                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4438                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4439                 .fc_expires = rtmsg->rtmsg_info,
4440                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4441                 .fc_src_len = rtmsg->rtmsg_src_len,
4442                 .fc_flags = rtmsg->rtmsg_flags,
4443                 .fc_type = rtmsg->rtmsg_type,
4444
4445                 .fc_nlinfo.nl_net = net,
4446
4447                 .fc_dst = rtmsg->rtmsg_dst,
4448                 .fc_src = rtmsg->rtmsg_src,
4449                 .fc_gateway = rtmsg->rtmsg_gateway,
4450         };
4451 }
4452
4453 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4454 {
4455         struct fib6_config cfg;
4456         int err;
4457
4458         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4459                 return -EINVAL;
4460         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4461                 return -EPERM;
4462
4463         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4464
4465         rtnl_lock();
4466         switch (cmd) {
4467         case SIOCADDRT:
4468                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4469                 break;
4470         case SIOCDELRT:
4471                 err = ip6_route_del(&cfg, NULL);
4472                 break;
4473         }
4474         rtnl_unlock();
4475         return err;
4476 }
4477
4478 /*
4479  *      Drop the packet on the floor
4480  */
4481
4482 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4483 {
4484         struct dst_entry *dst = skb_dst(skb);
4485         struct net *net = dev_net(dst->dev);
4486         struct inet6_dev *idev;
4487         SKB_DR(reason);
4488         int type;
4489
4490         if (netif_is_l3_master(skb->dev) ||
4491             dst->dev == net->loopback_dev)
4492                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4493         else
4494                 idev = ip6_dst_idev(dst);
4495
4496         switch (ipstats_mib_noroutes) {
4497         case IPSTATS_MIB_INNOROUTES:
4498                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4499                 if (type == IPV6_ADDR_ANY) {
4500                         SKB_DR_SET(reason, IP_INADDRERRORS);
4501                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4502                         break;
4503                 }
4504                 SKB_DR_SET(reason, IP_INNOROUTES);
4505                 fallthrough;
4506         case IPSTATS_MIB_OUTNOROUTES:
4507                 SKB_DR_OR(reason, IP_OUTNOROUTES);
4508                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4509                 break;
4510         }
4511
4512         /* Start over by dropping the dst for l3mdev case */
4513         if (netif_is_l3_master(skb->dev))
4514                 skb_dst_drop(skb);
4515
4516         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4517         kfree_skb_reason(skb, reason);
4518         return 0;
4519 }
4520
4521 static int ip6_pkt_discard(struct sk_buff *skb)
4522 {
4523         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4524 }
4525
4526 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4527 {
4528         skb->dev = skb_dst(skb)->dev;
4529         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4530 }
4531
4532 static int ip6_pkt_prohibit(struct sk_buff *skb)
4533 {
4534         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4535 }
4536
4537 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4538 {
4539         skb->dev = skb_dst(skb)->dev;
4540         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4541 }
4542
4543 /*
4544  *      Allocate a dst for local (unicast / anycast) address.
4545  */
4546
4547 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4548                                      struct inet6_dev *idev,
4549                                      const struct in6_addr *addr,
4550                                      bool anycast, gfp_t gfp_flags)
4551 {
4552         struct fib6_config cfg = {
4553                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4554                 .fc_ifindex = idev->dev->ifindex,
4555                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4556                 .fc_dst = *addr,
4557                 .fc_dst_len = 128,
4558                 .fc_protocol = RTPROT_KERNEL,
4559                 .fc_nlinfo.nl_net = net,
4560                 .fc_ignore_dev_down = true,
4561         };
4562         struct fib6_info *f6i;
4563
4564         if (anycast) {
4565                 cfg.fc_type = RTN_ANYCAST;
4566                 cfg.fc_flags |= RTF_ANYCAST;
4567         } else {
4568                 cfg.fc_type = RTN_LOCAL;
4569                 cfg.fc_flags |= RTF_LOCAL;
4570         }
4571
4572         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4573         if (!IS_ERR(f6i)) {
4574                 f6i->dst_nocount = true;
4575
4576                 if (!anycast &&
4577                     (net->ipv6.devconf_all->disable_policy ||
4578                      idev->cnf.disable_policy))
4579                         f6i->dst_nopolicy = true;
4580         }
4581
4582         return f6i;
4583 }
4584
4585 /* remove deleted ip from prefsrc entries */
4586 struct arg_dev_net_ip {
4587         struct net_device *dev;
4588         struct net *net;
4589         struct in6_addr *addr;
4590 };
4591
4592 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4593 {
4594         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4595         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4596         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4597
4598         if (!rt->nh &&
4599             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4600             rt != net->ipv6.fib6_null_entry &&
4601             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4602                 spin_lock_bh(&rt6_exception_lock);
4603                 /* remove prefsrc entry */
4604                 rt->fib6_prefsrc.plen = 0;
4605                 spin_unlock_bh(&rt6_exception_lock);
4606         }
4607         return 0;
4608 }
4609
4610 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4611 {
4612         struct net *net = dev_net(ifp->idev->dev);
4613         struct arg_dev_net_ip adni = {
4614                 .dev = ifp->idev->dev,
4615                 .net = net,
4616                 .addr = &ifp->addr,
4617         };
4618         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4619 }
4620
4621 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4622
4623 /* Remove routers and update dst entries when gateway turn into host. */
4624 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4625 {
4626         struct in6_addr *gateway = (struct in6_addr *)arg;
4627         struct fib6_nh *nh;
4628
4629         /* RA routes do not use nexthops */
4630         if (rt->nh)
4631                 return 0;
4632
4633         nh = rt->fib6_nh;
4634         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4635             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4636                 return -1;
4637
4638         /* Further clean up cached routes in exception table.
4639          * This is needed because cached route may have a different
4640          * gateway than its 'parent' in the case of an ip redirect.
4641          */
4642         fib6_nh_exceptions_clean_tohost(nh, gateway);
4643
4644         return 0;
4645 }
4646
4647 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4648 {
4649         fib6_clean_all(net, fib6_clean_tohost, gateway);
4650 }
4651
4652 struct arg_netdev_event {
4653         const struct net_device *dev;
4654         union {
4655                 unsigned char nh_flags;
4656                 unsigned long event;
4657         };
4658 };
4659
4660 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4661 {
4662         struct fib6_info *iter;
4663         struct fib6_node *fn;
4664
4665         fn = rcu_dereference_protected(rt->fib6_node,
4666                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4667         iter = rcu_dereference_protected(fn->leaf,
4668                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4669         while (iter) {
4670                 if (iter->fib6_metric == rt->fib6_metric &&
4671                     rt6_qualify_for_ecmp(iter))
4672                         return iter;
4673                 iter = rcu_dereference_protected(iter->fib6_next,
4674                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4675         }
4676
4677         return NULL;
4678 }
4679
4680 /* only called for fib entries with builtin fib6_nh */
4681 static bool rt6_is_dead(const struct fib6_info *rt)
4682 {
4683         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4684             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4685              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4686                 return true;
4687
4688         return false;
4689 }
4690
4691 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4692 {
4693         struct fib6_info *iter;
4694         int total = 0;
4695
4696         if (!rt6_is_dead(rt))
4697                 total += rt->fib6_nh->fib_nh_weight;
4698
4699         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4700                 if (!rt6_is_dead(iter))
4701                         total += iter->fib6_nh->fib_nh_weight;
4702         }
4703
4704         return total;
4705 }
4706
4707 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4708 {
4709         int upper_bound = -1;
4710
4711         if (!rt6_is_dead(rt)) {
4712                 *weight += rt->fib6_nh->fib_nh_weight;
4713                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4714                                                     total) - 1;
4715         }
4716         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4717 }
4718
4719 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4720 {
4721         struct fib6_info *iter;
4722         int weight = 0;
4723
4724         rt6_upper_bound_set(rt, &weight, total);
4725
4726         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4727                 rt6_upper_bound_set(iter, &weight, total);
4728 }
4729
4730 void rt6_multipath_rebalance(struct fib6_info *rt)
4731 {
4732         struct fib6_info *first;
4733         int total;
4734
4735         /* In case the entire multipath route was marked for flushing,
4736          * then there is no need to rebalance upon the removal of every
4737          * sibling route.
4738          */
4739         if (!rt->fib6_nsiblings || rt->should_flush)
4740                 return;
4741
4742         /* During lookup routes are evaluated in order, so we need to
4743          * make sure upper bounds are assigned from the first sibling
4744          * onwards.
4745          */
4746         first = rt6_multipath_first_sibling(rt);
4747         if (WARN_ON_ONCE(!first))
4748                 return;
4749
4750         total = rt6_multipath_total_weight(first);
4751         rt6_multipath_upper_bound_set(first, total);
4752 }
4753
4754 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4755 {
4756         const struct arg_netdev_event *arg = p_arg;
4757         struct net *net = dev_net(arg->dev);
4758
4759         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4760             rt->fib6_nh->fib_nh_dev == arg->dev) {
4761                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4762                 fib6_update_sernum_upto_root(net, rt);
4763                 rt6_multipath_rebalance(rt);
4764         }
4765
4766         return 0;
4767 }
4768
4769 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4770 {
4771         struct arg_netdev_event arg = {
4772                 .dev = dev,
4773                 {
4774                         .nh_flags = nh_flags,
4775                 },
4776         };
4777
4778         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4779                 arg.nh_flags |= RTNH_F_LINKDOWN;
4780
4781         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4782 }
4783
4784 /* only called for fib entries with inline fib6_nh */
4785 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4786                                    const struct net_device *dev)
4787 {
4788         struct fib6_info *iter;
4789
4790         if (rt->fib6_nh->fib_nh_dev == dev)
4791                 return true;
4792         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4793                 if (iter->fib6_nh->fib_nh_dev == dev)
4794                         return true;
4795
4796         return false;
4797 }
4798
4799 static void rt6_multipath_flush(struct fib6_info *rt)
4800 {
4801         struct fib6_info *iter;
4802
4803         rt->should_flush = 1;
4804         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4805                 iter->should_flush = 1;
4806 }
4807
4808 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4809                                              const struct net_device *down_dev)
4810 {
4811         struct fib6_info *iter;
4812         unsigned int dead = 0;
4813
4814         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4815             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4816                 dead++;
4817         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4818                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4819                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4820                         dead++;
4821
4822         return dead;
4823 }
4824
4825 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4826                                        const struct net_device *dev,
4827                                        unsigned char nh_flags)
4828 {
4829         struct fib6_info *iter;
4830
4831         if (rt->fib6_nh->fib_nh_dev == dev)
4832                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4833         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4834                 if (iter->fib6_nh->fib_nh_dev == dev)
4835                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4836 }
4837
4838 /* called with write lock held for table with rt */
4839 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4840 {
4841         const struct arg_netdev_event *arg = p_arg;
4842         const struct net_device *dev = arg->dev;
4843         struct net *net = dev_net(dev);
4844
4845         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4846                 return 0;
4847
4848         switch (arg->event) {
4849         case NETDEV_UNREGISTER:
4850                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4851         case NETDEV_DOWN:
4852                 if (rt->should_flush)
4853                         return -1;
4854                 if (!rt->fib6_nsiblings)
4855                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4856                 if (rt6_multipath_uses_dev(rt, dev)) {
4857                         unsigned int count;
4858
4859                         count = rt6_multipath_dead_count(rt, dev);
4860                         if (rt->fib6_nsiblings + 1 == count) {
4861                                 rt6_multipath_flush(rt);
4862                                 return -1;
4863                         }
4864                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4865                                                    RTNH_F_LINKDOWN);
4866                         fib6_update_sernum(net, rt);
4867                         rt6_multipath_rebalance(rt);
4868                 }
4869                 return -2;
4870         case NETDEV_CHANGE:
4871                 if (rt->fib6_nh->fib_nh_dev != dev ||
4872                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4873                         break;
4874                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4875                 rt6_multipath_rebalance(rt);
4876                 break;
4877         }
4878
4879         return 0;
4880 }
4881
4882 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4883 {
4884         struct arg_netdev_event arg = {
4885                 .dev = dev,
4886                 {
4887                         .event = event,
4888                 },
4889         };
4890         struct net *net = dev_net(dev);
4891
4892         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4893                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4894         else
4895                 fib6_clean_all(net, fib6_ifdown, &arg);
4896 }
4897
4898 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4899 {
4900         rt6_sync_down_dev(dev, event);
4901         rt6_uncached_list_flush_dev(dev);
4902         neigh_ifdown(&nd_tbl, dev);
4903 }
4904
4905 struct rt6_mtu_change_arg {
4906         struct net_device *dev;
4907         unsigned int mtu;
4908         struct fib6_info *f6i;
4909 };
4910
4911 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4912 {
4913         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4914         struct fib6_info *f6i = arg->f6i;
4915
4916         /* For administrative MTU increase, there is no way to discover
4917          * IPv6 PMTU increase, so PMTU increase should be updated here.
4918          * Since RFC 1981 doesn't include administrative MTU increase
4919          * update PMTU increase is a MUST. (i.e. jumbo frame)
4920          */
4921         if (nh->fib_nh_dev == arg->dev) {
4922                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4923                 u32 mtu = f6i->fib6_pmtu;
4924
4925                 if (mtu >= arg->mtu ||
4926                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4927                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4928
4929                 spin_lock_bh(&rt6_exception_lock);
4930                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4931                 spin_unlock_bh(&rt6_exception_lock);
4932         }
4933
4934         return 0;
4935 }
4936
4937 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4938 {
4939         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4940         struct inet6_dev *idev;
4941
4942         /* In IPv6 pmtu discovery is not optional,
4943            so that RTAX_MTU lock cannot disable it.
4944            We still use this lock to block changes
4945            caused by addrconf/ndisc.
4946         */
4947
4948         idev = __in6_dev_get(arg->dev);
4949         if (!idev)
4950                 return 0;
4951
4952         if (fib6_metric_locked(f6i, RTAX_MTU))
4953                 return 0;
4954
4955         arg->f6i = f6i;
4956         if (f6i->nh) {
4957                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4958                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4959                                                 arg);
4960         }
4961
4962         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4963 }
4964
4965 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4966 {
4967         struct rt6_mtu_change_arg arg = {
4968                 .dev = dev,
4969                 .mtu = mtu,
4970         };
4971
4972         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4973 }
4974
4975 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4976         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4977         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4978         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4979         [RTA_OIF]               = { .type = NLA_U32 },
4980         [RTA_IIF]               = { .type = NLA_U32 },
4981         [RTA_PRIORITY]          = { .type = NLA_U32 },
4982         [RTA_METRICS]           = { .type = NLA_NESTED },
4983         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4984         [RTA_PREF]              = { .type = NLA_U8 },
4985         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4986         [RTA_ENCAP]             = { .type = NLA_NESTED },
4987         [RTA_EXPIRES]           = { .type = NLA_U32 },
4988         [RTA_UID]               = { .type = NLA_U32 },
4989         [RTA_MARK]              = { .type = NLA_U32 },
4990         [RTA_TABLE]             = { .type = NLA_U32 },
4991         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4992         [RTA_SPORT]             = { .type = NLA_U16 },
4993         [RTA_DPORT]             = { .type = NLA_U16 },
4994         [RTA_NH_ID]             = { .type = NLA_U32 },
4995 };
4996
4997 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4998                               struct fib6_config *cfg,
4999                               struct netlink_ext_ack *extack)
5000 {
5001         struct rtmsg *rtm;
5002         struct nlattr *tb[RTA_MAX+1];
5003         unsigned int pref;
5004         int err;
5005
5006         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5007                                      rtm_ipv6_policy, extack);
5008         if (err < 0)
5009                 goto errout;
5010
5011         err = -EINVAL;
5012         rtm = nlmsg_data(nlh);
5013
5014         if (rtm->rtm_tos) {
5015                 NL_SET_ERR_MSG(extack,
5016                                "Invalid dsfield (tos): option not available for IPv6");
5017                 goto errout;
5018         }
5019
5020         *cfg = (struct fib6_config){
5021                 .fc_table = rtm->rtm_table,
5022                 .fc_dst_len = rtm->rtm_dst_len,
5023                 .fc_src_len = rtm->rtm_src_len,
5024                 .fc_flags = RTF_UP,
5025                 .fc_protocol = rtm->rtm_protocol,
5026                 .fc_type = rtm->rtm_type,
5027
5028                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5029                 .fc_nlinfo.nlh = nlh,
5030                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5031         };
5032
5033         if (rtm->rtm_type == RTN_UNREACHABLE ||
5034             rtm->rtm_type == RTN_BLACKHOLE ||
5035             rtm->rtm_type == RTN_PROHIBIT ||
5036             rtm->rtm_type == RTN_THROW)
5037                 cfg->fc_flags |= RTF_REJECT;
5038
5039         if (rtm->rtm_type == RTN_LOCAL)
5040                 cfg->fc_flags |= RTF_LOCAL;
5041
5042         if (rtm->rtm_flags & RTM_F_CLONED)
5043                 cfg->fc_flags |= RTF_CACHE;
5044
5045         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5046
5047         if (tb[RTA_NH_ID]) {
5048                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5049                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5050                         NL_SET_ERR_MSG(extack,
5051                                        "Nexthop specification and nexthop id are mutually exclusive");
5052                         goto errout;
5053                 }
5054                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5055         }
5056
5057         if (tb[RTA_GATEWAY]) {
5058                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5059                 cfg->fc_flags |= RTF_GATEWAY;
5060         }
5061         if (tb[RTA_VIA]) {
5062                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5063                 goto errout;
5064         }
5065
5066         if (tb[RTA_DST]) {
5067                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5068
5069                 if (nla_len(tb[RTA_DST]) < plen)
5070                         goto errout;
5071
5072                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5073         }
5074
5075         if (tb[RTA_SRC]) {
5076                 int plen = (rtm->rtm_src_len + 7) >> 3;
5077
5078                 if (nla_len(tb[RTA_SRC]) < plen)
5079                         goto errout;
5080
5081                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5082         }
5083
5084         if (tb[RTA_PREFSRC])
5085                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5086
5087         if (tb[RTA_OIF])
5088                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5089
5090         if (tb[RTA_PRIORITY])
5091                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5092
5093         if (tb[RTA_METRICS]) {
5094                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5095                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5096         }
5097
5098         if (tb[RTA_TABLE])
5099                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5100
5101         if (tb[RTA_MULTIPATH]) {
5102                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5103                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5104
5105                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5106                                                      cfg->fc_mp_len, extack);
5107                 if (err < 0)
5108                         goto errout;
5109         }
5110
5111         if (tb[RTA_PREF]) {
5112                 pref = nla_get_u8(tb[RTA_PREF]);
5113                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5114                     pref != ICMPV6_ROUTER_PREF_HIGH)
5115                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5116                 cfg->fc_flags |= RTF_PREF(pref);
5117         }
5118
5119         if (tb[RTA_ENCAP])
5120                 cfg->fc_encap = tb[RTA_ENCAP];
5121
5122         if (tb[RTA_ENCAP_TYPE]) {
5123                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5124
5125                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5126                 if (err < 0)
5127                         goto errout;
5128         }
5129
5130         if (tb[RTA_EXPIRES]) {
5131                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5132
5133                 if (addrconf_finite_timeout(timeout)) {
5134                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5135                         cfg->fc_flags |= RTF_EXPIRES;
5136                 }
5137         }
5138
5139         err = 0;
5140 errout:
5141         return err;
5142 }
5143
5144 struct rt6_nh {
5145         struct fib6_info *fib6_info;
5146         struct fib6_config r_cfg;
5147         struct list_head next;
5148 };
5149
5150 static int ip6_route_info_append(struct net *net,
5151                                  struct list_head *rt6_nh_list,
5152                                  struct fib6_info *rt,
5153                                  struct fib6_config *r_cfg)
5154 {
5155         struct rt6_nh *nh;
5156         int err = -EEXIST;
5157
5158         list_for_each_entry(nh, rt6_nh_list, next) {
5159                 /* check if fib6_info already exists */
5160                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5161                         return err;
5162         }
5163
5164         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5165         if (!nh)
5166                 return -ENOMEM;
5167         nh->fib6_info = rt;
5168         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5169         list_add_tail(&nh->next, rt6_nh_list);
5170
5171         return 0;
5172 }
5173
5174 static void ip6_route_mpath_notify(struct fib6_info *rt,
5175                                    struct fib6_info *rt_last,
5176                                    struct nl_info *info,
5177                                    __u16 nlflags)
5178 {
5179         /* if this is an APPEND route, then rt points to the first route
5180          * inserted and rt_last points to last route inserted. Userspace
5181          * wants a consistent dump of the route which starts at the first
5182          * nexthop. Since sibling routes are always added at the end of
5183          * the list, find the first sibling of the last route appended
5184          */
5185         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5186                 rt = list_first_entry(&rt_last->fib6_siblings,
5187                                       struct fib6_info,
5188                                       fib6_siblings);
5189         }
5190
5191         if (rt)
5192                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5193 }
5194
5195 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5196 {
5197         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5198         bool should_notify = false;
5199         struct fib6_info *leaf;
5200         struct fib6_node *fn;
5201
5202         rcu_read_lock();
5203         fn = rcu_dereference(rt->fib6_node);
5204         if (!fn)
5205                 goto out;
5206
5207         leaf = rcu_dereference(fn->leaf);
5208         if (!leaf)
5209                 goto out;
5210
5211         if (rt == leaf ||
5212             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5213              rt6_qualify_for_ecmp(leaf)))
5214                 should_notify = true;
5215 out:
5216         rcu_read_unlock();
5217
5218         return should_notify;
5219 }
5220
5221 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5222                              struct netlink_ext_ack *extack)
5223 {
5224         if (nla_len(nla) < sizeof(*gw)) {
5225                 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5226                 return -EINVAL;
5227         }
5228
5229         *gw = nla_get_in6_addr(nla);
5230
5231         return 0;
5232 }
5233
5234 static int ip6_route_multipath_add(struct fib6_config *cfg,
5235                                    struct netlink_ext_ack *extack)
5236 {
5237         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5238         struct nl_info *info = &cfg->fc_nlinfo;
5239         struct fib6_config r_cfg;
5240         struct rtnexthop *rtnh;
5241         struct fib6_info *rt;
5242         struct rt6_nh *err_nh;
5243         struct rt6_nh *nh, *nh_safe;
5244         __u16 nlflags;
5245         int remaining;
5246         int attrlen;
5247         int err = 1;
5248         int nhn = 0;
5249         int replace = (cfg->fc_nlinfo.nlh &&
5250                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5251         LIST_HEAD(rt6_nh_list);
5252
5253         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5254         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5255                 nlflags |= NLM_F_APPEND;
5256
5257         remaining = cfg->fc_mp_len;
5258         rtnh = (struct rtnexthop *)cfg->fc_mp;
5259
5260         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5261          * fib6_info structs per nexthop
5262          */
5263         while (rtnh_ok(rtnh, remaining)) {
5264                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5265                 if (rtnh->rtnh_ifindex)
5266                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5267
5268                 attrlen = rtnh_attrlen(rtnh);
5269                 if (attrlen > 0) {
5270                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5271
5272                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5273                         if (nla) {
5274                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5275                                                         extack);
5276                                 if (err)
5277                                         goto cleanup;
5278
5279                                 r_cfg.fc_flags |= RTF_GATEWAY;
5280                         }
5281                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5282
5283                         /* RTA_ENCAP_TYPE length checked in
5284                          * lwtunnel_valid_encap_type_attr
5285                          */
5286                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5287                         if (nla)
5288                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5289                 }
5290
5291                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5292                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5293                 if (IS_ERR(rt)) {
5294                         err = PTR_ERR(rt);
5295                         rt = NULL;
5296                         goto cleanup;
5297                 }
5298                 if (!rt6_qualify_for_ecmp(rt)) {
5299                         err = -EINVAL;
5300                         NL_SET_ERR_MSG(extack,
5301                                        "Device only routes can not be added for IPv6 using the multipath API.");
5302                         fib6_info_release(rt);
5303                         goto cleanup;
5304                 }
5305
5306                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5307
5308                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5309                                             rt, &r_cfg);
5310                 if (err) {
5311                         fib6_info_release(rt);
5312                         goto cleanup;
5313                 }
5314
5315                 rtnh = rtnh_next(rtnh, &remaining);
5316         }
5317
5318         if (list_empty(&rt6_nh_list)) {
5319                 NL_SET_ERR_MSG(extack,
5320                                "Invalid nexthop configuration - no valid nexthops");
5321                 return -EINVAL;
5322         }
5323
5324         /* for add and replace send one notification with all nexthops.
5325          * Skip the notification in fib6_add_rt2node and send one with
5326          * the full route when done
5327          */
5328         info->skip_notify = 1;
5329
5330         /* For add and replace, send one notification with all nexthops. For
5331          * append, send one notification with all appended nexthops.
5332          */
5333         info->skip_notify_kernel = 1;
5334
5335         err_nh = NULL;
5336         list_for_each_entry(nh, &rt6_nh_list, next) {
5337                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5338
5339                 if (err) {
5340                         if (replace && nhn)
5341                                 NL_SET_ERR_MSG_MOD(extack,
5342                                                    "multipath route replace failed (check consistency of installed routes)");
5343                         err_nh = nh;
5344                         goto add_errout;
5345                 }
5346                 /* save reference to last route successfully inserted */
5347                 rt_last = nh->fib6_info;
5348
5349                 /* save reference to first route for notification */
5350                 if (!rt_notif)
5351                         rt_notif = nh->fib6_info;
5352
5353                 /* Because each route is added like a single route we remove
5354                  * these flags after the first nexthop: if there is a collision,
5355                  * we have already failed to add the first nexthop:
5356                  * fib6_add_rt2node() has rejected it; when replacing, old
5357                  * nexthops have been replaced by first new, the rest should
5358                  * be added to it.
5359                  */
5360                 if (cfg->fc_nlinfo.nlh) {
5361                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5362                                                              NLM_F_REPLACE);
5363                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5364                 }
5365                 nhn++;
5366         }
5367
5368         /* An in-kernel notification should only be sent in case the new
5369          * multipath route is added as the first route in the node, or if
5370          * it was appended to it. We pass 'rt_notif' since it is the first
5371          * sibling and might allow us to skip some checks in the replace case.
5372          */
5373         if (ip6_route_mpath_should_notify(rt_notif)) {
5374                 enum fib_event_type fib_event;
5375
5376                 if (rt_notif->fib6_nsiblings != nhn - 1)
5377                         fib_event = FIB_EVENT_ENTRY_APPEND;
5378                 else
5379                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5380
5381                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5382                                                           fib_event, rt_notif,
5383                                                           nhn - 1, extack);
5384                 if (err) {
5385                         /* Delete all the siblings that were just added */
5386                         err_nh = NULL;
5387                         goto add_errout;
5388                 }
5389         }
5390
5391         /* success ... tell user about new route */
5392         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5393         goto cleanup;
5394
5395 add_errout:
5396         /* send notification for routes that were added so that
5397          * the delete notifications sent by ip6_route_del are
5398          * coherent
5399          */
5400         if (rt_notif)
5401                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5402
5403         /* Delete routes that were already added */
5404         list_for_each_entry(nh, &rt6_nh_list, next) {
5405                 if (err_nh == nh)
5406                         break;
5407                 ip6_route_del(&nh->r_cfg, extack);
5408         }
5409
5410 cleanup:
5411         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5412                 fib6_info_release(nh->fib6_info);
5413                 list_del(&nh->next);
5414                 kfree(nh);
5415         }
5416
5417         return err;
5418 }
5419
5420 static int ip6_route_multipath_del(struct fib6_config *cfg,
5421                                    struct netlink_ext_ack *extack)
5422 {
5423         struct fib6_config r_cfg;
5424         struct rtnexthop *rtnh;
5425         int last_err = 0;
5426         int remaining;
5427         int attrlen;
5428         int err;
5429
5430         remaining = cfg->fc_mp_len;
5431         rtnh = (struct rtnexthop *)cfg->fc_mp;
5432
5433         /* Parse a Multipath Entry */
5434         while (rtnh_ok(rtnh, remaining)) {
5435                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5436                 if (rtnh->rtnh_ifindex)
5437                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5438
5439                 attrlen = rtnh_attrlen(rtnh);
5440                 if (attrlen > 0) {
5441                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5442
5443                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5444                         if (nla) {
5445                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5446                                                         extack);
5447                                 if (err) {
5448                                         last_err = err;
5449                                         goto next_rtnh;
5450                                 }
5451
5452                                 r_cfg.fc_flags |= RTF_GATEWAY;
5453                         }
5454                 }
5455                 err = ip6_route_del(&r_cfg, extack);
5456                 if (err)
5457                         last_err = err;
5458
5459 next_rtnh:
5460                 rtnh = rtnh_next(rtnh, &remaining);
5461         }
5462
5463         return last_err;
5464 }
5465
5466 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5467                               struct netlink_ext_ack *extack)
5468 {
5469         struct fib6_config cfg;
5470         int err;
5471
5472         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5473         if (err < 0)
5474                 return err;
5475
5476         if (cfg.fc_nh_id &&
5477             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5478                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5479                 return -EINVAL;
5480         }
5481
5482         if (cfg.fc_mp)
5483                 return ip6_route_multipath_del(&cfg, extack);
5484         else {
5485                 cfg.fc_delete_all_nh = 1;
5486                 return ip6_route_del(&cfg, extack);
5487         }
5488 }
5489
5490 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5491                               struct netlink_ext_ack *extack)
5492 {
5493         struct fib6_config cfg;
5494         int err;
5495
5496         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5497         if (err < 0)
5498                 return err;
5499
5500         if (cfg.fc_metric == 0)
5501                 cfg.fc_metric = IP6_RT_PRIO_USER;
5502
5503         if (cfg.fc_mp)
5504                 return ip6_route_multipath_add(&cfg, extack);
5505         else
5506                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5507 }
5508
5509 /* add the overhead of this fib6_nh to nexthop_len */
5510 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5511 {
5512         int *nexthop_len = arg;
5513
5514         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5515                      + NLA_ALIGN(sizeof(struct rtnexthop))
5516                      + nla_total_size(16); /* RTA_GATEWAY */
5517
5518         if (nh->fib_nh_lws) {
5519                 /* RTA_ENCAP_TYPE */
5520                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5521                 /* RTA_ENCAP */
5522                 *nexthop_len += nla_total_size(2);
5523         }
5524
5525         return 0;
5526 }
5527
5528 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5529 {
5530         int nexthop_len;
5531
5532         if (f6i->nh) {
5533                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5534                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5535                                          &nexthop_len);
5536         } else {
5537                 struct fib6_info *sibling, *next_sibling;
5538                 struct fib6_nh *nh = f6i->fib6_nh;
5539
5540                 nexthop_len = 0;
5541                 if (f6i->fib6_nsiblings) {
5542                         rt6_nh_nlmsg_size(nh, &nexthop_len);
5543
5544                         list_for_each_entry_safe(sibling, next_sibling,
5545                                                  &f6i->fib6_siblings, fib6_siblings) {
5546                                 rt6_nh_nlmsg_size(sibling->fib6_nh, &nexthop_len);
5547                         }
5548                 }
5549                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5550         }
5551
5552         return NLMSG_ALIGN(sizeof(struct rtmsg))
5553                + nla_total_size(16) /* RTA_SRC */
5554                + nla_total_size(16) /* RTA_DST */
5555                + nla_total_size(16) /* RTA_GATEWAY */
5556                + nla_total_size(16) /* RTA_PREFSRC */
5557                + nla_total_size(4) /* RTA_TABLE */
5558                + nla_total_size(4) /* RTA_IIF */
5559                + nla_total_size(4) /* RTA_OIF */
5560                + nla_total_size(4) /* RTA_PRIORITY */
5561                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5562                + nla_total_size(sizeof(struct rta_cacheinfo))
5563                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5564                + nla_total_size(1) /* RTA_PREF */
5565                + nexthop_len;
5566 }
5567
5568 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5569                                  unsigned char *flags)
5570 {
5571         if (nexthop_is_multipath(nh)) {
5572                 struct nlattr *mp;
5573
5574                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5575                 if (!mp)
5576                         goto nla_put_failure;
5577
5578                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5579                         goto nla_put_failure;
5580
5581                 nla_nest_end(skb, mp);
5582         } else {
5583                 struct fib6_nh *fib6_nh;
5584
5585                 fib6_nh = nexthop_fib6_nh(nh);
5586                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5587                                      flags, false) < 0)
5588                         goto nla_put_failure;
5589         }
5590
5591         return 0;
5592
5593 nla_put_failure:
5594         return -EMSGSIZE;
5595 }
5596
5597 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5598                          struct fib6_info *rt, struct dst_entry *dst,
5599                          struct in6_addr *dest, struct in6_addr *src,
5600                          int iif, int type, u32 portid, u32 seq,
5601                          unsigned int flags)
5602 {
5603         struct rt6_info *rt6 = (struct rt6_info *)dst;
5604         struct rt6key *rt6_dst, *rt6_src;
5605         u32 *pmetrics, table, rt6_flags;
5606         unsigned char nh_flags = 0;
5607         struct nlmsghdr *nlh;
5608         struct rtmsg *rtm;
5609         long expires = 0;
5610
5611         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5612         if (!nlh)
5613                 return -EMSGSIZE;
5614
5615         if (rt6) {
5616                 rt6_dst = &rt6->rt6i_dst;
5617                 rt6_src = &rt6->rt6i_src;
5618                 rt6_flags = rt6->rt6i_flags;
5619         } else {
5620                 rt6_dst = &rt->fib6_dst;
5621                 rt6_src = &rt->fib6_src;
5622                 rt6_flags = rt->fib6_flags;
5623         }
5624
5625         rtm = nlmsg_data(nlh);
5626         rtm->rtm_family = AF_INET6;
5627         rtm->rtm_dst_len = rt6_dst->plen;
5628         rtm->rtm_src_len = rt6_src->plen;
5629         rtm->rtm_tos = 0;
5630         if (rt->fib6_table)
5631                 table = rt->fib6_table->tb6_id;
5632         else
5633                 table = RT6_TABLE_UNSPEC;
5634         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5635         if (nla_put_u32(skb, RTA_TABLE, table))
5636                 goto nla_put_failure;
5637
5638         rtm->rtm_type = rt->fib6_type;
5639         rtm->rtm_flags = 0;
5640         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5641         rtm->rtm_protocol = rt->fib6_protocol;
5642
5643         if (rt6_flags & RTF_CACHE)
5644                 rtm->rtm_flags |= RTM_F_CLONED;
5645
5646         if (dest) {
5647                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5648                         goto nla_put_failure;
5649                 rtm->rtm_dst_len = 128;
5650         } else if (rtm->rtm_dst_len)
5651                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5652                         goto nla_put_failure;
5653 #ifdef CONFIG_IPV6_SUBTREES
5654         if (src) {
5655                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5656                         goto nla_put_failure;
5657                 rtm->rtm_src_len = 128;
5658         } else if (rtm->rtm_src_len &&
5659                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5660                 goto nla_put_failure;
5661 #endif
5662         if (iif) {
5663 #ifdef CONFIG_IPV6_MROUTE
5664                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5665                         int err = ip6mr_get_route(net, skb, rtm, portid);
5666
5667                         if (err == 0)
5668                                 return 0;
5669                         if (err < 0)
5670                                 goto nla_put_failure;
5671                 } else
5672 #endif
5673                         if (nla_put_u32(skb, RTA_IIF, iif))
5674                                 goto nla_put_failure;
5675         } else if (dest) {
5676                 struct in6_addr saddr_buf;
5677                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5678                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5679                         goto nla_put_failure;
5680         }
5681
5682         if (rt->fib6_prefsrc.plen) {
5683                 struct in6_addr saddr_buf;
5684                 saddr_buf = rt->fib6_prefsrc.addr;
5685                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5686                         goto nla_put_failure;
5687         }
5688
5689         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5690         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5691                 goto nla_put_failure;
5692
5693         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5694                 goto nla_put_failure;
5695
5696         /* For multipath routes, walk the siblings list and add
5697          * each as a nexthop within RTA_MULTIPATH.
5698          */
5699         if (rt6) {
5700                 if (rt6_flags & RTF_GATEWAY &&
5701                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5702                         goto nla_put_failure;
5703
5704                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5705                         goto nla_put_failure;
5706
5707                 if (dst->lwtstate &&
5708                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5709                         goto nla_put_failure;
5710         } else if (rt->fib6_nsiblings) {
5711                 struct fib6_info *sibling, *next_sibling;
5712                 struct nlattr *mp;
5713
5714                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5715                 if (!mp)
5716                         goto nla_put_failure;
5717
5718                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5719                                     rt->fib6_nh->fib_nh_weight, AF_INET6,
5720                                     0) < 0)
5721                         goto nla_put_failure;
5722
5723                 list_for_each_entry_safe(sibling, next_sibling,
5724                                          &rt->fib6_siblings, fib6_siblings) {
5725                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5726                                             sibling->fib6_nh->fib_nh_weight,
5727                                             AF_INET6, 0) < 0)
5728                                 goto nla_put_failure;
5729                 }
5730
5731                 nla_nest_end(skb, mp);
5732         } else if (rt->nh) {
5733                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5734                         goto nla_put_failure;
5735
5736                 if (nexthop_is_blackhole(rt->nh))
5737                         rtm->rtm_type = RTN_BLACKHOLE;
5738
5739                 if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5740                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5741                         goto nla_put_failure;
5742
5743                 rtm->rtm_flags |= nh_flags;
5744         } else {
5745                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5746                                      &nh_flags, false) < 0)
5747                         goto nla_put_failure;
5748
5749                 rtm->rtm_flags |= nh_flags;
5750         }
5751
5752         if (rt6_flags & RTF_EXPIRES) {
5753                 expires = dst ? dst->expires : rt->expires;
5754                 expires -= jiffies;
5755         }
5756
5757         if (!dst) {
5758                 if (READ_ONCE(rt->offload))
5759                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5760                 if (READ_ONCE(rt->trap))
5761                         rtm->rtm_flags |= RTM_F_TRAP;
5762                 if (READ_ONCE(rt->offload_failed))
5763                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5764         }
5765
5766         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5767                 goto nla_put_failure;
5768
5769         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5770                 goto nla_put_failure;
5771
5772
5773         nlmsg_end(skb, nlh);
5774         return 0;
5775
5776 nla_put_failure:
5777         nlmsg_cancel(skb, nlh);
5778         return -EMSGSIZE;
5779 }
5780
5781 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5782 {
5783         const struct net_device *dev = arg;
5784
5785         if (nh->fib_nh_dev == dev)
5786                 return 1;
5787
5788         return 0;
5789 }
5790
5791 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5792                                const struct net_device *dev)
5793 {
5794         if (f6i->nh) {
5795                 struct net_device *_dev = (struct net_device *)dev;
5796
5797                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5798                                                   fib6_info_nh_uses_dev,
5799                                                   _dev);
5800         }
5801
5802         if (f6i->fib6_nh->fib_nh_dev == dev)
5803                 return true;
5804
5805         if (f6i->fib6_nsiblings) {
5806                 struct fib6_info *sibling, *next_sibling;
5807
5808                 list_for_each_entry_safe(sibling, next_sibling,
5809                                          &f6i->fib6_siblings, fib6_siblings) {
5810                         if (sibling->fib6_nh->fib_nh_dev == dev)
5811                                 return true;
5812                 }
5813         }
5814
5815         return false;
5816 }
5817
5818 struct fib6_nh_exception_dump_walker {
5819         struct rt6_rtnl_dump_arg *dump;
5820         struct fib6_info *rt;
5821         unsigned int flags;
5822         unsigned int skip;
5823         unsigned int count;
5824 };
5825
5826 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5827 {
5828         struct fib6_nh_exception_dump_walker *w = arg;
5829         struct rt6_rtnl_dump_arg *dump = w->dump;
5830         struct rt6_exception_bucket *bucket;
5831         struct rt6_exception *rt6_ex;
5832         int i, err;
5833
5834         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5835         if (!bucket)
5836                 return 0;
5837
5838         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5839                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5840                         if (w->skip) {
5841                                 w->skip--;
5842                                 continue;
5843                         }
5844
5845                         /* Expiration of entries doesn't bump sernum, insertion
5846                          * does. Removal is triggered by insertion, so we can
5847                          * rely on the fact that if entries change between two
5848                          * partial dumps, this node is scanned again completely,
5849                          * see rt6_insert_exception() and fib6_dump_table().
5850                          *
5851                          * Count expired entries we go through as handled
5852                          * entries that we'll skip next time, in case of partial
5853                          * node dump. Otherwise, if entries expire meanwhile,
5854                          * we'll skip the wrong amount.
5855                          */
5856                         if (rt6_check_expired(rt6_ex->rt6i)) {
5857                                 w->count++;
5858                                 continue;
5859                         }
5860
5861                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5862                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5863                                             RTM_NEWROUTE,
5864                                             NETLINK_CB(dump->cb->skb).portid,
5865                                             dump->cb->nlh->nlmsg_seq, w->flags);
5866                         if (err)
5867                                 return err;
5868
5869                         w->count++;
5870                 }
5871                 bucket++;
5872         }
5873
5874         return 0;
5875 }
5876
5877 /* Return -1 if done with node, number of handled routes on partial dump */
5878 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5879 {
5880         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5881         struct fib_dump_filter *filter = &arg->filter;
5882         unsigned int flags = NLM_F_MULTI;
5883         struct net *net = arg->net;
5884         int count = 0;
5885
5886         if (rt == net->ipv6.fib6_null_entry)
5887                 return -1;
5888
5889         if ((filter->flags & RTM_F_PREFIX) &&
5890             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5891                 /* success since this is not a prefix route */
5892                 return -1;
5893         }
5894         if (filter->filter_set &&
5895             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5896              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5897              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5898                 return -1;
5899         }
5900
5901         if (filter->filter_set ||
5902             !filter->dump_routes || !filter->dump_exceptions) {
5903                 flags |= NLM_F_DUMP_FILTERED;
5904         }
5905
5906         if (filter->dump_routes) {
5907                 if (skip) {
5908                         skip--;
5909                 } else {
5910                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5911                                           0, RTM_NEWROUTE,
5912                                           NETLINK_CB(arg->cb->skb).portid,
5913                                           arg->cb->nlh->nlmsg_seq, flags)) {
5914                                 return 0;
5915                         }
5916                         count++;
5917                 }
5918         }
5919
5920         if (filter->dump_exceptions) {
5921                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5922                                                            .rt = rt,
5923                                                            .flags = flags,
5924                                                            .skip = skip,
5925                                                            .count = 0 };
5926                 int err;
5927
5928                 rcu_read_lock();
5929                 if (rt->nh) {
5930                         err = nexthop_for_each_fib6_nh(rt->nh,
5931                                                        rt6_nh_dump_exceptions,
5932                                                        &w);
5933                 } else {
5934                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5935                 }
5936                 rcu_read_unlock();
5937
5938                 if (err)
5939                         return count + w.count;
5940         }
5941
5942         return -1;
5943 }
5944
5945 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5946                                         const struct nlmsghdr *nlh,
5947                                         struct nlattr **tb,
5948                                         struct netlink_ext_ack *extack)
5949 {
5950         struct rtmsg *rtm;
5951         int i, err;
5952
5953         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5954                 NL_SET_ERR_MSG_MOD(extack,
5955                                    "Invalid header for get route request");
5956                 return -EINVAL;
5957         }
5958
5959         if (!netlink_strict_get_check(skb))
5960                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5961                                               rtm_ipv6_policy, extack);
5962
5963         rtm = nlmsg_data(nlh);
5964         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5965             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5966             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5967             rtm->rtm_type) {
5968                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5969                 return -EINVAL;
5970         }
5971         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5972                 NL_SET_ERR_MSG_MOD(extack,
5973                                    "Invalid flags for get route request");
5974                 return -EINVAL;
5975         }
5976
5977         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5978                                             rtm_ipv6_policy, extack);
5979         if (err)
5980                 return err;
5981
5982         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5983             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5984                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5985                 return -EINVAL;
5986         }
5987
5988         for (i = 0; i <= RTA_MAX; i++) {
5989                 if (!tb[i])
5990                         continue;
5991
5992                 switch (i) {
5993                 case RTA_SRC:
5994                 case RTA_DST:
5995                 case RTA_IIF:
5996                 case RTA_OIF:
5997                 case RTA_MARK:
5998                 case RTA_UID:
5999                 case RTA_SPORT:
6000                 case RTA_DPORT:
6001                 case RTA_IP_PROTO:
6002                         break;
6003                 default:
6004                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6005                         return -EINVAL;
6006                 }
6007         }
6008
6009         return 0;
6010 }
6011
6012 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6013                               struct netlink_ext_ack *extack)
6014 {
6015         struct net *net = sock_net(in_skb->sk);
6016         struct nlattr *tb[RTA_MAX+1];
6017         int err, iif = 0, oif = 0;
6018         struct fib6_info *from;
6019         struct dst_entry *dst;
6020         struct rt6_info *rt;
6021         struct sk_buff *skb;
6022         struct rtmsg *rtm;
6023         struct flowi6 fl6 = {};
6024         bool fibmatch;
6025
6026         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6027         if (err < 0)
6028                 goto errout;
6029
6030         err = -EINVAL;
6031         rtm = nlmsg_data(nlh);
6032         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6033         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6034
6035         if (tb[RTA_SRC]) {
6036                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6037                         goto errout;
6038
6039                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6040         }
6041
6042         if (tb[RTA_DST]) {
6043                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6044                         goto errout;
6045
6046                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6047         }
6048
6049         if (tb[RTA_IIF])
6050                 iif = nla_get_u32(tb[RTA_IIF]);
6051
6052         if (tb[RTA_OIF])
6053                 oif = nla_get_u32(tb[RTA_OIF]);
6054
6055         if (tb[RTA_MARK])
6056                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6057
6058         if (tb[RTA_UID])
6059                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6060                                            nla_get_u32(tb[RTA_UID]));
6061         else
6062                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6063
6064         if (tb[RTA_SPORT])
6065                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6066
6067         if (tb[RTA_DPORT])
6068                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6069
6070         if (tb[RTA_IP_PROTO]) {
6071                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6072                                                   &fl6.flowi6_proto, AF_INET6,
6073                                                   extack);
6074                 if (err)
6075                         goto errout;
6076         }
6077
6078         if (iif) {
6079                 struct net_device *dev;
6080                 int flags = 0;
6081
6082                 rcu_read_lock();
6083
6084                 dev = dev_get_by_index_rcu(net, iif);
6085                 if (!dev) {
6086                         rcu_read_unlock();
6087                         err = -ENODEV;
6088                         goto errout;
6089                 }
6090
6091                 fl6.flowi6_iif = iif;
6092
6093                 if (!ipv6_addr_any(&fl6.saddr))
6094                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6095
6096                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6097
6098                 rcu_read_unlock();
6099         } else {
6100                 fl6.flowi6_oif = oif;
6101
6102                 dst = ip6_route_output(net, NULL, &fl6);
6103         }
6104
6105
6106         rt = container_of(dst, struct rt6_info, dst);
6107         if (rt->dst.error) {
6108                 err = rt->dst.error;
6109                 ip6_rt_put(rt);
6110                 goto errout;
6111         }
6112
6113         if (rt == net->ipv6.ip6_null_entry) {
6114                 err = rt->dst.error;
6115                 ip6_rt_put(rt);
6116                 goto errout;
6117         }
6118
6119         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6120         if (!skb) {
6121                 ip6_rt_put(rt);
6122                 err = -ENOBUFS;
6123                 goto errout;
6124         }
6125
6126         skb_dst_set(skb, &rt->dst);
6127
6128         rcu_read_lock();
6129         from = rcu_dereference(rt->from);
6130         if (from) {
6131                 if (fibmatch)
6132                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6133                                             iif, RTM_NEWROUTE,
6134                                             NETLINK_CB(in_skb).portid,
6135                                             nlh->nlmsg_seq, 0);
6136                 else
6137                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6138                                             &fl6.saddr, iif, RTM_NEWROUTE,
6139                                             NETLINK_CB(in_skb).portid,
6140                                             nlh->nlmsg_seq, 0);
6141         } else {
6142                 err = -ENETUNREACH;
6143         }
6144         rcu_read_unlock();
6145
6146         if (err < 0) {
6147                 kfree_skb(skb);
6148                 goto errout;
6149         }
6150
6151         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6152 errout:
6153         return err;
6154 }
6155
6156 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6157                      unsigned int nlm_flags)
6158 {
6159         struct sk_buff *skb;
6160         struct net *net = info->nl_net;
6161         u32 seq;
6162         int err;
6163
6164         err = -ENOBUFS;
6165         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6166
6167         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6168         if (!skb)
6169                 goto errout;
6170
6171         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6172                             event, info->portid, seq, nlm_flags);
6173         if (err < 0) {
6174                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6175                 WARN_ON(err == -EMSGSIZE);
6176                 kfree_skb(skb);
6177                 goto errout;
6178         }
6179         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6180                     info->nlh, gfp_any());
6181         return;
6182 errout:
6183         if (err < 0)
6184                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6185 }
6186
6187 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6188                     struct nl_info *info)
6189 {
6190         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6191         struct sk_buff *skb;
6192         int err = -ENOBUFS;
6193
6194         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6195         if (!skb)
6196                 goto errout;
6197
6198         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6199                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6200         if (err < 0) {
6201                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6202                 WARN_ON(err == -EMSGSIZE);
6203                 kfree_skb(skb);
6204                 goto errout;
6205         }
6206         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6207                     info->nlh, gfp_any());
6208         return;
6209 errout:
6210         if (err < 0)
6211                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6212 }
6213
6214 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6215                             bool offload, bool trap, bool offload_failed)
6216 {
6217         struct sk_buff *skb;
6218         int err;
6219
6220         if (READ_ONCE(f6i->offload) == offload &&
6221             READ_ONCE(f6i->trap) == trap &&
6222             READ_ONCE(f6i->offload_failed) == offload_failed)
6223                 return;
6224
6225         WRITE_ONCE(f6i->offload, offload);
6226         WRITE_ONCE(f6i->trap, trap);
6227
6228         /* 2 means send notifications only if offload_failed was changed. */
6229         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6230             READ_ONCE(f6i->offload_failed) == offload_failed)
6231                 return;
6232
6233         WRITE_ONCE(f6i->offload_failed, offload_failed);
6234
6235         if (!rcu_access_pointer(f6i->fib6_node))
6236                 /* The route was removed from the tree, do not send
6237                  * notification.
6238                  */
6239                 return;
6240
6241         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6242                 return;
6243
6244         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6245         if (!skb) {
6246                 err = -ENOBUFS;
6247                 goto errout;
6248         }
6249
6250         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6251                             0, 0);
6252         if (err < 0) {
6253                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6254                 WARN_ON(err == -EMSGSIZE);
6255                 kfree_skb(skb);
6256                 goto errout;
6257         }
6258
6259         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6260         return;
6261
6262 errout:
6263         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6264 }
6265 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6266
6267 static int ip6_route_dev_notify(struct notifier_block *this,
6268                                 unsigned long event, void *ptr)
6269 {
6270         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6271         struct net *net = dev_net(dev);
6272
6273         if (!(dev->flags & IFF_LOOPBACK))
6274                 return NOTIFY_OK;
6275
6276         if (event == NETDEV_REGISTER) {
6277                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6278                 net->ipv6.ip6_null_entry->dst.dev = dev;
6279                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6280 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6281                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6282                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6283                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6284                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6285 #endif
6286          } else if (event == NETDEV_UNREGISTER &&
6287                     dev->reg_state != NETREG_UNREGISTERED) {
6288                 /* NETDEV_UNREGISTER could be fired for multiple times by
6289                  * netdev_wait_allrefs(). Make sure we only call this once.
6290                  */
6291                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6292 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6293                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6294                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6295 #endif
6296         }
6297
6298         return NOTIFY_OK;
6299 }
6300
6301 /*
6302  *      /proc
6303  */
6304
6305 #ifdef CONFIG_PROC_FS
6306 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6307 {
6308         struct net *net = (struct net *)seq->private;
6309         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6310                    net->ipv6.rt6_stats->fib_nodes,
6311                    net->ipv6.rt6_stats->fib_route_nodes,
6312                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6313                    net->ipv6.rt6_stats->fib_rt_entries,
6314                    net->ipv6.rt6_stats->fib_rt_cache,
6315                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6316                    net->ipv6.rt6_stats->fib_discarded_routes);
6317
6318         return 0;
6319 }
6320 #endif  /* CONFIG_PROC_FS */
6321
6322 #ifdef CONFIG_SYSCTL
6323
6324 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6325                               void *buffer, size_t *lenp, loff_t *ppos)
6326 {
6327         struct net *net;
6328         int delay;
6329         int ret;
6330         if (!write)
6331                 return -EINVAL;
6332
6333         net = (struct net *)ctl->extra1;
6334         delay = net->ipv6.sysctl.flush_delay;
6335         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6336         if (ret)
6337                 return ret;
6338
6339         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6340         return 0;
6341 }
6342
6343 static struct ctl_table ipv6_route_table_template[] = {
6344         {
6345                 .procname       =       "max_size",
6346                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6347                 .maxlen         =       sizeof(int),
6348                 .mode           =       0644,
6349                 .proc_handler   =       proc_dointvec,
6350         },
6351         {
6352                 .procname       =       "gc_thresh",
6353                 .data           =       &ip6_dst_ops_template.gc_thresh,
6354                 .maxlen         =       sizeof(int),
6355                 .mode           =       0644,
6356                 .proc_handler   =       proc_dointvec,
6357         },
6358         {
6359                 .procname       =       "flush",
6360                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6361                 .maxlen         =       sizeof(int),
6362                 .mode           =       0200,
6363                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6364         },
6365         {
6366                 .procname       =       "gc_min_interval",
6367                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6368                 .maxlen         =       sizeof(int),
6369                 .mode           =       0644,
6370                 .proc_handler   =       proc_dointvec_jiffies,
6371         },
6372         {
6373                 .procname       =       "gc_timeout",
6374                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6375                 .maxlen         =       sizeof(int),
6376                 .mode           =       0644,
6377                 .proc_handler   =       proc_dointvec_jiffies,
6378         },
6379         {
6380                 .procname       =       "gc_interval",
6381                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6382                 .maxlen         =       sizeof(int),
6383                 .mode           =       0644,
6384                 .proc_handler   =       proc_dointvec_jiffies,
6385         },
6386         {
6387                 .procname       =       "gc_elasticity",
6388                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6389                 .maxlen         =       sizeof(int),
6390                 .mode           =       0644,
6391                 .proc_handler   =       proc_dointvec,
6392         },
6393         {
6394                 .procname       =       "mtu_expires",
6395                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6396                 .maxlen         =       sizeof(int),
6397                 .mode           =       0644,
6398                 .proc_handler   =       proc_dointvec_jiffies,
6399         },
6400         {
6401                 .procname       =       "min_adv_mss",
6402                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6403                 .maxlen         =       sizeof(int),
6404                 .mode           =       0644,
6405                 .proc_handler   =       proc_dointvec,
6406         },
6407         {
6408                 .procname       =       "gc_min_interval_ms",
6409                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6410                 .maxlen         =       sizeof(int),
6411                 .mode           =       0644,
6412                 .proc_handler   =       proc_dointvec_ms_jiffies,
6413         },
6414         {
6415                 .procname       =       "skip_notify_on_dev_down",
6416                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6417                 .maxlen         =       sizeof(int),
6418                 .mode           =       0644,
6419                 .proc_handler   =       proc_dointvec_minmax,
6420                 .extra1         =       SYSCTL_ZERO,
6421                 .extra2         =       SYSCTL_ONE,
6422         },
6423         { }
6424 };
6425
6426 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6427 {
6428         struct ctl_table *table;
6429
6430         table = kmemdup(ipv6_route_table_template,
6431                         sizeof(ipv6_route_table_template),
6432                         GFP_KERNEL);
6433
6434         if (table) {
6435                 table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6436                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6437                 table[2].data = &net->ipv6.sysctl.flush_delay;
6438                 table[2].extra1 = net;
6439                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6440                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6441                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6442                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6443                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6444                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6445                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6446                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6447
6448                 /* Don't export sysctls to unprivileged users */
6449                 if (net->user_ns != &init_user_ns)
6450                         table[1].procname = NULL;
6451         }
6452
6453         return table;
6454 }
6455 #endif
6456
6457 static int __net_init ip6_route_net_init(struct net *net)
6458 {
6459         int ret = -ENOMEM;
6460
6461         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6462                sizeof(net->ipv6.ip6_dst_ops));
6463
6464         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6465                 goto out_ip6_dst_ops;
6466
6467         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6468         if (!net->ipv6.fib6_null_entry)
6469                 goto out_ip6_dst_entries;
6470         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6471                sizeof(*net->ipv6.fib6_null_entry));
6472
6473         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6474                                            sizeof(*net->ipv6.ip6_null_entry),
6475                                            GFP_KERNEL);
6476         if (!net->ipv6.ip6_null_entry)
6477                 goto out_fib6_null_entry;
6478         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6479         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6480                          ip6_template_metrics, true);
6481         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6482
6483 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6484         net->ipv6.fib6_has_custom_rules = false;
6485         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6486                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6487                                                GFP_KERNEL);
6488         if (!net->ipv6.ip6_prohibit_entry)
6489                 goto out_ip6_null_entry;
6490         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6491         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6492                          ip6_template_metrics, true);
6493         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6494
6495         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6496                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6497                                                GFP_KERNEL);
6498         if (!net->ipv6.ip6_blk_hole_entry)
6499                 goto out_ip6_prohibit_entry;
6500         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6501         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6502                          ip6_template_metrics, true);
6503         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6504 #ifdef CONFIG_IPV6_SUBTREES
6505         net->ipv6.fib6_routes_require_src = 0;
6506 #endif
6507 #endif
6508
6509         net->ipv6.sysctl.flush_delay = 0;
6510         net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
6511         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6512         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6513         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6514         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6515         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6516         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6517         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6518
6519         atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6520
6521         ret = 0;
6522 out:
6523         return ret;
6524
6525 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6526 out_ip6_prohibit_entry:
6527         kfree(net->ipv6.ip6_prohibit_entry);
6528 out_ip6_null_entry:
6529         kfree(net->ipv6.ip6_null_entry);
6530 #endif
6531 out_fib6_null_entry:
6532         kfree(net->ipv6.fib6_null_entry);
6533 out_ip6_dst_entries:
6534         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6535 out_ip6_dst_ops:
6536         goto out;
6537 }
6538
6539 static void __net_exit ip6_route_net_exit(struct net *net)
6540 {
6541         kfree(net->ipv6.fib6_null_entry);
6542         kfree(net->ipv6.ip6_null_entry);
6543 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6544         kfree(net->ipv6.ip6_prohibit_entry);
6545         kfree(net->ipv6.ip6_blk_hole_entry);
6546 #endif
6547         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6548 }
6549
6550 static int __net_init ip6_route_net_init_late(struct net *net)
6551 {
6552 #ifdef CONFIG_PROC_FS
6553         if (!proc_create_net("ipv6_route", 0, net->proc_net,
6554                              &ipv6_route_seq_ops,
6555                              sizeof(struct ipv6_route_iter)))
6556                 return -ENOMEM;
6557
6558         if (!proc_create_net_single("rt6_stats", 0444, net->proc_net,
6559                                     rt6_stats_seq_show, NULL)) {
6560                 remove_proc_entry("ipv6_route", net->proc_net);
6561                 return -ENOMEM;
6562         }
6563 #endif
6564         return 0;
6565 }
6566
6567 static void __net_exit ip6_route_net_exit_late(struct net *net)
6568 {
6569 #ifdef CONFIG_PROC_FS
6570         remove_proc_entry("ipv6_route", net->proc_net);
6571         remove_proc_entry("rt6_stats", net->proc_net);
6572 #endif
6573 }
6574
6575 static struct pernet_operations ip6_route_net_ops = {
6576         .init = ip6_route_net_init,
6577         .exit = ip6_route_net_exit,
6578 };
6579
6580 static int __net_init ipv6_inetpeer_init(struct net *net)
6581 {
6582         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6583
6584         if (!bp)
6585                 return -ENOMEM;
6586         inet_peer_base_init(bp);
6587         net->ipv6.peers = bp;
6588         return 0;
6589 }
6590
6591 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6592 {
6593         struct inet_peer_base *bp = net->ipv6.peers;
6594
6595         net->ipv6.peers = NULL;
6596         inetpeer_invalidate_tree(bp);
6597         kfree(bp);
6598 }
6599
6600 static struct pernet_operations ipv6_inetpeer_ops = {
6601         .init   =       ipv6_inetpeer_init,
6602         .exit   =       ipv6_inetpeer_exit,
6603 };
6604
6605 static struct pernet_operations ip6_route_net_late_ops = {
6606         .init = ip6_route_net_init_late,
6607         .exit = ip6_route_net_exit_late,
6608 };
6609
6610 static struct notifier_block ip6_route_dev_notifier = {
6611         .notifier_call = ip6_route_dev_notify,
6612         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6613 };
6614
6615 void __init ip6_route_init_special_entries(void)
6616 {
6617         /* Registering of the loopback is done before this portion of code,
6618          * the loopback reference in rt6_info will not be taken, do it
6619          * manually for init_net */
6620         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6621         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6622         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6623   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6624         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6625         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6626         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6627         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6628   #endif
6629 }
6630
6631 #if IS_BUILTIN(CONFIG_IPV6)
6632 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6633 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6634
6635 BTF_ID_LIST(btf_fib6_info_id)
6636 BTF_ID(struct, fib6_info)
6637
6638 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6639         .seq_ops                = &ipv6_route_seq_ops,
6640         .init_seq_private       = bpf_iter_init_seq_net,
6641         .fini_seq_private       = bpf_iter_fini_seq_net,
6642         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6643 };
6644
6645 static struct bpf_iter_reg ipv6_route_reg_info = {
6646         .target                 = "ipv6_route",
6647         .ctx_arg_info_size      = 1,
6648         .ctx_arg_info           = {
6649                 { offsetof(struct bpf_iter__ipv6_route, rt),
6650                   PTR_TO_BTF_ID_OR_NULL },
6651         },
6652         .seq_info               = &ipv6_route_seq_info,
6653 };
6654
6655 static int __init bpf_iter_register(void)
6656 {
6657         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6658         return bpf_iter_reg_target(&ipv6_route_reg_info);
6659 }
6660
6661 static void bpf_iter_unregister(void)
6662 {
6663         bpf_iter_unreg_target(&ipv6_route_reg_info);
6664 }
6665 #endif
6666 #endif
6667
6668 int __init ip6_route_init(void)
6669 {
6670         int ret;
6671         int cpu;
6672
6673         ret = -ENOMEM;
6674         ip6_dst_ops_template.kmem_cachep =
6675                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6676                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6677         if (!ip6_dst_ops_template.kmem_cachep)
6678                 goto out;
6679
6680         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6681         if (ret)
6682                 goto out_kmem_cache;
6683
6684         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6685         if (ret)
6686                 goto out_dst_entries;
6687
6688         ret = register_pernet_subsys(&ip6_route_net_ops);
6689         if (ret)
6690                 goto out_register_inetpeer;
6691
6692         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6693
6694         ret = fib6_init();
6695         if (ret)
6696                 goto out_register_subsys;
6697
6698         ret = xfrm6_init();
6699         if (ret)
6700                 goto out_fib6_init;
6701
6702         ret = fib6_rules_init();
6703         if (ret)
6704                 goto xfrm6_init;
6705
6706         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6707         if (ret)
6708                 goto fib6_rules_init;
6709
6710         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6711                                    inet6_rtm_newroute, NULL, 0);
6712         if (ret < 0)
6713                 goto out_register_late_subsys;
6714
6715         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6716                                    inet6_rtm_delroute, NULL, 0);
6717         if (ret < 0)
6718                 goto out_register_late_subsys;
6719
6720         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6721                                    inet6_rtm_getroute, NULL,
6722                                    RTNL_FLAG_DOIT_UNLOCKED);
6723         if (ret < 0)
6724                 goto out_register_late_subsys;
6725
6726         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6727         if (ret)
6728                 goto out_register_late_subsys;
6729
6730 #if IS_BUILTIN(CONFIG_IPV6)
6731 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6732         ret = bpf_iter_register();
6733         if (ret)
6734                 goto out_register_late_subsys;
6735 #endif
6736 #endif
6737
6738         for_each_possible_cpu(cpu) {
6739                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6740
6741                 INIT_LIST_HEAD(&ul->head);
6742                 INIT_LIST_HEAD(&ul->quarantine);
6743                 spin_lock_init(&ul->lock);
6744         }
6745
6746 out:
6747         return ret;
6748
6749 out_register_late_subsys:
6750         rtnl_unregister_all(PF_INET6);
6751         unregister_pernet_subsys(&ip6_route_net_late_ops);
6752 fib6_rules_init:
6753         fib6_rules_cleanup();
6754 xfrm6_init:
6755         xfrm6_fini();
6756 out_fib6_init:
6757         fib6_gc_cleanup();
6758 out_register_subsys:
6759         unregister_pernet_subsys(&ip6_route_net_ops);
6760 out_register_inetpeer:
6761         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6762 out_dst_entries:
6763         dst_entries_destroy(&ip6_dst_blackhole_ops);
6764 out_kmem_cache:
6765         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6766         goto out;
6767 }
6768
6769 void ip6_route_cleanup(void)
6770 {
6771 #if IS_BUILTIN(CONFIG_IPV6)
6772 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6773         bpf_iter_unregister();
6774 #endif
6775 #endif
6776         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6777         unregister_pernet_subsys(&ip6_route_net_late_ops);
6778         fib6_rules_cleanup();
6779         xfrm6_fini();
6780         fib6_gc_cleanup();
6781         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6782         unregister_pernet_subsys(&ip6_route_net_ops);
6783         dst_entries_destroy(&ip6_dst_blackhole_ops);
6784         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6785 }