net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <linux/siphash.h>
  45 #include <net/net_namespace.h>
  46 #include <net/snmp.h>
  47 #include <net/ipv6.h>
  48 #include <net/ip6_fib.h>
  49 #include <net/ip6_route.h>
  50 #include <net/ndisc.h>
  51 #include <net/addrconf.h>
  52 #include <net/tcp.h>
  53 #include <linux/rtnetlink.h>
  54 #include <net/dst.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/xfrm.h>
  57 #include <net/netevent.h>
  58 #include <net/netlink.h>
  59 #include <net/rtnh.h>
  60 #include <net/lwtunnel.h>
  61 #include <net/ip_tunnels.h>
  62 #include <net/l3mdev.h>
  63 #include <net/ip.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/btf_ids.h>
  66
  67 #ifdef CONFIG_SYSCTL
  68 #include <linux/sysctl.h>
  69 #endif
  70
  71 static int ip6_rt_type_to_error(u8 fib6_type);
  72
  73 #define CREATE_TRACE_POINTS
  74 #include <trace/events/fib6.h>
  75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  76 #undef CREATE_TRACE_POINTS
  77
  78 enum rt6_nud_state {
  79         RT6_NUD_FAIL_HARD = -3,
  80         RT6_NUD_FAIL_PROBE = -2,
  81         RT6_NUD_FAIL_DO_RR = -1,
  82         RT6_NUD_SUCCEED = 1
  83 };
  84
  85 INDIRECT_CALLABLE_SCOPE
  86 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  87 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  88 INDIRECT_CALLABLE_SCOPE
  89 unsigned int            ip6_mtu(const struct dst_entry *dst);
  90 static void             ip6_negative_advice(struct sock *sk,
  91                                             struct dst_entry *dst);
  92 static void             ip6_dst_destroy(struct dst_entry *);
  93 static void             ip6_dst_ifdown(struct dst_entry *,
  94                                        struct net_device *dev);
  95 static void              ip6_dst_gc(struct dst_ops *ops);
  96
  97 static int              ip6_pkt_discard(struct sk_buff *skb);
  98 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  99 static int              ip6_pkt_prohibit(struct sk_buff *skb);
 100 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
 101 static void             ip6_link_failure(struct sk_buff *skb);
 102 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 103                                            struct sk_buff *skb, u32 mtu,
 104                                            bool confirm_neigh);
 105 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 106                                         struct sk_buff *skb);
 107 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 108                            int strict);
 109 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 110 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 111                          struct fib6_info *rt, struct dst_entry *dst,
 112                          struct in6_addr *dest, struct in6_addr *src,
 113                          int iif, int type, u32 portid, u32 seq,
 114                          unsigned int flags);
 115 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 116                                            const struct in6_addr *daddr,
 117                                            const struct in6_addr *saddr);
 118
 119 #ifdef CONFIG_IPV6_ROUTE_INFO
 120 static struct fib6_info *rt6_add_route_info(struct net *net,
 121                                            const struct in6_addr *prefix, int prefixlen,
 122                                            const struct in6_addr *gwaddr,
 123                                            struct net_device *dev,
 124                                            unsigned int pref);
 125 static struct fib6_info *rt6_get_route_info(struct net *net,
 126                                            const struct in6_addr *prefix, int prefixlen,
 127                                            const struct in6_addr *gwaddr,
 128                                            struct net_device *dev);
 129 #endif
 130
 131 struct uncached_list {
 132         spinlock_t              lock;
 133         struct list_head        head;
 134 };
 135
 136 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 137
 138 void rt6_uncached_list_add(struct rt6_info *rt)
 139 {
 140         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 141
 142         rt->dst.rt_uncached_list = ul;
 143
 144         spin_lock_bh(&ul->lock);
 145         list_add_tail(&rt->dst.rt_uncached, &ul->head);
 146         spin_unlock_bh(&ul->lock);
 147 }
 148
 149 void rt6_uncached_list_del(struct rt6_info *rt)
 150 {
 151         if (!list_empty(&rt->dst.rt_uncached)) {
 152                 struct uncached_list *ul = rt->dst.rt_uncached_list;
 153
 154                 spin_lock_bh(&ul->lock);
 155                 list_del_init(&rt->dst.rt_uncached);
 156                 spin_unlock_bh(&ul->lock);
 157         }
 158 }
 159
 160 static void rt6_uncached_list_flush_dev(struct net_device *dev)
 161 {
 162         int cpu;
 163
 164         for_each_possible_cpu(cpu) {
 165                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 166                 struct rt6_info *rt, *safe;
 167
 168                 if (list_empty(&ul->head))
 169                         continue;
 170
 171                 spin_lock_bh(&ul->lock);
 172                 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
 173                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 174                         struct net_device *rt_dev = rt->dst.dev;
 175                         bool handled = false;
 176
 177                         if (rt_idev && 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_del_init(&rt->dst.rt_uncached);
 192                 }
 193                 spin_unlock_bh(&ul->lock);
 194         }
 195 }
 196
 197 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 198                                              struct sk_buff *skb,
 199                                              const void *daddr)
 200 {
 201         if (!ipv6_addr_any(p))
 202                 return (const void *) p;
 203         else if (skb)
 204                 return &ipv6_hdr(skb)->daddr;
 205         return daddr;
 206 }
 207
 208 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 209                                    struct net_device *dev,
 210                                    struct sk_buff *skb,
 211                                    const void *daddr)
 212 {
 213         struct neighbour *n;
 214
 215         daddr = choose_neigh_daddr(gw, skb, daddr);
 216         n = __ipv6_neigh_lookup(dev, daddr);
 217         if (n)
 218                 return n;
 219
 220         n = neigh_create(&nd_tbl, daddr, dev);
 221         return IS_ERR(n) ? NULL : n;
 222 }
 223
 224 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 225                                               struct sk_buff *skb,
 226                                               const void *daddr)
 227 {
 228         const struct rt6_info *rt = dst_rt6_info(dst);
 229
 230         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 231                                 dst->dev, skb, daddr);
 232 }
 233
 234 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 235 {
 236         const struct rt6_info *rt = dst_rt6_info(dst);
 237         struct net_device *dev = dst->dev;
 238
 239         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 240         if (!daddr)
 241                 return;
 242         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 243                 return;
 244         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 245                 return;
 246         __ipv6_confirm_neigh(dev, daddr);
 247 }
 248
 249 static struct dst_ops ip6_dst_ops_template = {
 250         .family                 =       AF_INET6,
 251         .gc                     =       ip6_dst_gc,
 252         .gc_thresh              =       1024,
 253         .check                  =       ip6_dst_check,
 254         .default_advmss         =       ip6_default_advmss,
 255         .mtu                    =       ip6_mtu,
 256         .cow_metrics            =       dst_cow_metrics_generic,
 257         .destroy                =       ip6_dst_destroy,
 258         .ifdown                 =       ip6_dst_ifdown,
 259         .negative_advice        =       ip6_negative_advice,
 260         .link_failure           =       ip6_link_failure,
 261         .update_pmtu            =       ip6_rt_update_pmtu,
 262         .redirect               =       rt6_do_redirect,
 263         .local_out              =       __ip6_local_out,
 264         .neigh_lookup           =       ip6_dst_neigh_lookup,
 265         .confirm_neigh          =       ip6_confirm_neigh,
 266 };
 267
 268 static struct dst_ops ip6_dst_blackhole_ops = {
 269         .family                 = AF_INET6,
 270         .default_advmss         = ip6_default_advmss,
 271         .neigh_lookup           = ip6_dst_neigh_lookup,
 272         .check                  = ip6_dst_check,
 273         .destroy                = ip6_dst_destroy,
 274         .cow_metrics            = dst_cow_metrics_generic,
 275         .update_pmtu            = dst_blackhole_update_pmtu,
 276         .redirect               = dst_blackhole_redirect,
 277         .mtu                    = dst_blackhole_mtu,
 278 };
 279
 280 static const u32 ip6_template_metrics[RTAX_MAX] = {
 281         [RTAX_HOPLIMIT - 1] = 0,
 282 };
 283
 284 static const struct fib6_info fib6_null_entry_template = {
 285         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 286         .fib6_protocol  = RTPROT_KERNEL,
 287         .fib6_metric    = ~(u32)0,
 288         .fib6_ref       = REFCOUNT_INIT(1),
 289         .fib6_type      = RTN_UNREACHABLE,
 290         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 291 };
 292
 293 static const struct rt6_info ip6_null_entry_template = {
 294         .dst = {
 295                 .__rcuref       = RCUREF_INIT(1),
 296                 .__use          = 1,
 297                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 298                 .error          = -ENETUNREACH,
 299                 .input          = ip6_pkt_discard,
 300                 .output         = ip6_pkt_discard_out,
 301         },
 302         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 303 };
 304
 305 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 306
 307 static const struct rt6_info ip6_prohibit_entry_template = {
 308         .dst = {
 309                 .__rcuref       = RCUREF_INIT(1),
 310                 .__use          = 1,
 311                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 312                 .error          = -EACCES,
 313                 .input          = ip6_pkt_prohibit,
 314                 .output         = ip6_pkt_prohibit_out,
 315         },
 316         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 317 };
 318
 319 static const struct rt6_info ip6_blk_hole_entry_template = {
 320         .dst = {
 321                 .__rcuref       = RCUREF_INIT(1),
 322                 .__use          = 1,
 323                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 324                 .error          = -EINVAL,
 325                 .input          = dst_discard,
 326                 .output         = dst_discard_out,
 327         },
 328         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 329 };
 330
 331 #endif
 332
 333 static void rt6_info_init(struct rt6_info *rt)
 334 {
 335         memset_after(rt, 0, dst);
 336 }
 337
 338 /* allocate dst with ip6_dst_ops */
 339 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 340                                int flags)
 341 {
 342         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 343                                         DST_OBSOLETE_FORCE_CHK, flags);
 344
 345         if (rt) {
 346                 rt6_info_init(rt);
 347                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 348         }
 349
 350         return rt;
 351 }
 352 EXPORT_SYMBOL(ip6_dst_alloc);
 353
 354 static void ip6_dst_destroy(struct dst_entry *dst)
 355 {
 356         struct rt6_info *rt = dst_rt6_info(dst);
 357         struct fib6_info *from;
 358         struct inet6_dev *idev;
 359
 360         ip_dst_metrics_put(dst);
 361         rt6_uncached_list_del(rt);
 362
 363         idev = rt->rt6i_idev;
 364         if (idev) {
 365                 rt->rt6i_idev = NULL;
 366                 in6_dev_put(idev);
 367         }
 368
 369         from = unrcu_pointer(xchg(&rt->from, NULL));
 370         fib6_info_release(from);
 371 }
 372
 373 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
 374 {
 375         struct rt6_info *rt = dst_rt6_info(dst);
 376         struct inet6_dev *idev = rt->rt6i_idev;
 377         struct fib6_info *from;
 378
 379         if (idev && idev->dev != blackhole_netdev) {
 380                 struct inet6_dev *blackhole_idev = in6_dev_get(blackhole_netdev);
 381
 382                 if (blackhole_idev) {
 383                         rt->rt6i_idev = blackhole_idev;
 384                         in6_dev_put(idev);
 385                 }
 386         }
 387         from = unrcu_pointer(xchg(&rt->from, NULL));
 388         fib6_info_release(from);
 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 static struct fib6_info *
 416 rt6_multipath_first_sibling_rcu(const struct fib6_info *rt)
 417 {
 418         struct fib6_info *iter;
 419         struct fib6_node *fn;
 420
 421         fn = rcu_dereference(rt->fib6_node);
 422         if (!fn)
 423                 goto out;
 424         iter = rcu_dereference(fn->leaf);
 425         if (!iter)
 426                 goto out;
 427
 428         while (iter) {
 429                 if (iter->fib6_metric == rt->fib6_metric &&
 430                     rt6_qualify_for_ecmp(iter))
 431                         return iter;
 432                 iter = rcu_dereference(iter->fib6_next);
 433         }
 434
 435 out:
 436         return NULL;
 437 }
 438
 439 void fib6_select_path(const struct net *net, struct fib6_result *res,
 440                       struct flowi6 *fl6, int oif, bool have_oif_match,
 441                       const struct sk_buff *skb, int strict)
 442 {
 443         struct fib6_info *first, *match = res->f6i;
 444         struct fib6_info *sibling;
 445         int hash;
 446
 447         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 448                 goto out;
 449
 450         if (match->nh && have_oif_match && res->nh)
 451                 return;
 452
 453         if (skb)
 454                 IP6CB(skb)->flags |= IP6SKB_MULTIPATH;
 455
 456         /* We might have already computed the hash for ICMPv6 errors. In such
 457          * case it will always be non-zero. Otherwise now is the time to do it.
 458          */
 459         if (!fl6->mp_hash &&
 460             (!match->nh || nexthop_is_multipath(match->nh)))
 461                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 462
 463         if (unlikely(match->nh)) {
 464                 nexthop_path_fib6_result(res, fl6->mp_hash);
 465                 return;
 466         }
 467
 468         first = rt6_multipath_first_sibling_rcu(match);
 469         if (!first)
 470                 goto out;
 471
 472         hash = fl6->mp_hash;
 473         if (hash <= atomic_read(&first->fib6_nh->fib_nh_upper_bound)) {
 474                 if (rt6_score_route(first->fib6_nh, first->fib6_flags, oif,
 475                                     strict) >= 0)
 476                         match = first;
 477                 goto out;
 478         }
 479
 480         list_for_each_entry_rcu(sibling, &first->fib6_siblings,
 481                                 fib6_siblings) {
 482                 const struct fib6_nh *nh = sibling->fib6_nh;
 483                 int nh_upper_bound;
 484
 485                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 486                 if (hash > nh_upper_bound)
 487                         continue;
 488                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 489                         break;
 490                 match = sibling;
 491                 break;
 492         }
 493
 494 out:
 495         res->f6i = match;
 496         res->nh = match->fib6_nh;
 497 }
 498
 499 /*
 500  *      Route lookup. rcu_read_lock() should be held.
 501  */
 502
 503 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 504                                const struct in6_addr *saddr, int oif, int flags)
 505 {
 506         const struct net_device *dev;
 507
 508         if (nh->fib_nh_flags & RTNH_F_DEAD)
 509                 return false;
 510
 511         dev = nh->fib_nh_dev;
 512         if (oif) {
 513                 if (dev->ifindex == oif)
 514                         return true;
 515         } else {
 516                 if (ipv6_chk_addr(net, saddr, dev,
 517                                   flags & RT6_LOOKUP_F_IFACE))
 518                         return true;
 519         }
 520
 521         return false;
 522 }
 523
 524 struct fib6_nh_dm_arg {
 525         struct net              *net;
 526         const struct in6_addr   *saddr;
 527         int                     oif;
 528         int                     flags;
 529         struct fib6_nh          *nh;
 530 };
 531
 532 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 533 {
 534         struct fib6_nh_dm_arg *arg = _arg;
 535
 536         arg->nh = nh;
 537         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 538                                   arg->flags);
 539 }
 540
 541 /* returns fib6_nh from nexthop or NULL */
 542 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 543                                         struct fib6_result *res,
 544                                         const struct in6_addr *saddr,
 545                                         int oif, int flags)
 546 {
 547         struct fib6_nh_dm_arg arg = {
 548                 .net   = net,
 549                 .saddr = saddr,
 550                 .oif   = oif,
 551                 .flags = flags,
 552         };
 553
 554         if (nexthop_is_blackhole(nh))
 555                 return NULL;
 556
 557         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 558                 return arg.nh;
 559
 560         return NULL;
 561 }
 562
 563 static void rt6_device_match(struct net *net, struct fib6_result *res,
 564                              const struct in6_addr *saddr, int oif, int flags)
 565 {
 566         struct fib6_info *f6i = res->f6i;
 567         struct fib6_info *spf6i;
 568         struct fib6_nh *nh;
 569
 570         if (!oif && ipv6_addr_any(saddr)) {
 571                 if (unlikely(f6i->nh)) {
 572                         nh = nexthop_fib6_nh(f6i->nh);
 573                         if (nexthop_is_blackhole(f6i->nh))
 574                                 goto out_blackhole;
 575                 } else {
 576                         nh = f6i->fib6_nh;
 577                 }
 578                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 579                         goto out;
 580         }
 581
 582         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 583                 bool matched = false;
 584
 585                 if (unlikely(spf6i->nh)) {
 586                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 587                                               oif, flags);
 588                         if (nh)
 589                                 matched = true;
 590                 } else {
 591                         nh = spf6i->fib6_nh;
 592                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 593                                 matched = true;
 594                 }
 595                 if (matched) {
 596                         res->f6i = spf6i;
 597                         goto out;
 598                 }
 599         }
 600
 601         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 602                 res->f6i = net->ipv6.fib6_null_entry;
 603                 nh = res->f6i->fib6_nh;
 604                 goto out;
 605         }
 606
 607         if (unlikely(f6i->nh)) {
 608                 nh = nexthop_fib6_nh(f6i->nh);
 609                 if (nexthop_is_blackhole(f6i->nh))
 610                         goto out_blackhole;
 611         } else {
 612                 nh = f6i->fib6_nh;
 613         }
 614
 615         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 616                 res->f6i = net->ipv6.fib6_null_entry;
 617                 nh = res->f6i->fib6_nh;
 618         }
 619 out:
 620         res->nh = nh;
 621         res->fib6_type = res->f6i->fib6_type;
 622         res->fib6_flags = res->f6i->fib6_flags;
 623         return;
 624
 625 out_blackhole:
 626         res->fib6_flags |= RTF_REJECT;
 627         res->fib6_type = RTN_BLACKHOLE;
 628         res->nh = nh;
 629 }
 630
 631 #ifdef CONFIG_IPV6_ROUTER_PREF
 632 struct __rt6_probe_work {
 633         struct work_struct work;
 634         struct in6_addr target;
 635         struct net_device *dev;
 636         netdevice_tracker dev_tracker;
 637 };
 638
 639 static void rt6_probe_deferred(struct work_struct *w)
 640 {
 641         struct in6_addr mcaddr;
 642         struct __rt6_probe_work *work =
 643                 container_of(w, struct __rt6_probe_work, work);
 644
 645         addrconf_addr_solict_mult(&work->target, &mcaddr);
 646         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 647         netdev_put(work->dev, &work->dev_tracker);
 648         kfree(work);
 649 }
 650
 651 static void rt6_probe(struct fib6_nh *fib6_nh)
 652 {
 653         struct __rt6_probe_work *work = NULL;
 654         const struct in6_addr *nh_gw;
 655         unsigned long last_probe;
 656         struct neighbour *neigh;
 657         struct net_device *dev;
 658         struct inet6_dev *idev;
 659
 660         /*
 661          * Okay, this does not seem to be appropriate
 662          * for now, however, we need to check if it
 663          * is really so; aka Router Reachability Probing.
 664          *
 665          * Router Reachability Probe MUST be rate-limited
 666          * to no more than one per minute.
 667          */
 668         if (!fib6_nh->fib_nh_gw_family)
 669                 return;
 670
 671         nh_gw = &fib6_nh->fib_nh_gw6;
 672         dev = fib6_nh->fib_nh_dev;
 673         rcu_read_lock();
 674         last_probe = READ_ONCE(fib6_nh->last_probe);
 675         idev = __in6_dev_get(dev);
 676         if (!idev)
 677                 goto out;
 678         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 679         if (neigh) {
 680                 if (READ_ONCE(neigh->nud_state) & NUD_VALID)
 681                         goto out;
 682
 683                 write_lock_bh(&neigh->lock);
 684                 if (!(neigh->nud_state & NUD_VALID) &&
 685                     time_after(jiffies,
 686                                neigh->updated +
 687                                READ_ONCE(idev->cnf.rtr_probe_interval))) {
 688                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 689                         if (work)
 690                                 __neigh_set_probe_once(neigh);
 691                 }
 692                 write_unlock_bh(&neigh->lock);
 693         } else if (time_after(jiffies, last_probe +
 694                                        READ_ONCE(idev->cnf.rtr_probe_interval))) {
 695                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 696         }
 697
 698         if (!work || cmpxchg(&fib6_nh->last_probe,
 699                              last_probe, jiffies) != last_probe) {
 700                 kfree(work);
 701         } else {
 702                 INIT_WORK(&work->work, rt6_probe_deferred);
 703                 work->target = *nh_gw;
 704                 netdev_hold(dev, &work->dev_tracker, GFP_ATOMIC);
 705                 work->dev = dev;
 706                 schedule_work(&work->work);
 707         }
 708
 709 out:
 710         rcu_read_unlock();
 711 }
 712 #else
 713 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 714 {
 715 }
 716 #endif
 717
 718 /*
 719  * Default Router Selection (RFC 2461 6.3.6)
 720  */
 721 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 722 {
 723         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 724         struct neighbour *neigh;
 725
 726         rcu_read_lock();
 727         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 728                                           &fib6_nh->fib_nh_gw6);
 729         if (neigh) {
 730                 u8 nud_state = READ_ONCE(neigh->nud_state);
 731
 732                 if (nud_state & NUD_VALID)
 733                         ret = RT6_NUD_SUCCEED;
 734 #ifdef CONFIG_IPV6_ROUTER_PREF
 735                 else if (!(nud_state & NUD_FAILED))
 736                         ret = RT6_NUD_SUCCEED;
 737                 else
 738                         ret = RT6_NUD_FAIL_PROBE;
 739 #endif
 740         } else {
 741                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 742                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 743         }
 744         rcu_read_unlock();
 745
 746         return ret;
 747 }
 748
 749 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 750                            int strict)
 751 {
 752         int m = 0;
 753
 754         if (!oif || nh->fib_nh_dev->ifindex == oif)
 755                 m = 2;
 756
 757         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 758                 return RT6_NUD_FAIL_HARD;
 759 #ifdef CONFIG_IPV6_ROUTER_PREF
 760         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 761 #endif
 762         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 763             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 764                 int n = rt6_check_neigh(nh);
 765                 if (n < 0)
 766                         return n;
 767         }
 768         return m;
 769 }
 770
 771 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 772                        int oif, int strict, int *mpri, bool *do_rr)
 773 {
 774         bool match_do_rr = false;
 775         bool rc = false;
 776         int m;
 777
 778         if (nh->fib_nh_flags & RTNH_F_DEAD)
 779                 goto out;
 780
 781         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 782             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 783             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 784                 goto out;
 785
 786         m = rt6_score_route(nh, fib6_flags, oif, strict);
 787         if (m == RT6_NUD_FAIL_DO_RR) {
 788                 match_do_rr = true;
 789                 m = 0; /* lowest valid score */
 790         } else if (m == RT6_NUD_FAIL_HARD) {
 791                 goto out;
 792         }
 793
 794         if (strict & RT6_LOOKUP_F_REACHABLE)
 795                 rt6_probe(nh);
 796
 797         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 798         if (m > *mpri) {
 799                 *do_rr = match_do_rr;
 800                 *mpri = m;
 801                 rc = true;
 802         }
 803 out:
 804         return rc;
 805 }
 806
 807 struct fib6_nh_frl_arg {
 808         u32             flags;
 809         int             oif;
 810         int             strict;
 811         int             *mpri;
 812         bool            *do_rr;
 813         struct fib6_nh  *nh;
 814 };
 815
 816 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 817 {
 818         struct fib6_nh_frl_arg *arg = _arg;
 819
 820         arg->nh = nh;
 821         return find_match(nh, arg->flags, arg->oif, arg->strict,
 822                           arg->mpri, arg->do_rr);
 823 }
 824
 825 static void __find_rr_leaf(struct fib6_info *f6i_start,
 826                            struct fib6_info *nomatch, u32 metric,
 827                            struct fib6_result *res, struct fib6_info **cont,
 828                            int oif, int strict, bool *do_rr, int *mpri)
 829 {
 830         struct fib6_info *f6i;
 831
 832         for (f6i = f6i_start;
 833              f6i && f6i != nomatch;
 834              f6i = rcu_dereference(f6i->fib6_next)) {
 835                 bool matched = false;
 836                 struct fib6_nh *nh;
 837
 838                 if (cont && f6i->fib6_metric != metric) {
 839                         *cont = f6i;
 840                         return;
 841                 }
 842
 843                 if (fib6_check_expired(f6i))
 844                         continue;
 845
 846                 if (unlikely(f6i->nh)) {
 847                         struct fib6_nh_frl_arg arg = {
 848                                 .flags  = f6i->fib6_flags,
 849                                 .oif    = oif,
 850                                 .strict = strict,
 851                                 .mpri   = mpri,
 852                                 .do_rr  = do_rr
 853                         };
 854
 855                         if (nexthop_is_blackhole(f6i->nh)) {
 856                                 res->fib6_flags = RTF_REJECT;
 857                                 res->fib6_type = RTN_BLACKHOLE;
 858                                 res->f6i = f6i;
 859                                 res->nh = nexthop_fib6_nh(f6i->nh);
 860                                 return;
 861                         }
 862                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 863                                                      &arg)) {
 864                                 matched = true;
 865                                 nh = arg.nh;
 866                         }
 867                 } else {
 868                         nh = f6i->fib6_nh;
 869                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 870                                        mpri, do_rr))
 871                                 matched = true;
 872                 }
 873                 if (matched) {
 874                         res->f6i = f6i;
 875                         res->nh = nh;
 876                         res->fib6_flags = f6i->fib6_flags;
 877                         res->fib6_type = f6i->fib6_type;
 878                 }
 879         }
 880 }
 881
 882 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 883                          struct fib6_info *rr_head, int oif, int strict,
 884                          bool *do_rr, struct fib6_result *res)
 885 {
 886         u32 metric = rr_head->fib6_metric;
 887         struct fib6_info *cont = NULL;
 888         int mpri = -1;
 889
 890         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 891                        oif, strict, do_rr, &mpri);
 892
 893         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 894                        oif, strict, do_rr, &mpri);
 895
 896         if (res->f6i || !cont)
 897                 return;
 898
 899         __find_rr_leaf(cont, NULL, metric, res, NULL,
 900                        oif, strict, do_rr, &mpri);
 901 }
 902
 903 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 904                        struct fib6_result *res, int strict)
 905 {
 906         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 907         struct fib6_info *rt0;
 908         bool do_rr = false;
 909         int key_plen;
 910
 911         /* make sure this function or its helpers sets f6i */
 912         res->f6i = NULL;
 913
 914         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 915                 goto out;
 916
 917         rt0 = rcu_dereference(fn->rr_ptr);
 918         if (!rt0)
 919                 rt0 = leaf;
 920
 921         /* Double check to make sure fn is not an intermediate node
 922          * and fn->leaf does not points to its child's leaf
 923          * (This might happen if all routes under fn are deleted from
 924          * the tree and fib6_repair_tree() is called on the node.)
 925          */
 926         key_plen = rt0->fib6_dst.plen;
 927 #ifdef CONFIG_IPV6_SUBTREES
 928         if (rt0->fib6_src.plen)
 929                 key_plen = rt0->fib6_src.plen;
 930 #endif
 931         if (fn->fn_bit != key_plen)
 932                 goto out;
 933
 934         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 935         if (do_rr) {
 936                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 937
 938                 /* no entries matched; do round-robin */
 939                 if (!next || next->fib6_metric != rt0->fib6_metric)
 940                         next = leaf;
 941
 942                 if (next != rt0) {
 943                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 944                         /* make sure next is not being deleted from the tree */
 945                         if (next->fib6_node)
 946                                 rcu_assign_pointer(fn->rr_ptr, next);
 947                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 948                 }
 949         }
 950
 951 out:
 952         if (!res->f6i) {
 953                 res->f6i = net->ipv6.fib6_null_entry;
 954                 res->nh = res->f6i->fib6_nh;
 955                 res->fib6_flags = res->f6i->fib6_flags;
 956                 res->fib6_type = res->f6i->fib6_type;
 957         }
 958 }
 959
 960 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 961 {
 962         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 963                res->nh->fib_nh_gw_family;
 964 }
 965
 966 #ifdef CONFIG_IPV6_ROUTE_INFO
 967 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 968                   const struct in6_addr *gwaddr)
 969 {
 970         struct net *net = dev_net(dev);
 971         struct route_info *rinfo = (struct route_info *) opt;
 972         struct in6_addr prefix_buf, *prefix;
 973         struct fib6_table *table;
 974         unsigned int pref;
 975         unsigned long lifetime;
 976         struct fib6_info *rt;
 977
 978         if (len < sizeof(struct route_info)) {
 979                 return -EINVAL;
 980         }
 981
 982         /* Sanity check for prefix_len and length */
 983         if (rinfo->length > 3) {
 984                 return -EINVAL;
 985         } else if (rinfo->prefix_len > 128) {
 986                 return -EINVAL;
 987         } else if (rinfo->prefix_len > 64) {
 988                 if (rinfo->length < 2) {
 989                         return -EINVAL;
 990                 }
 991         } else if (rinfo->prefix_len > 0) {
 992                 if (rinfo->length < 1) {
 993                         return -EINVAL;
 994                 }
 995         }
 996
 997         pref = rinfo->route_pref;
 998         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 999                 return -EINVAL;
1000
1001         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
1002
1003         if (rinfo->length == 3)
1004                 prefix = (struct in6_addr *)rinfo->prefix;
1005         else {
1006                 /* this function is safe */
1007                 ipv6_addr_prefix(&prefix_buf,
1008                                  (struct in6_addr *)rinfo->prefix,
1009                                  rinfo->prefix_len);
1010                 prefix = &prefix_buf;
1011         }
1012
1013         if (rinfo->prefix_len == 0)
1014                 rt = rt6_get_dflt_router(net, gwaddr, dev);
1015         else
1016                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
1017                                         gwaddr, dev);
1018
1019         if (rt && !lifetime) {
1020                 ip6_del_rt(net, rt, false);
1021                 rt = NULL;
1022         }
1023
1024         if (!rt && lifetime)
1025                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
1026                                         dev, pref);
1027         else if (rt)
1028                 rt->fib6_flags = RTF_ROUTEINFO |
1029                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
1030
1031         if (rt) {
1032                 table = rt->fib6_table;
1033                 spin_lock_bh(&table->tb6_lock);
1034
1035                 if (!addrconf_finite_timeout(lifetime)) {
1036                         fib6_clean_expires(rt);
1037                         fib6_remove_gc_list(rt);
1038                 } else {
1039                         fib6_set_expires(rt, jiffies + HZ * lifetime);
1040                         fib6_add_gc_list(rt);
1041                 }
1042
1043                 spin_unlock_bh(&table->tb6_lock);
1044
1045                 fib6_info_release(rt);
1046         }
1047         return 0;
1048 }
1049 #endif
1050
1051 /*
1052  *      Misc support functions
1053  */
1054
1055 /* called with rcu_lock held */
1056 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1057 {
1058         struct net_device *dev = res->nh->fib_nh_dev;
1059
1060         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1061                 /* for copies of local routes, dst->dev needs to be the
1062                  * device if it is a master device, the master device if
1063                  * device is enslaved, and the loopback as the default
1064                  */
1065                 if (netif_is_l3_slave(dev) &&
1066                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1067                         dev = l3mdev_master_dev_rcu(dev);
1068                 else if (!netif_is_l3_master(dev))
1069                         dev = dev_net(dev)->loopback_dev;
1070                 /* last case is netif_is_l3_master(dev) is true in which
1071                  * case we want dev returned to be dev
1072                  */
1073         }
1074
1075         return dev;
1076 }
1077
1078 static const int fib6_prop[RTN_MAX + 1] = {
1079         [RTN_UNSPEC]    = 0,
1080         [RTN_UNICAST]   = 0,
1081         [RTN_LOCAL]     = 0,
1082         [RTN_BROADCAST] = 0,
1083         [RTN_ANYCAST]   = 0,
1084         [RTN_MULTICAST] = 0,
1085         [RTN_BLACKHOLE] = -EINVAL,
1086         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1087         [RTN_PROHIBIT]  = -EACCES,
1088         [RTN_THROW]     = -EAGAIN,
1089         [RTN_NAT]       = -EINVAL,
1090         [RTN_XRESOLVE]  = -EINVAL,
1091 };
1092
1093 static int ip6_rt_type_to_error(u8 fib6_type)
1094 {
1095         return fib6_prop[fib6_type];
1096 }
1097
1098 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1099 {
1100         unsigned short flags = 0;
1101
1102         if (rt->dst_nocount)
1103                 flags |= DST_NOCOUNT;
1104         if (rt->dst_nopolicy)
1105                 flags |= DST_NOPOLICY;
1106
1107         return flags;
1108 }
1109
1110 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1111 {
1112         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1113
1114         switch (fib6_type) {
1115         case RTN_BLACKHOLE:
1116                 rt->dst.output = dst_discard_out;
1117                 rt->dst.input = dst_discard;
1118                 break;
1119         case RTN_PROHIBIT:
1120                 rt->dst.output = ip6_pkt_prohibit_out;
1121                 rt->dst.input = ip6_pkt_prohibit;
1122                 break;
1123         case RTN_THROW:
1124         case RTN_UNREACHABLE:
1125         default:
1126                 rt->dst.output = ip6_pkt_discard_out;
1127                 rt->dst.input = ip6_pkt_discard;
1128                 break;
1129         }
1130 }
1131
1132 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1133 {
1134         struct fib6_info *f6i = res->f6i;
1135
1136         if (res->fib6_flags & RTF_REJECT) {
1137                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1138                 return;
1139         }
1140
1141         rt->dst.error = 0;
1142         rt->dst.output = ip6_output;
1143
1144         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1145                 rt->dst.input = ip6_input;
1146         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1147                 rt->dst.input = ip6_mc_input;
1148         } else {
1149                 rt->dst.input = ip6_forward;
1150         }
1151
1152         if (res->nh->fib_nh_lws) {
1153                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1154                 lwtunnel_set_redirect(&rt->dst);
1155         }
1156
1157         rt->dst.lastuse = jiffies;
1158 }
1159
1160 /* Caller must already hold reference to @from */
1161 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1162 {
1163         rt->rt6i_flags &= ~RTF_EXPIRES;
1164         rcu_assign_pointer(rt->from, from);
1165         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1166 }
1167
1168 /* Caller must already hold reference to f6i in result */
1169 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1170 {
1171         const struct fib6_nh *nh = res->nh;
1172         const struct net_device *dev = nh->fib_nh_dev;
1173         struct fib6_info *f6i = res->f6i;
1174
1175         ip6_rt_init_dst(rt, res);
1176
1177         rt->rt6i_dst = f6i->fib6_dst;
1178         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1179         rt->rt6i_flags = res->fib6_flags;
1180         if (nh->fib_nh_gw_family) {
1181                 rt->rt6i_gateway = nh->fib_nh_gw6;
1182                 rt->rt6i_flags |= RTF_GATEWAY;
1183         }
1184         rt6_set_from(rt, f6i);
1185 #ifdef CONFIG_IPV6_SUBTREES
1186         rt->rt6i_src = f6i->fib6_src;
1187 #endif
1188 }
1189
1190 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1191                                         struct in6_addr *saddr)
1192 {
1193         struct fib6_node *pn, *sn;
1194         while (1) {
1195                 if (fn->fn_flags & RTN_TL_ROOT)
1196                         return NULL;
1197                 pn = rcu_dereference(fn->parent);
1198                 sn = FIB6_SUBTREE(pn);
1199                 if (sn && sn != fn)
1200                         fn = fib6_node_lookup(sn, NULL, saddr);
1201                 else
1202                         fn = pn;
1203                 if (fn->fn_flags & RTN_RTINFO)
1204                         return fn;
1205         }
1206 }
1207
1208 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1209 {
1210         struct rt6_info *rt = *prt;
1211
1212         if (dst_hold_safe(&rt->dst))
1213                 return true;
1214         if (net) {
1215                 rt = net->ipv6.ip6_null_entry;
1216                 dst_hold(&rt->dst);
1217         } else {
1218                 rt = NULL;
1219         }
1220         *prt = rt;
1221         return false;
1222 }
1223
1224 /* called with rcu_lock held */
1225 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1226 {
1227         struct net_device *dev = res->nh->fib_nh_dev;
1228         struct fib6_info *f6i = res->f6i;
1229         unsigned short flags;
1230         struct rt6_info *nrt;
1231
1232         if (!fib6_info_hold_safe(f6i))
1233                 goto fallback;
1234
1235         flags = fib6_info_dst_flags(f6i);
1236         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1237         if (!nrt) {
1238                 fib6_info_release(f6i);
1239                 goto fallback;
1240         }
1241
1242         ip6_rt_copy_init(nrt, res);
1243         return nrt;
1244
1245 fallback:
1246         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1247         dst_hold(&nrt->dst);
1248         return nrt;
1249 }
1250
1251 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1252                                              struct fib6_table *table,
1253                                              struct flowi6 *fl6,
1254                                              const struct sk_buff *skb,
1255                                              int flags)
1256 {
1257         struct fib6_result res = {};
1258         struct fib6_node *fn;
1259         struct rt6_info *rt;
1260
1261         rcu_read_lock();
1262         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1263 restart:
1264         res.f6i = rcu_dereference(fn->leaf);
1265         if (!res.f6i)
1266                 res.f6i = net->ipv6.fib6_null_entry;
1267         else
1268                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1269                                  flags);
1270
1271         if (res.f6i == net->ipv6.fib6_null_entry) {
1272                 fn = fib6_backtrack(fn, &fl6->saddr);
1273                 if (fn)
1274                         goto restart;
1275
1276                 rt = net->ipv6.ip6_null_entry;
1277                 dst_hold(&rt->dst);
1278                 goto out;
1279         } else if (res.fib6_flags & RTF_REJECT) {
1280                 goto do_create;
1281         }
1282
1283         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1284                          fl6->flowi6_oif != 0, skb, flags);
1285
1286         /* Search through exception table */
1287         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1288         if (rt) {
1289                 if (ip6_hold_safe(net, &rt))
1290                         dst_use_noref(&rt->dst, jiffies);
1291         } else {
1292 do_create:
1293                 rt = ip6_create_rt_rcu(&res);
1294         }
1295
1296 out:
1297         trace_fib6_table_lookup(net, &res, table, fl6);
1298
1299         rcu_read_unlock();
1300
1301         return rt;
1302 }
1303
1304 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1305                                    const struct sk_buff *skb, int flags)
1306 {
1307         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1308 }
1309 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1310
1311 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1312                             const struct in6_addr *saddr, int oif,
1313                             const struct sk_buff *skb, int strict)
1314 {
1315         struct flowi6 fl6 = {
1316                 .flowi6_oif = oif,
1317                 .daddr = *daddr,
1318         };
1319         struct dst_entry *dst;
1320         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1321
1322         if (saddr) {
1323                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1324                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1325         }
1326
1327         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1328         if (dst->error == 0)
1329                 return dst_rt6_info(dst);
1330
1331         dst_release(dst);
1332
1333         return NULL;
1334 }
1335 EXPORT_SYMBOL(rt6_lookup);
1336
1337 /* ip6_ins_rt is called with FREE table->tb6_lock.
1338  * It takes new route entry, the addition fails by any reason the
1339  * route is released.
1340  * Caller must hold dst before calling it.
1341  */
1342
1343 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1344                         struct netlink_ext_ack *extack)
1345 {
1346         int err;
1347         struct fib6_table *table;
1348
1349         table = rt->fib6_table;
1350         spin_lock_bh(&table->tb6_lock);
1351         err = fib6_add(&table->tb6_root, rt, info, extack);
1352         spin_unlock_bh(&table->tb6_lock);
1353
1354         return err;
1355 }
1356
1357 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1358 {
1359         struct nl_info info = { .nl_net = net, };
1360
1361         return __ip6_ins_rt(rt, &info, NULL);
1362 }
1363
1364 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1365                                            const struct in6_addr *daddr,
1366                                            const struct in6_addr *saddr)
1367 {
1368         struct fib6_info *f6i = res->f6i;
1369         struct net_device *dev;
1370         struct rt6_info *rt;
1371
1372         /*
1373          *      Clone the route.
1374          */
1375
1376         if (!fib6_info_hold_safe(f6i))
1377                 return NULL;
1378
1379         dev = ip6_rt_get_dev_rcu(res);
1380         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1381         if (!rt) {
1382                 fib6_info_release(f6i);
1383                 return NULL;
1384         }
1385
1386         ip6_rt_copy_init(rt, res);
1387         rt->rt6i_flags |= RTF_CACHE;
1388         rt->rt6i_dst.addr = *daddr;
1389         rt->rt6i_dst.plen = 128;
1390
1391         if (!rt6_is_gw_or_nonexthop(res)) {
1392                 if (f6i->fib6_dst.plen != 128 &&
1393                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1394                         rt->rt6i_flags |= RTF_ANYCAST;
1395 #ifdef CONFIG_IPV6_SUBTREES
1396                 if (rt->rt6i_src.plen && saddr) {
1397                         rt->rt6i_src.addr = *saddr;
1398                         rt->rt6i_src.plen = 128;
1399                 }
1400 #endif
1401         }
1402
1403         return rt;
1404 }
1405
1406 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1407 {
1408         struct fib6_info *f6i = res->f6i;
1409         unsigned short flags = fib6_info_dst_flags(f6i);
1410         struct net_device *dev;
1411         struct rt6_info *pcpu_rt;
1412
1413         if (!fib6_info_hold_safe(f6i))
1414                 return NULL;
1415
1416         rcu_read_lock();
1417         dev = ip6_rt_get_dev_rcu(res);
1418         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1419         rcu_read_unlock();
1420         if (!pcpu_rt) {
1421                 fib6_info_release(f6i);
1422                 return NULL;
1423         }
1424         ip6_rt_copy_init(pcpu_rt, res);
1425         pcpu_rt->rt6i_flags |= RTF_PCPU;
1426
1427         if (f6i->nh)
1428                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1429
1430         return pcpu_rt;
1431 }
1432
1433 static bool rt6_is_valid(const struct rt6_info *rt6)
1434 {
1435         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1436 }
1437
1438 /* It should be called with rcu_read_lock() acquired */
1439 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1440 {
1441         struct rt6_info *pcpu_rt;
1442
1443         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1444
1445         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1446                 struct rt6_info *prev, **p;
1447
1448                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1449                 /* Paired with READ_ONCE() in __fib6_drop_pcpu_from() */
1450                 prev = xchg(p, NULL);
1451                 if (prev) {
1452                         dst_dev_put(&prev->dst);
1453                         dst_release(&prev->dst);
1454                 }
1455
1456                 pcpu_rt = NULL;
1457         }
1458
1459         return pcpu_rt;
1460 }
1461
1462 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1463                                             const struct fib6_result *res)
1464 {
1465         struct rt6_info *pcpu_rt, *prev, **p;
1466
1467         pcpu_rt = ip6_rt_pcpu_alloc(res);
1468         if (!pcpu_rt)
1469                 return NULL;
1470
1471         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1472         prev = cmpxchg(p, NULL, pcpu_rt);
1473         BUG_ON(prev);
1474
1475         if (res->f6i->fib6_destroying) {
1476                 struct fib6_info *from;
1477
1478                 from = unrcu_pointer(xchg(&pcpu_rt->from, NULL));
1479                 fib6_info_release(from);
1480         }
1481
1482         return pcpu_rt;
1483 }
1484
1485 /* exception hash table implementation
1486  */
1487 static DEFINE_SPINLOCK(rt6_exception_lock);
1488
1489 /* Remove rt6_ex from hash table and free the memory
1490  * Caller must hold rt6_exception_lock
1491  */
1492 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1493                                  struct rt6_exception *rt6_ex)
1494 {
1495         struct net *net;
1496
1497         if (!bucket || !rt6_ex)
1498                 return;
1499
1500         net = dev_net(rt6_ex->rt6i->dst.dev);
1501         net->ipv6.rt6_stats->fib_rt_cache--;
1502
1503         /* purge completely the exception to allow releasing the held resources:
1504          * some [sk] cache may keep the dst around for unlimited time
1505          */
1506         dst_dev_put(&rt6_ex->rt6i->dst);
1507
1508         hlist_del_rcu(&rt6_ex->hlist);
1509         dst_release(&rt6_ex->rt6i->dst);
1510         kfree_rcu(rt6_ex, rcu);
1511         WARN_ON_ONCE(!bucket->depth);
1512         bucket->depth--;
1513 }
1514
1515 /* Remove oldest rt6_ex in bucket and free the memory
1516  * Caller must hold rt6_exception_lock
1517  */
1518 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1519 {
1520         struct rt6_exception *rt6_ex, *oldest = NULL;
1521
1522         if (!bucket)
1523                 return;
1524
1525         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1526                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1527                         oldest = rt6_ex;
1528         }
1529         rt6_remove_exception(bucket, oldest);
1530 }
1531
1532 static u32 rt6_exception_hash(const struct in6_addr *dst,
1533                               const struct in6_addr *src)
1534 {
1535         static siphash_aligned_key_t rt6_exception_key;
1536         struct {
1537                 struct in6_addr dst;
1538                 struct in6_addr src;
1539         } __aligned(SIPHASH_ALIGNMENT) combined = {
1540                 .dst = *dst,
1541         };
1542         u64 val;
1543
1544         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1545
1546 #ifdef CONFIG_IPV6_SUBTREES
1547         if (src)
1548                 combined.src = *src;
1549 #endif
1550         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1551
1552         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1553 }
1554
1555 /* Helper function to find the cached rt in the hash table
1556  * and update bucket pointer to point to the bucket for this
1557  * (daddr, saddr) pair
1558  * Caller must hold rt6_exception_lock
1559  */
1560 static struct rt6_exception *
1561 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1562                               const struct in6_addr *daddr,
1563                               const struct in6_addr *saddr)
1564 {
1565         struct rt6_exception *rt6_ex;
1566         u32 hval;
1567
1568         if (!(*bucket) || !daddr)
1569                 return NULL;
1570
1571         hval = rt6_exception_hash(daddr, saddr);
1572         *bucket += hval;
1573
1574         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1575                 struct rt6_info *rt6 = rt6_ex->rt6i;
1576                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1577
1578 #ifdef CONFIG_IPV6_SUBTREES
1579                 if (matched && saddr)
1580                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1581 #endif
1582                 if (matched)
1583                         return rt6_ex;
1584         }
1585         return NULL;
1586 }
1587
1588 /* Helper function to find the cached rt in the hash table
1589  * and update bucket pointer to point to the bucket for this
1590  * (daddr, saddr) pair
1591  * Caller must hold rcu_read_lock()
1592  */
1593 static struct rt6_exception *
1594 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1595                          const struct in6_addr *daddr,
1596                          const struct in6_addr *saddr)
1597 {
1598         struct rt6_exception *rt6_ex;
1599         u32 hval;
1600
1601         WARN_ON_ONCE(!rcu_read_lock_held());
1602
1603         if (!(*bucket) || !daddr)
1604                 return NULL;
1605
1606         hval = rt6_exception_hash(daddr, saddr);
1607         *bucket += hval;
1608
1609         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1610                 struct rt6_info *rt6 = rt6_ex->rt6i;
1611                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1612
1613 #ifdef CONFIG_IPV6_SUBTREES
1614                 if (matched && saddr)
1615                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1616 #endif
1617                 if (matched)
1618                         return rt6_ex;
1619         }
1620         return NULL;
1621 }
1622
1623 static unsigned int fib6_mtu(const struct fib6_result *res)
1624 {
1625         const struct fib6_nh *nh = res->nh;
1626         unsigned int mtu;
1627
1628         if (res->f6i->fib6_pmtu) {
1629                 mtu = res->f6i->fib6_pmtu;
1630         } else {
1631                 struct net_device *dev = nh->fib_nh_dev;
1632                 struct inet6_dev *idev;
1633
1634                 rcu_read_lock();
1635                 idev = __in6_dev_get(dev);
1636                 mtu = READ_ONCE(idev->cnf.mtu6);
1637                 rcu_read_unlock();
1638         }
1639
1640         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1641
1642         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1643 }
1644
1645 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1646
1647 /* used when the flushed bit is not relevant, only access to the bucket
1648  * (ie., all bucket users except rt6_insert_exception);
1649  *
1650  * called under rcu lock; sometimes called with rt6_exception_lock held
1651  */
1652 static
1653 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1654                                                        spinlock_t *lock)
1655 {
1656         struct rt6_exception_bucket *bucket;
1657
1658         if (lock)
1659                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1660                                                    lockdep_is_held(lock));
1661         else
1662                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1663
1664         /* remove bucket flushed bit if set */
1665         if (bucket) {
1666                 unsigned long p = (unsigned long)bucket;
1667
1668                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1669                 bucket = (struct rt6_exception_bucket *)p;
1670         }
1671
1672         return bucket;
1673 }
1674
1675 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1676 {
1677         unsigned long p = (unsigned long)bucket;
1678
1679         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1680 }
1681
1682 /* called with rt6_exception_lock held */
1683 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1684                                               spinlock_t *lock)
1685 {
1686         struct rt6_exception_bucket *bucket;
1687         unsigned long p;
1688
1689         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1690                                            lockdep_is_held(lock));
1691
1692         p = (unsigned long)bucket;
1693         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1694         bucket = (struct rt6_exception_bucket *)p;
1695         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1696 }
1697
1698 static int rt6_insert_exception(struct rt6_info *nrt,
1699                                 const struct fib6_result *res)
1700 {
1701         struct net *net = dev_net(nrt->dst.dev);
1702         struct rt6_exception_bucket *bucket;
1703         struct fib6_info *f6i = res->f6i;
1704         struct in6_addr *src_key = NULL;
1705         struct rt6_exception *rt6_ex;
1706         struct fib6_nh *nh = res->nh;
1707         int max_depth;
1708         int err = 0;
1709
1710         spin_lock_bh(&rt6_exception_lock);
1711
1712         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1713                                           lockdep_is_held(&rt6_exception_lock));
1714         if (!bucket) {
1715                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1716                                  GFP_ATOMIC);
1717                 if (!bucket) {
1718                         err = -ENOMEM;
1719                         goto out;
1720                 }
1721                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1722         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1723                 err = -EINVAL;
1724                 goto out;
1725         }
1726
1727 #ifdef CONFIG_IPV6_SUBTREES
1728         /* fib6_src.plen != 0 indicates f6i is in subtree
1729          * and exception table is indexed by a hash of
1730          * both fib6_dst and fib6_src.
1731          * Otherwise, the exception table is indexed by
1732          * a hash of only fib6_dst.
1733          */
1734         if (f6i->fib6_src.plen)
1735                 src_key = &nrt->rt6i_src.addr;
1736 #endif
1737         /* rt6_mtu_change() might lower mtu on f6i.
1738          * Only insert this exception route if its mtu
1739          * is less than f6i's mtu value.
1740          */
1741         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1742                 err = -EINVAL;
1743                 goto out;
1744         }
1745
1746         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1747                                                src_key);
1748         if (rt6_ex)
1749                 rt6_remove_exception(bucket, rt6_ex);
1750
1751         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1752         if (!rt6_ex) {
1753                 err = -ENOMEM;
1754                 goto out;
1755         }
1756         rt6_ex->rt6i = nrt;
1757         rt6_ex->stamp = jiffies;
1758         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1759         bucket->depth++;
1760         net->ipv6.rt6_stats->fib_rt_cache++;
1761
1762         /* Randomize max depth to avoid some side channels attacks. */
1763         max_depth = FIB6_MAX_DEPTH + get_random_u32_below(FIB6_MAX_DEPTH);
1764         while (bucket->depth > max_depth)
1765                 rt6_exception_remove_oldest(bucket);
1766
1767 out:
1768         spin_unlock_bh(&rt6_exception_lock);
1769
1770         /* Update fn->fn_sernum to invalidate all cached dst */
1771         if (!err) {
1772                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1773                 fib6_update_sernum(net, f6i);
1774                 fib6_add_gc_list(f6i);
1775                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1776                 fib6_force_start_gc(net);
1777         }
1778
1779         return err;
1780 }
1781
1782 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1783 {
1784         struct rt6_exception_bucket *bucket;
1785         struct rt6_exception *rt6_ex;
1786         struct hlist_node *tmp;
1787         int i;
1788
1789         spin_lock_bh(&rt6_exception_lock);
1790
1791         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1792         if (!bucket)
1793                 goto out;
1794
1795         /* Prevent rt6_insert_exception() to recreate the bucket list */
1796         if (!from)
1797                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1798
1799         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1800                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1801                         if (!from ||
1802                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1803                                 rt6_remove_exception(bucket, rt6_ex);
1804                 }
1805                 WARN_ON_ONCE(!from && bucket->depth);
1806                 bucket++;
1807         }
1808 out:
1809         spin_unlock_bh(&rt6_exception_lock);
1810 }
1811
1812 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1813 {
1814         struct fib6_info *f6i = arg;
1815
1816         fib6_nh_flush_exceptions(nh, f6i);
1817
1818         return 0;
1819 }
1820
1821 void rt6_flush_exceptions(struct fib6_info *f6i)
1822 {
1823         if (f6i->nh) {
1824                 rcu_read_lock();
1825                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions, f6i);
1826                 rcu_read_unlock();
1827         } else {
1828                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1829         }
1830 }
1831
1832 /* Find cached rt in the hash table inside passed in rt
1833  * Caller has to hold rcu_read_lock()
1834  */
1835 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1836                                            const struct in6_addr *daddr,
1837                                            const struct in6_addr *saddr)
1838 {
1839         const struct in6_addr *src_key = NULL;
1840         struct rt6_exception_bucket *bucket;
1841         struct rt6_exception *rt6_ex;
1842         struct rt6_info *ret = NULL;
1843
1844 #ifdef CONFIG_IPV6_SUBTREES
1845         /* fib6i_src.plen != 0 indicates f6i is in subtree
1846          * and exception table is indexed by a hash of
1847          * both fib6_dst and fib6_src.
1848          * However, the src addr used to create the hash
1849          * might not be exactly the passed in saddr which
1850          * is a /128 addr from the flow.
1851          * So we need to use f6i->fib6_src to redo lookup
1852          * if the passed in saddr does not find anything.
1853          * (See the logic in ip6_rt_cache_alloc() on how
1854          * rt->rt6i_src is updated.)
1855          */
1856         if (res->f6i->fib6_src.plen)
1857                 src_key = saddr;
1858 find_ex:
1859 #endif
1860         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1861         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1862
1863         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1864                 ret = rt6_ex->rt6i;
1865
1866 #ifdef CONFIG_IPV6_SUBTREES
1867         /* Use fib6_src as src_key and redo lookup */
1868         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1869                 src_key = &res->f6i->fib6_src.addr;
1870                 goto find_ex;
1871         }
1872 #endif
1873
1874         return ret;
1875 }
1876
1877 /* Remove the passed in cached rt from the hash table that contains it */
1878 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1879                                     const struct rt6_info *rt)
1880 {
1881         const struct in6_addr *src_key = NULL;
1882         struct rt6_exception_bucket *bucket;
1883         struct rt6_exception *rt6_ex;
1884         int err;
1885
1886         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1887                 return -ENOENT;
1888
1889         spin_lock_bh(&rt6_exception_lock);
1890         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1891
1892 #ifdef CONFIG_IPV6_SUBTREES
1893         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1894          * and exception table is indexed by a hash of
1895          * both rt6i_dst and rt6i_src.
1896          * Otherwise, the exception table is indexed by
1897          * a hash of only rt6i_dst.
1898          */
1899         if (plen)
1900                 src_key = &rt->rt6i_src.addr;
1901 #endif
1902         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1903                                                &rt->rt6i_dst.addr,
1904                                                src_key);
1905         if (rt6_ex) {
1906                 rt6_remove_exception(bucket, rt6_ex);
1907                 err = 0;
1908         } else {
1909                 err = -ENOENT;
1910         }
1911
1912         spin_unlock_bh(&rt6_exception_lock);
1913         return err;
1914 }
1915
1916 struct fib6_nh_excptn_arg {
1917         struct rt6_info *rt;
1918         int             plen;
1919 };
1920
1921 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1922 {
1923         struct fib6_nh_excptn_arg *arg = _arg;
1924         int err;
1925
1926         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1927         if (err == 0)
1928                 return 1;
1929
1930         return 0;
1931 }
1932
1933 static int rt6_remove_exception_rt(struct rt6_info *rt)
1934 {
1935         struct fib6_info *from;
1936
1937         from = rcu_dereference(rt->from);
1938         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1939                 return -EINVAL;
1940
1941         if (from->nh) {
1942                 struct fib6_nh_excptn_arg arg = {
1943                         .rt = rt,
1944                         .plen = from->fib6_src.plen
1945                 };
1946                 int rc;
1947
1948                 /* rc = 1 means an entry was found */
1949                 rc = nexthop_for_each_fib6_nh(from->nh,
1950                                               rt6_nh_remove_exception_rt,
1951                                               &arg);
1952                 return rc ? 0 : -ENOENT;
1953         }
1954
1955         return fib6_nh_remove_exception(from->fib6_nh,
1956                                         from->fib6_src.plen, rt);
1957 }
1958
1959 /* Find rt6_ex which contains the passed in rt cache and
1960  * refresh its stamp
1961  */
1962 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1963                                      const struct rt6_info *rt)
1964 {
1965         const struct in6_addr *src_key = NULL;
1966         struct rt6_exception_bucket *bucket;
1967         struct rt6_exception *rt6_ex;
1968
1969         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1970 #ifdef CONFIG_IPV6_SUBTREES
1971         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1972          * and exception table is indexed by a hash of
1973          * both rt6i_dst and rt6i_src.
1974          * Otherwise, the exception table is indexed by
1975          * a hash of only rt6i_dst.
1976          */
1977         if (plen)
1978                 src_key = &rt->rt6i_src.addr;
1979 #endif
1980         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1981         if (rt6_ex)
1982                 rt6_ex->stamp = jiffies;
1983 }
1984
1985 struct fib6_nh_match_arg {
1986         const struct net_device *dev;
1987         const struct in6_addr   *gw;
1988         struct fib6_nh          *match;
1989 };
1990
1991 /* determine if fib6_nh has given device and gateway */
1992 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1993 {
1994         struct fib6_nh_match_arg *arg = _arg;
1995
1996         if (arg->dev != nh->fib_nh_dev ||
1997             (arg->gw && !nh->fib_nh_gw_family) ||
1998             (!arg->gw && nh->fib_nh_gw_family) ||
1999             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
2000                 return 0;
2001
2002         arg->match = nh;
2003
2004         /* found a match, break the loop */
2005         return 1;
2006 }
2007
2008 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
2009 {
2010         struct fib6_info *from;
2011         struct fib6_nh *fib6_nh;
2012
2013         rcu_read_lock();
2014
2015         from = rcu_dereference(rt->from);
2016         if (!from || !(rt->rt6i_flags & RTF_CACHE))
2017                 goto unlock;
2018
2019         if (from->nh) {
2020                 struct fib6_nh_match_arg arg = {
2021                         .dev = rt->dst.dev,
2022                         .gw = &rt->rt6i_gateway,
2023                 };
2024
2025                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
2026
2027                 if (!arg.match)
2028                         goto unlock;
2029                 fib6_nh = arg.match;
2030         } else {
2031                 fib6_nh = from->fib6_nh;
2032         }
2033         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
2034 unlock:
2035         rcu_read_unlock();
2036 }
2037
2038 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
2039                                          struct rt6_info *rt, int mtu)
2040 {
2041         /* If the new MTU is lower than the route PMTU, this new MTU will be the
2042          * lowest MTU in the path: always allow updating the route PMTU to
2043          * reflect PMTU decreases.
2044          *
2045          * If the new MTU is higher, and the route PMTU is equal to the local
2046          * MTU, this means the old MTU is the lowest in the path, so allow
2047          * updating it: if other nodes now have lower MTUs, PMTU discovery will
2048          * handle this.
2049          */
2050
2051         if (dst_mtu(&rt->dst) >= mtu)
2052                 return true;
2053
2054         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2055                 return true;
2056
2057         return false;
2058 }
2059
2060 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2061                                        const struct fib6_nh *nh, int mtu)
2062 {
2063         struct rt6_exception_bucket *bucket;
2064         struct rt6_exception *rt6_ex;
2065         int i;
2066
2067         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2068         if (!bucket)
2069                 return;
2070
2071         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2072                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2073                         struct rt6_info *entry = rt6_ex->rt6i;
2074
2075                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2076                          * route), the metrics of its rt->from have already
2077                          * been updated.
2078                          */
2079                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2080                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2081                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2082                 }
2083                 bucket++;
2084         }
2085 }
2086
2087 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2088
2089 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2090                                             const struct in6_addr *gateway)
2091 {
2092         struct rt6_exception_bucket *bucket;
2093         struct rt6_exception *rt6_ex;
2094         struct hlist_node *tmp;
2095         int i;
2096
2097         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2098                 return;
2099
2100         spin_lock_bh(&rt6_exception_lock);
2101         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2102         if (bucket) {
2103                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2104                         hlist_for_each_entry_safe(rt6_ex, tmp,
2105                                                   &bucket->chain, hlist) {
2106                                 struct rt6_info *entry = rt6_ex->rt6i;
2107
2108                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2109                                     RTF_CACHE_GATEWAY &&
2110                                     ipv6_addr_equal(gateway,
2111                                                     &entry->rt6i_gateway)) {
2112                                         rt6_remove_exception(bucket, rt6_ex);
2113                                 }
2114                         }
2115                         bucket++;
2116                 }
2117         }
2118
2119         spin_unlock_bh(&rt6_exception_lock);
2120 }
2121
2122 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2123                                       struct rt6_exception *rt6_ex,
2124                                       struct fib6_gc_args *gc_args,
2125                                       unsigned long now)
2126 {
2127         struct rt6_info *rt = rt6_ex->rt6i;
2128
2129         /* we are pruning and obsoleting aged-out and non gateway exceptions
2130          * even if others have still references to them, so that on next
2131          * dst_check() such references can be dropped.
2132          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2133          * expired, independently from their aging, as per RFC 8201 section 4
2134          */
2135         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2136                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2137                         pr_debug("aging clone %p\n", rt);
2138                         rt6_remove_exception(bucket, rt6_ex);
2139                         return;
2140                 }
2141         } else if (time_after(jiffies, rt->dst.expires)) {
2142                 pr_debug("purging expired route %p\n", rt);
2143                 rt6_remove_exception(bucket, rt6_ex);
2144                 return;
2145         }
2146
2147         if (rt->rt6i_flags & RTF_GATEWAY) {
2148                 struct neighbour *neigh;
2149
2150                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2151
2152                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2153                         pr_debug("purging route %p via non-router but gateway\n",
2154                                  rt);
2155                         rt6_remove_exception(bucket, rt6_ex);
2156                         return;
2157                 }
2158         }
2159
2160         gc_args->more++;
2161 }
2162
2163 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2164                                    struct fib6_gc_args *gc_args,
2165                                    unsigned long now)
2166 {
2167         struct rt6_exception_bucket *bucket;
2168         struct rt6_exception *rt6_ex;
2169         struct hlist_node *tmp;
2170         int i;
2171
2172         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2173                 return;
2174
2175         rcu_read_lock_bh();
2176         spin_lock(&rt6_exception_lock);
2177         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2178         if (bucket) {
2179                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2180                         hlist_for_each_entry_safe(rt6_ex, tmp,
2181                                                   &bucket->chain, hlist) {
2182                                 rt6_age_examine_exception(bucket, rt6_ex,
2183                                                           gc_args, now);
2184                         }
2185                         bucket++;
2186                 }
2187         }
2188         spin_unlock(&rt6_exception_lock);
2189         rcu_read_unlock_bh();
2190 }
2191
2192 struct fib6_nh_age_excptn_arg {
2193         struct fib6_gc_args     *gc_args;
2194         unsigned long           now;
2195 };
2196
2197 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2198 {
2199         struct fib6_nh_age_excptn_arg *arg = _arg;
2200
2201         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2202         return 0;
2203 }
2204
2205 void rt6_age_exceptions(struct fib6_info *f6i,
2206                         struct fib6_gc_args *gc_args,
2207                         unsigned long now)
2208 {
2209         if (f6i->nh) {
2210                 struct fib6_nh_age_excptn_arg arg = {
2211                         .gc_args = gc_args,
2212                         .now = now
2213                 };
2214
2215                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2216                                          &arg);
2217         } else {
2218                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2219         }
2220 }
2221
2222 /* must be called with rcu lock held */
2223 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2224                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2225 {
2226         struct fib6_node *fn, *saved_fn;
2227
2228         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2229         saved_fn = fn;
2230
2231 redo_rt6_select:
2232         rt6_select(net, fn, oif, res, strict);
2233         if (res->f6i == net->ipv6.fib6_null_entry) {
2234                 fn = fib6_backtrack(fn, &fl6->saddr);
2235                 if (fn)
2236                         goto redo_rt6_select;
2237                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2238                         /* also consider unreachable route */
2239                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2240                         fn = saved_fn;
2241                         goto redo_rt6_select;
2242                 }
2243         }
2244
2245         trace_fib6_table_lookup(net, res, table, fl6);
2246
2247         return 0;
2248 }
2249
2250 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2251                                int oif, struct flowi6 *fl6,
2252                                const struct sk_buff *skb, int flags)
2253 {
2254         struct fib6_result res = {};
2255         struct rt6_info *rt = NULL;
2256         int strict = 0;
2257
2258         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2259                      !rcu_read_lock_held());
2260
2261         strict |= flags & RT6_LOOKUP_F_IFACE;
2262         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2263         if (READ_ONCE(net->ipv6.devconf_all->forwarding) == 0)
2264                 strict |= RT6_LOOKUP_F_REACHABLE;
2265
2266         rcu_read_lock();
2267
2268         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2269         if (res.f6i == net->ipv6.fib6_null_entry)
2270                 goto out;
2271
2272         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2273
2274         /*Search through exception table */
2275         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2276         if (rt) {
2277                 goto out;
2278         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2279                             !res.nh->fib_nh_gw_family)) {
2280                 /* Create a RTF_CACHE clone which will not be
2281                  * owned by the fib6 tree.  It is for the special case where
2282                  * the daddr in the skb during the neighbor look-up is different
2283                  * from the fl6->daddr used to look-up route here.
2284                  */
2285                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2286
2287                 if (rt) {
2288                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2289                          * As rt6_uncached_list_add() does not consume refcnt,
2290                          * this refcnt is always returned to the caller even
2291                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2292                          */
2293                         rt6_uncached_list_add(rt);
2294                         rcu_read_unlock();
2295
2296                         return rt;
2297                 }
2298         } else {
2299                 /* Get a percpu copy */
2300                 local_bh_disable();
2301                 rt = rt6_get_pcpu_route(&res);
2302
2303                 if (!rt)
2304                         rt = rt6_make_pcpu_route(net, &res);
2305
2306                 local_bh_enable();
2307         }
2308 out:
2309         if (!rt)
2310                 rt = net->ipv6.ip6_null_entry;
2311         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2312                 ip6_hold_safe(net, &rt);
2313         rcu_read_unlock();
2314
2315         return rt;
2316 }
2317 EXPORT_SYMBOL_GPL(ip6_pol_route);
2318
2319 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2320                                             struct fib6_table *table,
2321                                             struct flowi6 *fl6,
2322                                             const struct sk_buff *skb,
2323                                             int flags)
2324 {
2325         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2326 }
2327
2328 struct dst_entry *ip6_route_input_lookup(struct net *net,
2329                                          struct net_device *dev,
2330                                          struct flowi6 *fl6,
2331                                          const struct sk_buff *skb,
2332                                          int flags)
2333 {
2334         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2335                 flags |= RT6_LOOKUP_F_IFACE;
2336
2337         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2338 }
2339 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2340
2341 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2342                                   struct flow_keys *keys,
2343                                   struct flow_keys *flkeys)
2344 {
2345         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2346         const struct ipv6hdr *key_iph = outer_iph;
2347         struct flow_keys *_flkeys = flkeys;
2348         const struct ipv6hdr *inner_iph;
2349         const struct icmp6hdr *icmph;
2350         struct ipv6hdr _inner_iph;
2351         struct icmp6hdr _icmph;
2352
2353         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2354                 goto out;
2355
2356         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2357                                    sizeof(_icmph), &_icmph);
2358         if (!icmph)
2359                 goto out;
2360
2361         if (!icmpv6_is_err(icmph->icmp6_type))
2362                 goto out;
2363
2364         inner_iph = skb_header_pointer(skb,
2365                                        skb_transport_offset(skb) + sizeof(*icmph),
2366                                        sizeof(_inner_iph), &_inner_iph);
2367         if (!inner_iph)
2368                 goto out;
2369
2370         key_iph = inner_iph;
2371         _flkeys = NULL;
2372 out:
2373         if (_flkeys) {
2374                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2375                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2376                 keys->tags.flow_label = _flkeys->tags.flow_label;
2377                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2378         } else {
2379                 keys->addrs.v6addrs.src = key_iph->saddr;
2380                 keys->addrs.v6addrs.dst = key_iph->daddr;
2381                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2382                 keys->basic.ip_proto = key_iph->nexthdr;
2383         }
2384 }
2385
2386 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2387                                            const struct sk_buff *skb,
2388                                            bool *p_has_inner)
2389 {
2390         u32 hash_fields = ip6_multipath_hash_fields(net);
2391         struct flow_keys keys, hash_keys;
2392
2393         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2394                 return 0;
2395
2396         memset(&hash_keys, 0, sizeof(hash_keys));
2397         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2398
2399         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2400         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2401                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2402         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2403                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2404         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2405                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2406         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2407                 hash_keys.tags.flow_label = keys.tags.flow_label;
2408         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2409                 hash_keys.ports.src = keys.ports.src;
2410         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2411                 hash_keys.ports.dst = keys.ports.dst;
2412
2413         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2414         return fib_multipath_hash_from_keys(net, &hash_keys);
2415 }
2416
2417 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2418                                            const struct sk_buff *skb,
2419                                            bool has_inner)
2420 {
2421         u32 hash_fields = ip6_multipath_hash_fields(net);
2422         struct flow_keys keys, hash_keys;
2423
2424         /* We assume the packet carries an encapsulation, but if none was
2425          * encountered during dissection of the outer flow, then there is no
2426          * point in calling the flow dissector again.
2427          */
2428         if (!has_inner)
2429                 return 0;
2430
2431         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2432                 return 0;
2433
2434         memset(&hash_keys, 0, sizeof(hash_keys));
2435         skb_flow_dissect_flow_keys(skb, &keys, 0);
2436
2437         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2438                 return 0;
2439
2440         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2441                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2442                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2443                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2444                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2445                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2446         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2447                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2448                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2449                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2450                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2451                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2452                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2453                         hash_keys.tags.flow_label = keys.tags.flow_label;
2454         }
2455
2456         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2457                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2458         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2459                 hash_keys.ports.src = keys.ports.src;
2460         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2461                 hash_keys.ports.dst = keys.ports.dst;
2462
2463         return fib_multipath_hash_from_keys(net, &hash_keys);
2464 }
2465
2466 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2467                                          const struct sk_buff *skb)
2468 {
2469         u32 mhash, mhash_inner;
2470         bool has_inner = true;
2471
2472         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2473         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2474
2475         return jhash_2words(mhash, mhash_inner, 0);
2476 }
2477
2478 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2479                                          const struct flowi6 *fl6)
2480 {
2481         u32 hash_fields = ip6_multipath_hash_fields(net);
2482         struct flow_keys hash_keys;
2483
2484         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2485                 return 0;
2486
2487         memset(&hash_keys, 0, sizeof(hash_keys));
2488         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2489         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2490                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2491         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2492                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2493         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2494                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2495         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2496                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2497         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) {
2498                 if (fl6->flowi6_flags & FLOWI_FLAG_ANY_SPORT)
2499                         hash_keys.ports.src = (__force __be16)get_random_u16();
2500                 else
2501                         hash_keys.ports.src = fl6->fl6_sport;
2502         }
2503         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2504                 hash_keys.ports.dst = fl6->fl6_dport;
2505
2506         return fib_multipath_hash_from_keys(net, &hash_keys);
2507 }
2508
2509 /* if skb is set it will be used and fl6 can be NULL */
2510 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2511                        const struct sk_buff *skb, struct flow_keys *flkeys)
2512 {
2513         struct flow_keys hash_keys;
2514         u32 mhash = 0;
2515
2516         switch (ip6_multipath_hash_policy(net)) {
2517         case 0:
2518                 memset(&hash_keys, 0, sizeof(hash_keys));
2519                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2520                 if (skb) {
2521                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2522                 } else {
2523                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2524                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2525                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2526                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2527                 }
2528                 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2529                 break;
2530         case 1:
2531                 if (skb) {
2532                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2533                         struct flow_keys keys;
2534
2535                         /* short-circuit if we already have L4 hash present */
2536                         if (skb->l4_hash)
2537                                 return skb_get_hash_raw(skb) >> 1;
2538
2539                         memset(&hash_keys, 0, sizeof(hash_keys));
2540
2541                         if (!flkeys) {
2542                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2543                                 flkeys = &keys;
2544                         }
2545                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2546                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2547                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2548                         hash_keys.ports.src = flkeys->ports.src;
2549                         hash_keys.ports.dst = flkeys->ports.dst;
2550                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2551                 } else {
2552                         memset(&hash_keys, 0, sizeof(hash_keys));
2553                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2554                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2555                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2556                         if (fl6->flowi6_flags & FLOWI_FLAG_ANY_SPORT)
2557                                 hash_keys.ports.src = (__force __be16)get_random_u16();
2558                         else
2559                                 hash_keys.ports.src = fl6->fl6_sport;
2560                         hash_keys.ports.dst = fl6->fl6_dport;
2561                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2562                 }
2563                 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2564                 break;
2565         case 2:
2566                 memset(&hash_keys, 0, sizeof(hash_keys));
2567                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2568                 if (skb) {
2569                         struct flow_keys keys;
2570
2571                         if (!flkeys) {
2572                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2573                                 flkeys = &keys;
2574                         }
2575
2576                         /* Inner can be v4 or v6 */
2577                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2578                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2579                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2580                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2581                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2582                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2583                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2584                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2585                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2586                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2587                         } else {
2588                                 /* Same as case 0 */
2589                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2590                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2591                         }
2592                 } else {
2593                         /* Same as case 0 */
2594                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2595                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2596                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2597                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2598                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2599                 }
2600                 mhash = fib_multipath_hash_from_keys(net, &hash_keys);
2601                 break;
2602         case 3:
2603                 if (skb)
2604                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2605                 else
2606                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2607                 break;
2608         }
2609
2610         return mhash >> 1;
2611 }
2612
2613 /* Called with rcu held */
2614 void ip6_route_input(struct sk_buff *skb)
2615 {
2616         const struct ipv6hdr *iph = ipv6_hdr(skb);
2617         struct net *net = dev_net(skb->dev);
2618         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2619         struct ip_tunnel_info *tun_info;
2620         struct flowi6 fl6 = {
2621                 .flowi6_iif = skb->dev->ifindex,
2622                 .daddr = iph->daddr,
2623                 .saddr = iph->saddr,
2624                 .flowlabel = ip6_flowinfo(iph),
2625                 .flowi6_mark = skb->mark,
2626                 .flowi6_proto = iph->nexthdr,
2627         };
2628         struct flow_keys *flkeys = NULL, _flkeys;
2629
2630         tun_info = skb_tunnel_info(skb);
2631         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2632                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2633
2634         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2635                 flkeys = &_flkeys;
2636
2637         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2638                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2639         skb_dst_drop(skb);
2640         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2641                                                       &fl6, skb, flags));
2642 }
2643
2644 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2645                                              struct fib6_table *table,
2646                                              struct flowi6 *fl6,
2647                                              const struct sk_buff *skb,
2648                                              int flags)
2649 {
2650         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2651 }
2652
2653 static struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2654                                                       const struct sock *sk,
2655                                                       struct flowi6 *fl6,
2656                                                       int flags)
2657 {
2658         bool any_src;
2659
2660         if (ipv6_addr_type(&fl6->daddr) &
2661             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2662                 struct dst_entry *dst;
2663
2664                 /* This function does not take refcnt on the dst */
2665                 dst = l3mdev_link_scope_lookup(net, fl6);
2666                 if (dst)
2667                         return dst;
2668         }
2669
2670         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2671
2672         flags |= RT6_LOOKUP_F_DST_NOREF;
2673         any_src = ipv6_addr_any(&fl6->saddr);
2674         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2675             (fl6->flowi6_oif && any_src))
2676                 flags |= RT6_LOOKUP_F_IFACE;
2677
2678         if (!any_src)
2679                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2680         else if (sk)
2681                 flags |= rt6_srcprefs2flags(READ_ONCE(inet6_sk(sk)->srcprefs));
2682
2683         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2684 }
2685
2686 struct dst_entry *ip6_route_output_flags(struct net *net,
2687                                          const struct sock *sk,
2688                                          struct flowi6 *fl6,
2689                                          int flags)
2690 {
2691         struct dst_entry *dst;
2692         struct rt6_info *rt6;
2693
2694         rcu_read_lock();
2695         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2696         rt6 = dst_rt6_info(dst);
2697         /* For dst cached in uncached_list, refcnt is already taken. */
2698         if (list_empty(&rt6->dst.rt_uncached) && !dst_hold_safe(dst)) {
2699                 dst = &net->ipv6.ip6_null_entry->dst;
2700                 dst_hold(dst);
2701         }
2702         rcu_read_unlock();
2703
2704         return dst;
2705 }
2706 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2707
2708 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2709 {
2710         struct rt6_info *rt, *ort = dst_rt6_info(dst_orig);
2711         struct net_device *loopback_dev = net->loopback_dev;
2712         struct dst_entry *new = NULL;
2713
2714         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev,
2715                        DST_OBSOLETE_DEAD, 0);
2716         if (rt) {
2717                 rt6_info_init(rt);
2718                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2719
2720                 new = &rt->dst;
2721                 new->__use = 1;
2722                 new->input = dst_discard;
2723                 new->output = dst_discard_out;
2724
2725                 dst_copy_metrics(new, &ort->dst);
2726
2727                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2728                 rt->rt6i_gateway = ort->rt6i_gateway;
2729                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2730
2731                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2732 #ifdef CONFIG_IPV6_SUBTREES
2733                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2734 #endif
2735         }
2736
2737         dst_release(dst_orig);
2738         return new ? new : ERR_PTR(-ENOMEM);
2739 }
2740
2741 /*
2742  *      Destination cache support functions
2743  */
2744
2745 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2746 {
2747         u32 rt_cookie = 0;
2748
2749         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2750                 return false;
2751
2752         if (fib6_check_expired(f6i))
2753                 return false;
2754
2755         return true;
2756 }
2757
2758 static struct dst_entry *rt6_check(struct rt6_info *rt,
2759                                    struct fib6_info *from,
2760                                    u32 cookie)
2761 {
2762         u32 rt_cookie = 0;
2763
2764         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2765             rt_cookie != cookie)
2766                 return NULL;
2767
2768         if (rt6_check_expired(rt))
2769                 return NULL;
2770
2771         return &rt->dst;
2772 }
2773
2774 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2775                                             struct fib6_info *from,
2776                                             u32 cookie)
2777 {
2778         if (!__rt6_check_expired(rt) &&
2779             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2780             fib6_check(from, cookie))
2781                 return &rt->dst;
2782         else
2783                 return NULL;
2784 }
2785
2786 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2787                                                         u32 cookie)
2788 {
2789         struct dst_entry *dst_ret;
2790         struct fib6_info *from;
2791         struct rt6_info *rt;
2792
2793         rt = dst_rt6_info(dst);
2794
2795         if (rt->sernum)
2796                 return rt6_is_valid(rt) ? dst : NULL;
2797
2798         rcu_read_lock();
2799
2800         /* All IPV6 dsts are created with ->obsolete set to the value
2801          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2802          * into this function always.
2803          */
2804
2805         from = rcu_dereference(rt->from);
2806
2807         if (from && (rt->rt6i_flags & RTF_PCPU ||
2808             unlikely(!list_empty(&rt->dst.rt_uncached))))
2809                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2810         else
2811                 dst_ret = rt6_check(rt, from, cookie);
2812
2813         rcu_read_unlock();
2814
2815         return dst_ret;
2816 }
2817 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2818
2819 static void ip6_negative_advice(struct sock *sk,
2820                                 struct dst_entry *dst)
2821 {
2822         struct rt6_info *rt = dst_rt6_info(dst);
2823
2824         if (rt->rt6i_flags & RTF_CACHE) {
2825                 rcu_read_lock();
2826                 if (rt6_check_expired(rt)) {
2827                         /* rt/dst can not be destroyed yet,
2828                          * because of rcu_read_lock()
2829                          */
2830                         sk_dst_reset(sk);
2831                         rt6_remove_exception_rt(rt);
2832                 }
2833                 rcu_read_unlock();
2834                 return;
2835         }
2836         sk_dst_reset(sk);
2837 }
2838
2839 static void ip6_link_failure(struct sk_buff *skb)
2840 {
2841         struct rt6_info *rt;
2842
2843         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2844
2845         rt = dst_rt6_info(skb_dst(skb));
2846         if (rt) {
2847                 rcu_read_lock();
2848                 if (rt->rt6i_flags & RTF_CACHE) {
2849                         rt6_remove_exception_rt(rt);
2850                 } else {
2851                         struct fib6_info *from;
2852                         struct fib6_node *fn;
2853
2854                         from = rcu_dereference(rt->from);
2855                         if (from) {
2856                                 fn = rcu_dereference(from->fib6_node);
2857                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2858                                         WRITE_ONCE(fn->fn_sernum, -1);
2859                         }
2860                 }
2861                 rcu_read_unlock();
2862         }
2863 }
2864
2865 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2866 {
2867         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2868                 struct fib6_info *from;
2869
2870                 rcu_read_lock();
2871                 from = rcu_dereference(rt0->from);
2872                 if (from)
2873                         rt0->dst.expires = from->expires;
2874                 rcu_read_unlock();
2875         }
2876
2877         dst_set_expires(&rt0->dst, timeout);
2878         rt0->rt6i_flags |= RTF_EXPIRES;
2879 }
2880
2881 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2882 {
2883         struct net *net = dev_net(rt->dst.dev);
2884
2885         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2886         rt->rt6i_flags |= RTF_MODIFIED;
2887         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2888 }
2889
2890 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2891 {
2892         return !(rt->rt6i_flags & RTF_CACHE) &&
2893                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2894 }
2895
2896 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2897                                  const struct ipv6hdr *iph, u32 mtu,
2898                                  bool confirm_neigh)
2899 {
2900         const struct in6_addr *daddr, *saddr;
2901         struct rt6_info *rt6 = dst_rt6_info(dst);
2902
2903         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2904          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2905          * [see also comment in rt6_mtu_change_route()]
2906          */
2907
2908         if (iph) {
2909                 daddr = &iph->daddr;
2910                 saddr = &iph->saddr;
2911         } else if (sk) {
2912                 daddr = &sk->sk_v6_daddr;
2913                 saddr = &inet6_sk(sk)->saddr;
2914         } else {
2915                 daddr = NULL;
2916                 saddr = NULL;
2917         }
2918
2919         if (confirm_neigh)
2920                 dst_confirm_neigh(dst, daddr);
2921
2922         if (mtu < IPV6_MIN_MTU)
2923                 return;
2924         if (mtu >= dst_mtu(dst))
2925                 return;
2926
2927         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2928                 rt6_do_update_pmtu(rt6, mtu);
2929                 /* update rt6_ex->stamp for cache */
2930                 if (rt6->rt6i_flags & RTF_CACHE)
2931                         rt6_update_exception_stamp_rt(rt6);
2932         } else if (daddr) {
2933                 struct fib6_result res = {};
2934                 struct rt6_info *nrt6;
2935
2936                 rcu_read_lock();
2937                 res.f6i = rcu_dereference(rt6->from);
2938                 if (!res.f6i)
2939                         goto out_unlock;
2940
2941                 res.fib6_flags = res.f6i->fib6_flags;
2942                 res.fib6_type = res.f6i->fib6_type;
2943
2944                 if (res.f6i->nh) {
2945                         struct fib6_nh_match_arg arg = {
2946                                 .dev = dst->dev,
2947                                 .gw = &rt6->rt6i_gateway,
2948                         };
2949
2950                         nexthop_for_each_fib6_nh(res.f6i->nh,
2951                                                  fib6_nh_find_match, &arg);
2952
2953                         /* fib6_info uses a nexthop that does not have fib6_nh
2954                          * using the dst->dev + gw. Should be impossible.
2955                          */
2956                         if (!arg.match)
2957                                 goto out_unlock;
2958
2959                         res.nh = arg.match;
2960                 } else {
2961                         res.nh = res.f6i->fib6_nh;
2962                 }
2963
2964                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2965                 if (nrt6) {
2966                         rt6_do_update_pmtu(nrt6, mtu);
2967                         if (rt6_insert_exception(nrt6, &res))
2968                                 dst_release_immediate(&nrt6->dst);
2969                 }
2970 out_unlock:
2971                 rcu_read_unlock();
2972         }
2973 }
2974
2975 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2976                                struct sk_buff *skb, u32 mtu,
2977                                bool confirm_neigh)
2978 {
2979         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2980                              confirm_neigh);
2981 }
2982
2983 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2984                      int oif, u32 mark, kuid_t uid)
2985 {
2986         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2987         struct dst_entry *dst;
2988         struct flowi6 fl6 = {
2989                 .flowi6_oif = oif,
2990                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2991                 .daddr = iph->daddr,
2992                 .saddr = iph->saddr,
2993                 .flowlabel = ip6_flowinfo(iph),
2994                 .flowi6_uid = uid,
2995         };
2996
2997         dst = ip6_route_output(net, NULL, &fl6);
2998         if (!dst->error)
2999                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
3000         dst_release(dst);
3001 }
3002 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
3003
3004 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
3005 {
3006         int oif = sk->sk_bound_dev_if;
3007         struct dst_entry *dst;
3008
3009         if (!oif && skb->dev)
3010                 oif = l3mdev_master_ifindex(skb->dev);
3011
3012         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, READ_ONCE(sk->sk_mark),
3013                         sk->sk_uid);
3014
3015         dst = __sk_dst_get(sk);
3016         if (!dst || !dst->obsolete ||
3017             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
3018                 return;
3019
3020         bh_lock_sock(sk);
3021         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
3022                 ip6_datagram_dst_update(sk, false);
3023         bh_unlock_sock(sk);
3024 }
3025 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
3026
3027 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
3028                            const struct flowi6 *fl6)
3029 {
3030 #ifdef CONFIG_IPV6_SUBTREES
3031         struct ipv6_pinfo *np = inet6_sk(sk);
3032 #endif
3033
3034         ip6_dst_store(sk, dst,
3035                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
3036                       &sk->sk_v6_daddr : NULL,
3037 #ifdef CONFIG_IPV6_SUBTREES
3038                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
3039                       &np->saddr :
3040 #endif
3041                       NULL);
3042 }
3043
3044 static bool ip6_redirect_nh_match(const struct fib6_result *res,
3045                                   struct flowi6 *fl6,
3046                                   const struct in6_addr *gw,
3047                                   struct rt6_info **ret)
3048 {
3049         const struct fib6_nh *nh = res->nh;
3050
3051         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
3052             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
3053                 return false;
3054
3055         /* rt_cache's gateway might be different from its 'parent'
3056          * in the case of an ip redirect.
3057          * So we keep searching in the exception table if the gateway
3058          * is different.
3059          */
3060         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3061                 struct rt6_info *rt_cache;
3062
3063                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3064                 if (rt_cache &&
3065                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3066                         *ret = rt_cache;
3067                         return true;
3068                 }
3069                 return false;
3070         }
3071         return true;
3072 }
3073
3074 struct fib6_nh_rd_arg {
3075         struct fib6_result      *res;
3076         struct flowi6           *fl6;
3077         const struct in6_addr   *gw;
3078         struct rt6_info         **ret;
3079 };
3080
3081 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3082 {
3083         struct fib6_nh_rd_arg *arg = _arg;
3084
3085         arg->res->nh = nh;
3086         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3087 }
3088
3089 /* Handle redirects */
3090 struct ip6rd_flowi {
3091         struct flowi6 fl6;
3092         struct in6_addr gateway;
3093 };
3094
3095 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3096                                              struct fib6_table *table,
3097                                              struct flowi6 *fl6,
3098                                              const struct sk_buff *skb,
3099                                              int flags)
3100 {
3101         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3102         struct rt6_info *ret = NULL;
3103         struct fib6_result res = {};
3104         struct fib6_nh_rd_arg arg = {
3105                 .res = &res,
3106                 .fl6 = fl6,
3107                 .gw  = &rdfl->gateway,
3108                 .ret = &ret
3109         };
3110         struct fib6_info *rt;
3111         struct fib6_node *fn;
3112
3113         /* Get the "current" route for this destination and
3114          * check if the redirect has come from appropriate router.
3115          *
3116          * RFC 4861 specifies that redirects should only be
3117          * accepted if they come from the nexthop to the target.
3118          * Due to the way the routes are chosen, this notion
3119          * is a bit fuzzy and one might need to check all possible
3120          * routes.
3121          */
3122
3123         rcu_read_lock();
3124         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3125 restart:
3126         for_each_fib6_node_rt_rcu(fn) {
3127                 res.f6i = rt;
3128                 if (fib6_check_expired(rt))
3129                         continue;
3130                 if (rt->fib6_flags & RTF_REJECT)
3131                         break;
3132                 if (unlikely(rt->nh)) {
3133                         if (nexthop_is_blackhole(rt->nh))
3134                                 continue;
3135                         /* on match, res->nh is filled in and potentially ret */
3136                         if (nexthop_for_each_fib6_nh(rt->nh,
3137                                                      fib6_nh_redirect_match,
3138                                                      &arg))
3139                                 goto out;
3140                 } else {
3141                         res.nh = rt->fib6_nh;
3142                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3143                                                   &ret))
3144                                 goto out;
3145                 }
3146         }
3147
3148         if (!rt)
3149                 rt = net->ipv6.fib6_null_entry;
3150         else if (rt->fib6_flags & RTF_REJECT) {
3151                 ret = net->ipv6.ip6_null_entry;
3152                 goto out;
3153         }
3154
3155         if (rt == net->ipv6.fib6_null_entry) {
3156                 fn = fib6_backtrack(fn, &fl6->saddr);
3157                 if (fn)
3158                         goto restart;
3159         }
3160
3161         res.f6i = rt;
3162         res.nh = rt->fib6_nh;
3163 out:
3164         if (ret) {
3165                 ip6_hold_safe(net, &ret);
3166         } else {
3167                 res.fib6_flags = res.f6i->fib6_flags;
3168                 res.fib6_type = res.f6i->fib6_type;
3169                 ret = ip6_create_rt_rcu(&res);
3170         }
3171
3172         rcu_read_unlock();
3173
3174         trace_fib6_table_lookup(net, &res, table, fl6);
3175         return ret;
3176 };
3177
3178 static struct dst_entry *ip6_route_redirect(struct net *net,
3179                                             const struct flowi6 *fl6,
3180                                             const struct sk_buff *skb,
3181                                             const struct in6_addr *gateway)
3182 {
3183         int flags = RT6_LOOKUP_F_HAS_SADDR;
3184         struct ip6rd_flowi rdfl;
3185
3186         rdfl.fl6 = *fl6;
3187         rdfl.gateway = *gateway;
3188
3189         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3190                                 flags, __ip6_route_redirect);
3191 }
3192
3193 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3194                   kuid_t uid)
3195 {
3196         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3197         struct dst_entry *dst;
3198         struct flowi6 fl6 = {
3199                 .flowi6_iif = LOOPBACK_IFINDEX,
3200                 .flowi6_oif = oif,
3201                 .flowi6_mark = mark,
3202                 .daddr = iph->daddr,
3203                 .saddr = iph->saddr,
3204                 .flowlabel = ip6_flowinfo(iph),
3205                 .flowi6_uid = uid,
3206         };
3207
3208         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3209         rt6_do_redirect(dst, NULL, skb);
3210         dst_release(dst);
3211 }
3212 EXPORT_SYMBOL_GPL(ip6_redirect);
3213
3214 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3215 {
3216         const struct ipv6hdr *iph = ipv6_hdr(skb);
3217         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3218         struct dst_entry *dst;
3219         struct flowi6 fl6 = {
3220                 .flowi6_iif = LOOPBACK_IFINDEX,
3221                 .flowi6_oif = oif,
3222                 .daddr = msg->dest,
3223                 .saddr = iph->daddr,
3224                 .flowi6_uid = sock_net_uid(net, NULL),
3225         };
3226
3227         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3228         rt6_do_redirect(dst, NULL, skb);
3229         dst_release(dst);
3230 }
3231
3232 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3233 {
3234         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if,
3235                      READ_ONCE(sk->sk_mark), sk->sk_uid);
3236 }
3237 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3238
3239 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3240 {
3241         struct net_device *dev = dst->dev;
3242         unsigned int mtu = dst_mtu(dst);
3243         struct net *net;
3244
3245         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3246
3247         rcu_read_lock();
3248
3249         net = dev_net_rcu(dev);
3250         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3251                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3252
3253         rcu_read_unlock();
3254
3255         /*
3256          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3257          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3258          * IPV6_MAXPLEN is also valid and means: "any MSS,
3259          * rely only on pmtu discovery"
3260          */
3261         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3262                 mtu = IPV6_MAXPLEN;
3263         return mtu;
3264 }
3265
3266 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3267 {
3268         return ip6_dst_mtu_maybe_forward(dst, false);
3269 }
3270 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3271
3272 /* MTU selection:
3273  * 1. mtu on route is locked - use it
3274  * 2. mtu from nexthop exception
3275  * 3. mtu from egress device
3276  *
3277  * based on ip6_dst_mtu_forward and exception logic of
3278  * rt6_find_cached_rt; called with rcu_read_lock
3279  */
3280 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3281                       const struct in6_addr *daddr,
3282                       const struct in6_addr *saddr)
3283 {
3284         const struct fib6_nh *nh = res->nh;
3285         struct fib6_info *f6i = res->f6i;
3286         struct inet6_dev *idev;
3287         struct rt6_info *rt;
3288         u32 mtu = 0;
3289
3290         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3291                 mtu = f6i->fib6_pmtu;
3292                 if (mtu)
3293                         goto out;
3294         }
3295
3296         rt = rt6_find_cached_rt(res, daddr, saddr);
3297         if (unlikely(rt)) {
3298                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3299         } else {
3300                 struct net_device *dev = nh->fib_nh_dev;
3301
3302                 mtu = IPV6_MIN_MTU;
3303                 idev = __in6_dev_get(dev);
3304                 if (idev)
3305                         mtu = max_t(u32, mtu, READ_ONCE(idev->cnf.mtu6));
3306         }
3307
3308         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3309 out:
3310         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3311 }
3312
3313 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3314                                   struct flowi6 *fl6)
3315 {
3316         struct dst_entry *dst;
3317         struct rt6_info *rt;
3318         struct inet6_dev *idev = in6_dev_get(dev);
3319         struct net *net = dev_net(dev);
3320
3321         if (unlikely(!idev))
3322                 return ERR_PTR(-ENODEV);
3323
3324         rt = ip6_dst_alloc(net, dev, 0);
3325         if (unlikely(!rt)) {
3326                 in6_dev_put(idev);
3327                 dst = ERR_PTR(-ENOMEM);
3328                 goto out;
3329         }
3330
3331         rt->dst.input = ip6_input;
3332         rt->dst.output  = ip6_output;
3333         rt->rt6i_gateway  = fl6->daddr;
3334         rt->rt6i_dst.addr = fl6->daddr;
3335         rt->rt6i_dst.plen = 128;
3336         rt->rt6i_idev     = idev;
3337         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3338
3339         /* Add this dst into uncached_list so that rt6_disable_ip() can
3340          * do proper release of the net_device
3341          */
3342         rt6_uncached_list_add(rt);
3343
3344         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3345
3346 out:
3347         return dst;
3348 }
3349
3350 static void ip6_dst_gc(struct dst_ops *ops)
3351 {
3352         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3353         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3354         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3355         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3356         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3357         unsigned int val;
3358         int entries;
3359
3360         if (time_after(rt_last_gc + rt_min_interval, jiffies))
3361                 goto out;
3362
3363         fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3364         entries = dst_entries_get_slow(ops);
3365         if (entries < ops->gc_thresh)
3366                 atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3367 out:
3368         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3369         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3370 }
3371
3372 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3373                                const struct in6_addr *gw_addr, u32 tbid,
3374                                int flags, struct fib6_result *res)
3375 {
3376         struct flowi6 fl6 = {
3377                 .flowi6_oif = cfg->fc_ifindex,
3378                 .daddr = *gw_addr,
3379                 .saddr = cfg->fc_prefsrc,
3380         };
3381         struct fib6_table *table;
3382         int err;
3383
3384         table = fib6_get_table(net, tbid);
3385         if (!table)
3386                 return -EINVAL;
3387
3388         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3389                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3390
3391         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3392
3393         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3394         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3395                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3396                                  cfg->fc_ifindex != 0, NULL, flags);
3397
3398         return err;
3399 }
3400
3401 static int ip6_route_check_nh_onlink(struct net *net,
3402                                      struct fib6_config *cfg,
3403                                      const struct net_device *dev,
3404                                      struct netlink_ext_ack *extack)
3405 {
3406         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3407         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3408         struct fib6_result res = {};
3409         int err;
3410
3411         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3412         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3413             /* ignore match if it is the default route */
3414             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3415             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3416                 NL_SET_ERR_MSG(extack,
3417                                "Nexthop has invalid gateway or device mismatch");
3418                 err = -EINVAL;
3419         }
3420
3421         return err;
3422 }
3423
3424 static int ip6_route_check_nh(struct net *net,
3425                               struct fib6_config *cfg,
3426                               struct net_device **_dev,
3427                               netdevice_tracker *dev_tracker,
3428                               struct inet6_dev **idev)
3429 {
3430         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3431         struct net_device *dev = _dev ? *_dev : NULL;
3432         int flags = RT6_LOOKUP_F_IFACE;
3433         struct fib6_result res = {};
3434         int err = -EHOSTUNREACH;
3435
3436         if (cfg->fc_table) {
3437                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3438                                           cfg->fc_table, flags, &res);
3439                 /* gw_addr can not require a gateway or resolve to a reject
3440                  * route. If a device is given, it must match the result.
3441                  */
3442                 if (err || res.fib6_flags & RTF_REJECT ||
3443                     res.nh->fib_nh_gw_family ||
3444                     (dev && dev != res.nh->fib_nh_dev))
3445                         err = -EHOSTUNREACH;
3446         }
3447
3448         if (err < 0) {
3449                 struct flowi6 fl6 = {
3450                         .flowi6_oif = cfg->fc_ifindex,
3451                         .daddr = *gw_addr,
3452                 };
3453
3454                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3455                 if (err || res.fib6_flags & RTF_REJECT ||
3456                     res.nh->fib_nh_gw_family)
3457                         err = -EHOSTUNREACH;
3458
3459                 if (err)
3460                         return err;
3461
3462                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3463                                  cfg->fc_ifindex != 0, NULL, flags);
3464         }
3465
3466         err = 0;
3467         if (dev) {
3468                 if (dev != res.nh->fib_nh_dev)
3469                         err = -EHOSTUNREACH;
3470         } else {
3471                 *_dev = dev = res.nh->fib_nh_dev;
3472                 netdev_hold(dev, dev_tracker, GFP_ATOMIC);
3473                 *idev = in6_dev_get(dev);
3474         }
3475
3476         return err;
3477 }
3478
3479 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3480                            struct net_device **_dev,
3481                            netdevice_tracker *dev_tracker,
3482                            struct inet6_dev **idev,
3483                            struct netlink_ext_ack *extack)
3484 {
3485         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3486         int gwa_type = ipv6_addr_type(gw_addr);
3487         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3488         const struct net_device *dev = *_dev;
3489         bool need_addr_check = !dev;
3490         int err = -EINVAL;
3491
3492         /* if gw_addr is local we will fail to detect this in case
3493          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3494          * will return already-added prefix route via interface that
3495          * prefix route was assigned to, which might be non-loopback.
3496          */
3497         if (dev &&
3498             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3499                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3500                 goto out;
3501         }
3502
3503         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3504                 /* IPv6 strictly inhibits using not link-local
3505                  * addresses as nexthop address.
3506                  * Otherwise, router will not able to send redirects.
3507                  * It is very good, but in some (rare!) circumstances
3508                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3509                  * some exceptions. --ANK
3510                  * We allow IPv4-mapped nexthops to support RFC4798-type
3511                  * addressing
3512                  */
3513                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3514                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3515                         goto out;
3516                 }
3517
3518                 rcu_read_lock();
3519
3520                 if (cfg->fc_flags & RTNH_F_ONLINK)
3521                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3522                 else
3523                         err = ip6_route_check_nh(net, cfg, _dev, dev_tracker,
3524                                                  idev);
3525
3526                 rcu_read_unlock();
3527
3528                 if (err)
3529                         goto out;
3530         }
3531
3532         /* reload in case device was changed */
3533         dev = *_dev;
3534
3535         err = -EINVAL;
3536         if (!dev) {
3537                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3538                 goto out;
3539         } else if (dev->flags & IFF_LOOPBACK) {
3540                 NL_SET_ERR_MSG(extack,
3541                                "Egress device can not be loopback device for this route");
3542                 goto out;
3543         }
3544
3545         /* if we did not check gw_addr above, do so now that the
3546          * egress device has been resolved.
3547          */
3548         if (need_addr_check &&
3549             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3550                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3551                 goto out;
3552         }
3553
3554         err = 0;
3555 out:
3556         return err;
3557 }
3558
3559 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3560 {
3561         if ((flags & RTF_REJECT) ||
3562             (dev && (dev->flags & IFF_LOOPBACK) &&
3563              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3564              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3565                 return true;
3566
3567         return false;
3568 }
3569
3570 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3571                  struct fib6_config *cfg, gfp_t gfp_flags,
3572                  struct netlink_ext_ack *extack)
3573 {
3574         netdevice_tracker *dev_tracker = &fib6_nh->fib_nh_dev_tracker;
3575         struct net_device *dev = NULL;
3576         struct inet6_dev *idev = NULL;
3577         int addr_type;
3578         int err;
3579
3580         fib6_nh->fib_nh_family = AF_INET6;
3581 #ifdef CONFIG_IPV6_ROUTER_PREF
3582         fib6_nh->last_probe = jiffies;
3583 #endif
3584         if (cfg->fc_is_fdb) {
3585                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3586                 fib6_nh->fib_nh_gw_family = AF_INET6;
3587                 return 0;
3588         }
3589
3590         err = -ENODEV;
3591         if (cfg->fc_ifindex) {
3592                 dev = netdev_get_by_index(net, cfg->fc_ifindex,
3593                                           dev_tracker, gfp_flags);
3594                 if (!dev)
3595                         goto out;
3596                 idev = in6_dev_get(dev);
3597                 if (!idev)
3598                         goto out;
3599         }
3600
3601         if (cfg->fc_flags & RTNH_F_ONLINK) {
3602                 if (!dev) {
3603                         NL_SET_ERR_MSG(extack,
3604                                        "Nexthop device required for onlink");
3605                         goto out;
3606                 }
3607
3608                 if (!(dev->flags & IFF_UP)) {
3609                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3610                         err = -ENETDOWN;
3611                         goto out;
3612                 }
3613
3614                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3615         }
3616
3617         fib6_nh->fib_nh_weight = 1;
3618
3619         /* We cannot add true routes via loopback here,
3620          * they would result in kernel looping; promote them to reject routes
3621          */
3622         addr_type = ipv6_addr_type(&cfg->fc_dst);
3623         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3624                 /* hold loopback dev/idev if we haven't done so. */
3625                 if (dev != net->loopback_dev) {
3626                         if (dev) {
3627                                 netdev_put(dev, dev_tracker);
3628                                 in6_dev_put(idev);
3629                         }
3630                         dev = net->loopback_dev;
3631                         netdev_hold(dev, dev_tracker, gfp_flags);
3632                         idev = in6_dev_get(dev);
3633                         if (!idev) {
3634                                 err = -ENODEV;
3635                                 goto out;
3636                         }
3637                 }
3638                 goto pcpu_alloc;
3639         }
3640
3641         if (cfg->fc_flags & RTF_GATEWAY) {
3642                 err = ip6_validate_gw(net, cfg, &dev, dev_tracker,
3643                                       &idev, extack);
3644                 if (err)
3645                         goto out;
3646
3647                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3648                 fib6_nh->fib_nh_gw_family = AF_INET6;
3649         }
3650
3651         err = -ENODEV;
3652         if (!dev)
3653                 goto out;
3654
3655         if (!idev || idev->cnf.disable_ipv6) {
3656                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3657                 err = -EACCES;
3658                 goto out;
3659         }
3660
3661         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3662                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3663                 err = -ENETDOWN;
3664                 goto out;
3665         }
3666
3667         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3668             !netif_carrier_ok(dev))
3669                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3670
3671         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3672                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3673         if (err)
3674                 goto out;
3675
3676 pcpu_alloc:
3677         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3678         if (!fib6_nh->rt6i_pcpu) {
3679                 err = -ENOMEM;
3680                 goto out;
3681         }
3682
3683         fib6_nh->fib_nh_dev = dev;
3684         fib6_nh->fib_nh_oif = dev->ifindex;
3685         err = 0;
3686 out:
3687         if (idev)
3688                 in6_dev_put(idev);
3689
3690         if (err) {
3691                 fib_nh_common_release(&fib6_nh->nh_common);
3692                 fib6_nh->nh_common.nhc_pcpu_rth_output = NULL;
3693                 fib6_nh->fib_nh_lws = NULL;
3694                 netdev_put(dev, dev_tracker);
3695         }
3696
3697         return err;
3698 }
3699
3700 void fib6_nh_release(struct fib6_nh *fib6_nh)
3701 {
3702         struct rt6_exception_bucket *bucket;
3703
3704         rcu_read_lock();
3705
3706         fib6_nh_flush_exceptions(fib6_nh, NULL);
3707         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3708         if (bucket) {
3709                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3710                 kfree(bucket);
3711         }
3712
3713         rcu_read_unlock();
3714
3715         fib6_nh_release_dsts(fib6_nh);
3716         free_percpu(fib6_nh->rt6i_pcpu);
3717
3718         fib_nh_common_release(&fib6_nh->nh_common);
3719 }
3720
3721 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3722 {
3723         int cpu;
3724
3725         if (!fib6_nh->rt6i_pcpu)
3726                 return;
3727
3728         for_each_possible_cpu(cpu) {
3729                 struct rt6_info *pcpu_rt, **ppcpu_rt;
3730
3731                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3732                 pcpu_rt = xchg(ppcpu_rt, NULL);
3733                 if (pcpu_rt) {
3734                         dst_dev_put(&pcpu_rt->dst);
3735                         dst_release(&pcpu_rt->dst);
3736                 }
3737         }
3738 }
3739
3740 static int fib6_config_validate(struct fib6_config *cfg,
3741                                 struct netlink_ext_ack *extack)
3742 {
3743         /* RTF_PCPU is an internal flag; can not be set by userspace */
3744         if (cfg->fc_flags & RTF_PCPU) {
3745                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3746                 goto errout;
3747         }
3748
3749         /* RTF_CACHE is an internal flag; can not be set by userspace */
3750         if (cfg->fc_flags & RTF_CACHE) {
3751                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3752                 goto errout;
3753         }
3754
3755         if (cfg->fc_type > RTN_MAX) {
3756                 NL_SET_ERR_MSG(extack, "Invalid route type");
3757                 goto errout;
3758         }
3759
3760         if (cfg->fc_dst_len > 128) {
3761                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3762                 goto errout;
3763         }
3764
3765 #ifdef CONFIG_IPV6_SUBTREES
3766         if (cfg->fc_src_len > 128) {
3767                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3768                 goto errout;
3769         }
3770
3771         if (cfg->fc_nh_id && cfg->fc_src_len) {
3772                 NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3773                 goto errout;
3774         }
3775 #else
3776         if (cfg->fc_src_len) {
3777                 NL_SET_ERR_MSG(extack,
3778                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3779                 goto errout;
3780         }
3781 #endif
3782         return 0;
3783 errout:
3784         return -EINVAL;
3785 }
3786
3787 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3788                                                gfp_t gfp_flags,
3789                                                struct netlink_ext_ack *extack)
3790 {
3791         struct net *net = cfg->fc_nlinfo.nl_net;
3792         struct fib6_table *table;
3793         struct fib6_info *rt;
3794         int err;
3795
3796         if (cfg->fc_nlinfo.nlh &&
3797             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3798                 table = fib6_get_table(net, cfg->fc_table);
3799                 if (!table) {
3800                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3801                         table = fib6_new_table(net, cfg->fc_table);
3802                 }
3803         } else {
3804                 table = fib6_new_table(net, cfg->fc_table);
3805         }
3806         if (!table) {
3807                 err = -ENOBUFS;
3808                 goto err;
3809         }
3810
3811         rt = fib6_info_alloc(gfp_flags, !cfg->fc_nh_id);
3812         if (!rt) {
3813                 err = -ENOMEM;
3814                 goto err;
3815         }
3816
3817         rt->fib6_metrics = ip_fib_metrics_init(cfg->fc_mx, cfg->fc_mx_len,
3818                                                extack);
3819         if (IS_ERR(rt->fib6_metrics)) {
3820                 err = PTR_ERR(rt->fib6_metrics);
3821                 goto free;
3822         }
3823
3824         if (cfg->fc_flags & RTF_ADDRCONF)
3825                 rt->dst_nocount = true;
3826
3827         if (cfg->fc_flags & RTF_EXPIRES)
3828                 fib6_set_expires(rt, jiffies +
3829                                  clock_t_to_jiffies(cfg->fc_expires));
3830
3831         if (cfg->fc_protocol == RTPROT_UNSPEC)
3832                 cfg->fc_protocol = RTPROT_BOOT;
3833
3834         rt->fib6_protocol = cfg->fc_protocol;
3835         rt->fib6_table = table;
3836         rt->fib6_metric = cfg->fc_metric;
3837         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3838         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3839
3840         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3841         rt->fib6_dst.plen = cfg->fc_dst_len;
3842
3843 #ifdef CONFIG_IPV6_SUBTREES
3844         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3845         rt->fib6_src.plen = cfg->fc_src_len;
3846 #endif
3847         return rt;
3848 free:
3849         kfree(rt);
3850 err:
3851         return ERR_PTR(err);
3852 }
3853
3854 static int ip6_route_info_create_nh(struct fib6_info *rt,
3855                                     struct fib6_config *cfg,
3856                                     gfp_t gfp_flags,
3857                                     struct netlink_ext_ack *extack)
3858 {
3859         struct net *net = cfg->fc_nlinfo.nl_net;
3860         struct fib6_nh *fib6_nh;
3861         int err;
3862
3863         if (cfg->fc_nh_id) {
3864                 struct nexthop *nh;
3865
3866                 rcu_read_lock();
3867
3868                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3869                 if (!nh) {
3870                         err = -EINVAL;
3871                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3872                         goto out_free;
3873                 }
3874
3875                 err = fib6_check_nexthop(nh, cfg, extack);
3876                 if (err)
3877                         goto out_free;
3878
3879                 if (!nexthop_get(nh)) {
3880                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3881                         err = -ENOENT;
3882                         goto out_free;
3883                 }
3884
3885                 rt->nh = nh;
3886                 fib6_nh = nexthop_fib6_nh(rt->nh);
3887
3888                 rcu_read_unlock();
3889         } else {
3890                 int addr_type;
3891
3892                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3893                 if (err)
3894                         goto out_release;
3895
3896                 fib6_nh = rt->fib6_nh;
3897
3898                 /* We cannot add true routes via loopback here, they would
3899                  * result in kernel looping; promote them to reject routes
3900                  */
3901                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3902                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3903                                    addr_type))
3904                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3905         }
3906
3907         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3908                 struct net_device *dev = fib6_nh->fib_nh_dev;
3909
3910                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3911                         NL_SET_ERR_MSG(extack, "Invalid source address");
3912                         err = -EINVAL;
3913                         goto out_release;
3914                 }
3915                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3916                 rt->fib6_prefsrc.plen = 128;
3917         }
3918
3919         return 0;
3920 out_release:
3921         fib6_info_release(rt);
3922         return err;
3923 out_free:
3924         rcu_read_unlock();
3925         ip_fib_metrics_put(rt->fib6_metrics);
3926         kfree(rt);
3927         return err;
3928 }
3929
3930 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3931                   struct netlink_ext_ack *extack)
3932 {
3933         struct fib6_info *rt;
3934         int err;
3935
3936         err = fib6_config_validate(cfg, extack);
3937         if (err)
3938                 return err;
3939
3940         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3941         if (IS_ERR(rt))
3942                 return PTR_ERR(rt);
3943
3944         err = ip6_route_info_create_nh(rt, cfg, gfp_flags, extack);
3945         if (err)
3946                 return err;
3947
3948         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3949         fib6_info_release(rt);
3950
3951         return err;
3952 }
3953
3954 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3955 {
3956         struct net *net = info->nl_net;
3957         struct fib6_table *table;
3958         int err;
3959
3960         if (rt == net->ipv6.fib6_null_entry) {
3961                 err = -ENOENT;
3962                 goto out;
3963         }
3964
3965         table = rt->fib6_table;
3966         spin_lock_bh(&table->tb6_lock);
3967         err = fib6_del(rt, info);
3968         spin_unlock_bh(&table->tb6_lock);
3969
3970 out:
3971         fib6_info_release(rt);
3972         return err;
3973 }
3974
3975 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3976 {
3977         struct nl_info info = {
3978                 .nl_net = net,
3979                 .skip_notify = skip_notify
3980         };
3981
3982         return __ip6_del_rt(rt, &info);
3983 }
3984
3985 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3986 {
3987         struct nl_info *info = &cfg->fc_nlinfo;
3988         struct net *net = info->nl_net;
3989         struct sk_buff *skb = NULL;
3990         struct fib6_table *table;
3991         int err = -ENOENT;
3992
3993         if (rt == net->ipv6.fib6_null_entry)
3994                 goto out_put;
3995         table = rt->fib6_table;
3996         spin_lock_bh(&table->tb6_lock);
3997
3998         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3999                 struct fib6_info *sibling, *next_sibling;
4000                 struct fib6_node *fn;
4001
4002                 /* prefer to send a single notification with all hops */
4003                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
4004                 if (skb) {
4005                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
4006
4007                         if (rt6_fill_node(net, skb, rt, NULL,
4008                                           NULL, NULL, 0, RTM_DELROUTE,
4009                                           info->portid, seq, 0) < 0) {
4010                                 kfree_skb(skb);
4011                                 skb = NULL;
4012                         } else
4013                                 info->skip_notify = 1;
4014                 }
4015
4016                 /* 'rt' points to the first sibling route. If it is not the
4017                  * leaf, then we do not need to send a notification. Otherwise,
4018                  * we need to check if the last sibling has a next route or not
4019                  * and emit a replace or delete notification, respectively.
4020                  */
4021                 info->skip_notify_kernel = 1;
4022                 fn = rcu_dereference_protected(rt->fib6_node,
4023                                             lockdep_is_held(&table->tb6_lock));
4024                 if (rcu_access_pointer(fn->leaf) == rt) {
4025                         struct fib6_info *last_sibling, *replace_rt;
4026
4027                         last_sibling = list_last_entry(&rt->fib6_siblings,
4028                                                        struct fib6_info,
4029                                                        fib6_siblings);
4030                         replace_rt = rcu_dereference_protected(
4031                                             last_sibling->fib6_next,
4032                                             lockdep_is_held(&table->tb6_lock));
4033                         if (replace_rt)
4034                                 call_fib6_entry_notifiers_replace(net,
4035                                                                   replace_rt);
4036                         else
4037                                 call_fib6_multipath_entry_notifiers(net,
4038                                                        FIB_EVENT_ENTRY_DEL,
4039                                                        rt, rt->fib6_nsiblings,
4040                                                        NULL);
4041                 }
4042                 list_for_each_entry_safe(sibling, next_sibling,
4043                                          &rt->fib6_siblings,
4044                                          fib6_siblings) {
4045                         err = fib6_del(sibling, info);
4046                         if (err)
4047                                 goto out_unlock;
4048                 }
4049         }
4050
4051         err = fib6_del(rt, info);
4052 out_unlock:
4053         spin_unlock_bh(&table->tb6_lock);
4054 out_put:
4055         fib6_info_release(rt);
4056
4057         if (skb) {
4058                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
4059                             info->nlh, gfp_any());
4060         }
4061         return err;
4062 }
4063
4064 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
4065 {
4066         int rc = -ESRCH;
4067
4068         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
4069                 goto out;
4070
4071         if (cfg->fc_flags & RTF_GATEWAY &&
4072             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
4073                 goto out;
4074
4075         rc = rt6_remove_exception_rt(rt);
4076 out:
4077         return rc;
4078 }
4079
4080 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
4081                              struct fib6_nh *nh)
4082 {
4083         struct fib6_result res = {
4084                 .f6i = rt,
4085                 .nh = nh,
4086         };
4087         struct rt6_info *rt_cache;
4088
4089         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
4090         if (rt_cache)
4091                 return __ip6_del_cached_rt(rt_cache, cfg);
4092
4093         return 0;
4094 }
4095
4096 struct fib6_nh_del_cached_rt_arg {
4097         struct fib6_config *cfg;
4098         struct fib6_info *f6i;
4099 };
4100
4101 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4102 {
4103         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4104         int rc;
4105
4106         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4107         return rc != -ESRCH ? rc : 0;
4108 }
4109
4110 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4111 {
4112         struct fib6_nh_del_cached_rt_arg arg = {
4113                 .cfg = cfg,
4114                 .f6i = f6i
4115         };
4116
4117         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4118 }
4119
4120 static int ip6_route_del(struct fib6_config *cfg,
4121                          struct netlink_ext_ack *extack)
4122 {
4123         struct fib6_table *table;
4124         struct fib6_info *rt;
4125         struct fib6_node *fn;
4126         int err = -ESRCH;
4127
4128         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4129         if (!table) {
4130                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4131                 return err;
4132         }
4133
4134         rcu_read_lock();
4135
4136         fn = fib6_locate(&table->tb6_root,
4137                          &cfg->fc_dst, cfg->fc_dst_len,
4138                          &cfg->fc_src, cfg->fc_src_len,
4139                          !(cfg->fc_flags & RTF_CACHE));
4140
4141         if (fn) {
4142                 for_each_fib6_node_rt_rcu(fn) {
4143                         struct fib6_nh *nh;
4144
4145                         if (rt->nh && cfg->fc_nh_id &&
4146                             rt->nh->id != cfg->fc_nh_id)
4147                                 continue;
4148
4149                         if (cfg->fc_flags & RTF_CACHE) {
4150                                 int rc = 0;
4151
4152                                 if (rt->nh) {
4153                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4154                                 } else if (cfg->fc_nh_id) {
4155                                         continue;
4156                                 } else {
4157                                         nh = rt->fib6_nh;
4158                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4159                                 }
4160                                 if (rc != -ESRCH) {
4161                                         rcu_read_unlock();
4162                                         return rc;
4163                                 }
4164                                 continue;
4165                         }
4166
4167                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4168                                 continue;
4169                         if (cfg->fc_protocol &&
4170                             cfg->fc_protocol != rt->fib6_protocol)
4171                                 continue;
4172
4173                         if (rt->nh) {
4174                                 if (!fib6_info_hold_safe(rt))
4175                                         continue;
4176
4177                                 err =  __ip6_del_rt(rt, &cfg->fc_nlinfo);
4178                                 break;
4179                         }
4180                         if (cfg->fc_nh_id)
4181                                 continue;
4182
4183                         nh = rt->fib6_nh;
4184                         if (cfg->fc_ifindex &&
4185                             (!nh->fib_nh_dev ||
4186                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4187                                 continue;
4188                         if (cfg->fc_flags & RTF_GATEWAY &&
4189                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4190                                 continue;
4191                         if (!fib6_info_hold_safe(rt))
4192                                 continue;
4193
4194                         /* if gateway was specified only delete the one hop */
4195                         if (cfg->fc_flags & RTF_GATEWAY)
4196                                 err = __ip6_del_rt(rt, &cfg->fc_nlinfo);
4197                         else
4198                                 err = __ip6_del_rt_siblings(rt, cfg);
4199                         break;
4200                 }
4201         }
4202         rcu_read_unlock();
4203
4204         return err;
4205 }
4206
4207 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4208 {
4209         struct netevent_redirect netevent;
4210         struct rt6_info *rt, *nrt = NULL;
4211         struct fib6_result res = {};
4212         struct ndisc_options ndopts;
4213         struct inet6_dev *in6_dev;
4214         struct neighbour *neigh;
4215         struct rd_msg *msg;
4216         int optlen, on_link;
4217         u8 *lladdr;
4218
4219         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4220         optlen -= sizeof(*msg);
4221
4222         if (optlen < 0) {
4223                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4224                 return;
4225         }
4226
4227         msg = (struct rd_msg *)icmp6_hdr(skb);
4228
4229         if (ipv6_addr_is_multicast(&msg->dest)) {
4230                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4231                 return;
4232         }
4233
4234         on_link = 0;
4235         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4236                 on_link = 1;
4237         } else if (ipv6_addr_type(&msg->target) !=
4238                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4239                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4240                 return;
4241         }
4242
4243         in6_dev = __in6_dev_get(skb->dev);
4244         if (!in6_dev)
4245                 return;
4246         if (READ_ONCE(in6_dev->cnf.forwarding) ||
4247             !READ_ONCE(in6_dev->cnf.accept_redirects))
4248                 return;
4249
4250         /* RFC2461 8.1:
4251          *      The IP source address of the Redirect MUST be the same as the current
4252          *      first-hop router for the specified ICMP Destination Address.
4253          */
4254
4255         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4256                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4257                 return;
4258         }
4259
4260         lladdr = NULL;
4261         if (ndopts.nd_opts_tgt_lladdr) {
4262                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4263                                              skb->dev);
4264                 if (!lladdr) {
4265                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4266                         return;
4267                 }
4268         }
4269
4270         rt = dst_rt6_info(dst);
4271         if (rt->rt6i_flags & RTF_REJECT) {
4272                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4273                 return;
4274         }
4275
4276         /* Redirect received -> path was valid.
4277          * Look, redirects are sent only in response to data packets,
4278          * so that this nexthop apparently is reachable. --ANK
4279          */
4280         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4281
4282         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4283         if (!neigh)
4284                 return;
4285
4286         /*
4287          *      We have finally decided to accept it.
4288          */
4289
4290         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4291                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4292                      NEIGH_UPDATE_F_OVERRIDE|
4293                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4294                                      NEIGH_UPDATE_F_ISROUTER)),
4295                      NDISC_REDIRECT, &ndopts);
4296
4297         rcu_read_lock();
4298         res.f6i = rcu_dereference(rt->from);
4299         if (!res.f6i)
4300                 goto out;
4301
4302         if (res.f6i->nh) {
4303                 struct fib6_nh_match_arg arg = {
4304                         .dev = dst->dev,
4305                         .gw = &rt->rt6i_gateway,
4306                 };
4307
4308                 nexthop_for_each_fib6_nh(res.f6i->nh,
4309                                          fib6_nh_find_match, &arg);
4310
4311                 /* fib6_info uses a nexthop that does not have fib6_nh
4312                  * using the dst->dev. Should be impossible
4313                  */
4314                 if (!arg.match)
4315                         goto out;
4316                 res.nh = arg.match;
4317         } else {
4318                 res.nh = res.f6i->fib6_nh;
4319         }
4320
4321         res.fib6_flags = res.f6i->fib6_flags;
4322         res.fib6_type = res.f6i->fib6_type;
4323         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4324         if (!nrt)
4325                 goto out;
4326
4327         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4328         if (on_link)
4329                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4330
4331         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4332
4333         /* rt6_insert_exception() will take care of duplicated exceptions */
4334         if (rt6_insert_exception(nrt, &res)) {
4335                 dst_release_immediate(&nrt->dst);
4336                 goto out;
4337         }
4338
4339         netevent.old = &rt->dst;
4340         netevent.new = &nrt->dst;
4341         netevent.daddr = &msg->dest;
4342         netevent.neigh = neigh;
4343         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4344
4345 out:
4346         rcu_read_unlock();
4347         neigh_release(neigh);
4348 }
4349
4350 #ifdef CONFIG_IPV6_ROUTE_INFO
4351 static struct fib6_info *rt6_get_route_info(struct net *net,
4352                                            const struct in6_addr *prefix, int prefixlen,
4353                                            const struct in6_addr *gwaddr,
4354                                            struct net_device *dev)
4355 {
4356         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4357         int ifindex = dev->ifindex;
4358         struct fib6_node *fn;
4359         struct fib6_info *rt = NULL;
4360         struct fib6_table *table;
4361
4362         table = fib6_get_table(net, tb_id);
4363         if (!table)
4364                 return NULL;
4365
4366         rcu_read_lock();
4367         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4368         if (!fn)
4369                 goto out;
4370
4371         for_each_fib6_node_rt_rcu(fn) {
4372                 /* these routes do not use nexthops */
4373                 if (rt->nh)
4374                         continue;
4375                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4376                         continue;
4377                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4378                     !rt->fib6_nh->fib_nh_gw_family)
4379                         continue;
4380                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4381                         continue;
4382                 if (!fib6_info_hold_safe(rt))
4383                         continue;
4384                 break;
4385         }
4386 out:
4387         rcu_read_unlock();
4388         return rt;
4389 }
4390
4391 static struct fib6_info *rt6_add_route_info(struct net *net,
4392                                            const struct in6_addr *prefix, int prefixlen,
4393                                            const struct in6_addr *gwaddr,
4394                                            struct net_device *dev,
4395                                            unsigned int pref)
4396 {
4397         struct fib6_config cfg = {
4398                 .fc_metric      = IP6_RT_PRIO_USER,
4399                 .fc_ifindex     = dev->ifindex,
4400                 .fc_dst_len     = prefixlen,
4401                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4402                                   RTF_UP | RTF_PREF(pref),
4403                 .fc_protocol = RTPROT_RA,
4404                 .fc_type = RTN_UNICAST,
4405                 .fc_nlinfo.portid = 0,
4406                 .fc_nlinfo.nlh = NULL,
4407                 .fc_nlinfo.nl_net = net,
4408         };
4409
4410         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4411         cfg.fc_dst = *prefix;
4412         cfg.fc_gateway = *gwaddr;
4413
4414         /* We should treat it as a default route if prefix length is 0. */
4415         if (!prefixlen)
4416                 cfg.fc_flags |= RTF_DEFAULT;
4417
4418         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4419
4420         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4421 }
4422 #endif
4423
4424 struct fib6_info *rt6_get_dflt_router(struct net *net,
4425                                      const struct in6_addr *addr,
4426                                      struct net_device *dev)
4427 {
4428         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4429         struct fib6_info *rt;
4430         struct fib6_table *table;
4431
4432         table = fib6_get_table(net, tb_id);
4433         if (!table)
4434                 return NULL;
4435
4436         rcu_read_lock();
4437         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4438                 struct fib6_nh *nh;
4439
4440                 /* RA routes do not use nexthops */
4441                 if (rt->nh)
4442                         continue;
4443
4444                 nh = rt->fib6_nh;
4445                 if (dev == nh->fib_nh_dev &&
4446                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4447                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4448                         break;
4449         }
4450         if (rt && !fib6_info_hold_safe(rt))
4451                 rt = NULL;
4452         rcu_read_unlock();
4453         return rt;
4454 }
4455
4456 struct fib6_info *rt6_add_dflt_router(struct net *net,
4457                                      const struct in6_addr *gwaddr,
4458                                      struct net_device *dev,
4459                                      unsigned int pref,
4460                                      u32 defrtr_usr_metric,
4461                                      int lifetime)
4462 {
4463         struct fib6_config cfg = {
4464                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4465                 .fc_metric      = defrtr_usr_metric,
4466                 .fc_ifindex     = dev->ifindex,
4467                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4468                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4469                 .fc_protocol = RTPROT_RA,
4470                 .fc_type = RTN_UNICAST,
4471                 .fc_nlinfo.portid = 0,
4472                 .fc_nlinfo.nlh = NULL,
4473                 .fc_nlinfo.nl_net = net,
4474                 .fc_expires = jiffies_to_clock_t(lifetime * HZ),
4475         };
4476
4477         cfg.fc_gateway = *gwaddr;
4478
4479         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4480                 struct fib6_table *table;
4481
4482                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4483                 if (table)
4484                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4485         }
4486
4487         return rt6_get_dflt_router(net, gwaddr, dev);
4488 }
4489
4490 static void __rt6_purge_dflt_routers(struct net *net,
4491                                      struct fib6_table *table)
4492 {
4493         struct fib6_info *rt;
4494
4495 restart:
4496         rcu_read_lock();
4497         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4498                 struct net_device *dev = fib6_info_nh_dev(rt);
4499                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4500
4501                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4502                     (!idev || idev->cnf.accept_ra != 2) &&
4503                     fib6_info_hold_safe(rt)) {
4504                         rcu_read_unlock();
4505                         ip6_del_rt(net, rt, false);
4506                         goto restart;
4507                 }
4508         }
4509         rcu_read_unlock();
4510
4511         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4512 }
4513
4514 void rt6_purge_dflt_routers(struct net *net)
4515 {
4516         struct fib6_table *table;
4517         struct hlist_head *head;
4518         unsigned int h;
4519
4520         rcu_read_lock();
4521
4522         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4523                 head = &net->ipv6.fib_table_hash[h];
4524                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4525                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4526                                 __rt6_purge_dflt_routers(net, table);
4527                 }
4528         }
4529
4530         rcu_read_unlock();
4531 }
4532
4533 static void rtmsg_to_fib6_config(struct net *net,
4534                                  struct in6_rtmsg *rtmsg,
4535                                  struct fib6_config *cfg)
4536 {
4537         *cfg = (struct fib6_config){
4538                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4539                          : RT6_TABLE_MAIN,
4540                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4541                 .fc_metric = rtmsg->rtmsg_metric,
4542                 .fc_expires = rtmsg->rtmsg_info,
4543                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4544                 .fc_src_len = rtmsg->rtmsg_src_len,
4545                 .fc_flags = rtmsg->rtmsg_flags,
4546                 .fc_type = rtmsg->rtmsg_type,
4547
4548                 .fc_nlinfo.nl_net = net,
4549
4550                 .fc_dst = rtmsg->rtmsg_dst,
4551                 .fc_src = rtmsg->rtmsg_src,
4552                 .fc_gateway = rtmsg->rtmsg_gateway,
4553         };
4554 }
4555
4556 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4557 {
4558         struct fib6_config cfg;
4559         int err;
4560
4561         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4562                 return -EINVAL;
4563         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4564                 return -EPERM;
4565
4566         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4567
4568         switch (cmd) {
4569         case SIOCADDRT:
4570                 /* Only do the default setting of fc_metric in route adding */
4571                 if (cfg.fc_metric == 0)
4572                         cfg.fc_metric = IP6_RT_PRIO_USER;
4573                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4574                 break;
4575         case SIOCDELRT:
4576                 err = ip6_route_del(&cfg, NULL);
4577                 break;
4578         }
4579
4580         return err;
4581 }
4582
4583 /*
4584  *      Drop the packet on the floor
4585  */
4586
4587 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4588 {
4589         struct dst_entry *dst = skb_dst(skb);
4590         struct net *net = dev_net(dst->dev);
4591         struct inet6_dev *idev;
4592         SKB_DR(reason);
4593         int type;
4594
4595         if (netif_is_l3_master(skb->dev) ||
4596             dst->dev == net->loopback_dev)
4597                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4598         else
4599                 idev = ip6_dst_idev(dst);
4600
4601         switch (ipstats_mib_noroutes) {
4602         case IPSTATS_MIB_INNOROUTES:
4603                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4604                 if (type == IPV6_ADDR_ANY) {
4605                         SKB_DR_SET(reason, IP_INADDRERRORS);
4606                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4607                         break;
4608                 }
4609                 SKB_DR_SET(reason, IP_INNOROUTES);
4610                 fallthrough;
4611         case IPSTATS_MIB_OUTNOROUTES:
4612                 SKB_DR_OR(reason, IP_OUTNOROUTES);
4613                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4614                 break;
4615         }
4616
4617         /* Start over by dropping the dst for l3mdev case */
4618         if (netif_is_l3_master(skb->dev))
4619                 skb_dst_drop(skb);
4620
4621         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4622         kfree_skb_reason(skb, reason);
4623         return 0;
4624 }
4625
4626 static int ip6_pkt_discard(struct sk_buff *skb)
4627 {
4628         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4629 }
4630
4631 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4632 {
4633         skb->dev = skb_dst(skb)->dev;
4634         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4635 }
4636
4637 static int ip6_pkt_prohibit(struct sk_buff *skb)
4638 {
4639         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4640 }
4641
4642 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4643 {
4644         skb->dev = skb_dst(skb)->dev;
4645         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4646 }
4647
4648 /*
4649  *      Allocate a dst for local (unicast / anycast) address.
4650  */
4651
4652 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4653                                      struct inet6_dev *idev,
4654                                      const struct in6_addr *addr,
4655                                      bool anycast, gfp_t gfp_flags,
4656                                      struct netlink_ext_ack *extack)
4657 {
4658         struct fib6_config cfg = {
4659                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4660                 .fc_ifindex = idev->dev->ifindex,
4661                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4662                 .fc_dst = *addr,
4663                 .fc_dst_len = 128,
4664                 .fc_protocol = RTPROT_KERNEL,
4665                 .fc_nlinfo.nl_net = net,
4666                 .fc_ignore_dev_down = true,
4667         };
4668         struct fib6_info *f6i;
4669         int err;
4670
4671         if (anycast) {
4672                 cfg.fc_type = RTN_ANYCAST;
4673                 cfg.fc_flags |= RTF_ANYCAST;
4674         } else {
4675                 cfg.fc_type = RTN_LOCAL;
4676                 cfg.fc_flags |= RTF_LOCAL;
4677         }
4678
4679         f6i = ip6_route_info_create(&cfg, gfp_flags, extack);
4680         if (IS_ERR(f6i))
4681                 return f6i;
4682
4683         err = ip6_route_info_create_nh(f6i, &cfg, gfp_flags, extack);
4684         if (err)
4685                 return ERR_PTR(err);
4686
4687         f6i->dst_nocount = true;
4688
4689         if (!anycast &&
4690             (READ_ONCE(net->ipv6.devconf_all->disable_policy) ||
4691              READ_ONCE(idev->cnf.disable_policy)))
4692                 f6i->dst_nopolicy = true;
4693
4694         return f6i;
4695 }
4696
4697 /* remove deleted ip from prefsrc entries */
4698 struct arg_dev_net_ip {
4699         struct net *net;
4700         struct in6_addr *addr;
4701 };
4702
4703 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4704 {
4705         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4706         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4707
4708         if (!rt->nh &&
4709             rt != net->ipv6.fib6_null_entry &&
4710             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr) &&
4711             !ipv6_chk_addr(net, addr, rt->fib6_nh->fib_nh_dev, 0)) {
4712                 spin_lock_bh(&rt6_exception_lock);
4713                 /* remove prefsrc entry */
4714                 rt->fib6_prefsrc.plen = 0;
4715                 spin_unlock_bh(&rt6_exception_lock);
4716         }
4717         return 0;
4718 }
4719
4720 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4721 {
4722         struct net *net = dev_net(ifp->idev->dev);
4723         struct arg_dev_net_ip adni = {
4724                 .net = net,
4725                 .addr = &ifp->addr,
4726         };
4727         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4728 }
4729
4730 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4731
4732 /* Remove routers and update dst entries when gateway turn into host. */
4733 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4734 {
4735         struct in6_addr *gateway = (struct in6_addr *)arg;
4736         struct fib6_nh *nh;
4737
4738         /* RA routes do not use nexthops */
4739         if (rt->nh)
4740                 return 0;
4741
4742         nh = rt->fib6_nh;
4743         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4744             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4745                 return -1;
4746
4747         /* Further clean up cached routes in exception table.
4748          * This is needed because cached route may have a different
4749          * gateway than its 'parent' in the case of an ip redirect.
4750          */
4751         fib6_nh_exceptions_clean_tohost(nh, gateway);
4752
4753         return 0;
4754 }
4755
4756 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4757 {
4758         fib6_clean_all(net, fib6_clean_tohost, gateway);
4759 }
4760
4761 struct arg_netdev_event {
4762         const struct net_device *dev;
4763         union {
4764                 unsigned char nh_flags;
4765                 unsigned long event;
4766         };
4767 };
4768
4769 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4770 {
4771         struct fib6_info *iter;
4772         struct fib6_node *fn;
4773
4774         fn = rcu_dereference_protected(rt->fib6_node,
4775                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4776         iter = rcu_dereference_protected(fn->leaf,
4777                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4778         while (iter) {
4779                 if (iter->fib6_metric == rt->fib6_metric &&
4780                     rt6_qualify_for_ecmp(iter))
4781                         return iter;
4782                 iter = rcu_dereference_protected(iter->fib6_next,
4783                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4784         }
4785
4786         return NULL;
4787 }
4788
4789 /* only called for fib entries with builtin fib6_nh */
4790 static bool rt6_is_dead(const struct fib6_info *rt)
4791 {
4792         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4793             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4794              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4795                 return true;
4796
4797         return false;
4798 }
4799
4800 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4801 {
4802         struct fib6_info *iter;
4803         int total = 0;
4804
4805         if (!rt6_is_dead(rt))
4806                 total += rt->fib6_nh->fib_nh_weight;
4807
4808         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4809                 if (!rt6_is_dead(iter))
4810                         total += iter->fib6_nh->fib_nh_weight;
4811         }
4812
4813         return total;
4814 }
4815
4816 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4817 {
4818         int upper_bound = -1;
4819
4820         if (!rt6_is_dead(rt)) {
4821                 *weight += rt->fib6_nh->fib_nh_weight;
4822                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4823                                                     total) - 1;
4824         }
4825         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4826 }
4827
4828 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4829 {
4830         struct fib6_info *iter;
4831         int weight = 0;
4832
4833         rt6_upper_bound_set(rt, &weight, total);
4834
4835         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4836                 rt6_upper_bound_set(iter, &weight, total);
4837 }
4838
4839 void rt6_multipath_rebalance(struct fib6_info *rt)
4840 {
4841         struct fib6_info *first;
4842         int total;
4843
4844         /* In case the entire multipath route was marked for flushing,
4845          * then there is no need to rebalance upon the removal of every
4846          * sibling route.
4847          */
4848         if (!rt->fib6_nsiblings || rt->should_flush)
4849                 return;
4850
4851         /* During lookup routes are evaluated in order, so we need to
4852          * make sure upper bounds are assigned from the first sibling
4853          * onwards.
4854          */
4855         first = rt6_multipath_first_sibling(rt);
4856         if (WARN_ON_ONCE(!first))
4857                 return;
4858
4859         total = rt6_multipath_total_weight(first);
4860         rt6_multipath_upper_bound_set(first, total);
4861 }
4862
4863 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4864 {
4865         const struct arg_netdev_event *arg = p_arg;
4866         struct net *net = dev_net(arg->dev);
4867
4868         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4869             rt->fib6_nh->fib_nh_dev == arg->dev) {
4870                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4871                 fib6_update_sernum_upto_root(net, rt);
4872                 rt6_multipath_rebalance(rt);
4873         }
4874
4875         return 0;
4876 }
4877
4878 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4879 {
4880         struct arg_netdev_event arg = {
4881                 .dev = dev,
4882                 {
4883                         .nh_flags = nh_flags,
4884                 },
4885         };
4886
4887         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4888                 arg.nh_flags |= RTNH_F_LINKDOWN;
4889
4890         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4891 }
4892
4893 /* only called for fib entries with inline fib6_nh */
4894 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4895                                    const struct net_device *dev)
4896 {
4897         struct fib6_info *iter;
4898
4899         if (rt->fib6_nh->fib_nh_dev == dev)
4900                 return true;
4901         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4902                 if (iter->fib6_nh->fib_nh_dev == dev)
4903                         return true;
4904
4905         return false;
4906 }
4907
4908 static void rt6_multipath_flush(struct fib6_info *rt)
4909 {
4910         struct fib6_info *iter;
4911
4912         rt->should_flush = 1;
4913         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4914                 iter->should_flush = 1;
4915 }
4916
4917 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4918                                              const struct net_device *down_dev)
4919 {
4920         struct fib6_info *iter;
4921         unsigned int dead = 0;
4922
4923         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4924             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4925                 dead++;
4926         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4927                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4928                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4929                         dead++;
4930
4931         return dead;
4932 }
4933
4934 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4935                                        const struct net_device *dev,
4936                                        unsigned char nh_flags)
4937 {
4938         struct fib6_info *iter;
4939
4940         if (rt->fib6_nh->fib_nh_dev == dev)
4941                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4942         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4943                 if (iter->fib6_nh->fib_nh_dev == dev)
4944                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4945 }
4946
4947 /* called with write lock held for table with rt */
4948 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4949 {
4950         const struct arg_netdev_event *arg = p_arg;
4951         const struct net_device *dev = arg->dev;
4952         struct net *net = dev_net(dev);
4953
4954         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4955                 return 0;
4956
4957         switch (arg->event) {
4958         case NETDEV_UNREGISTER:
4959                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4960         case NETDEV_DOWN:
4961                 if (rt->should_flush)
4962                         return -1;
4963                 if (!rt->fib6_nsiblings)
4964                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4965                 if (rt6_multipath_uses_dev(rt, dev)) {
4966                         unsigned int count;
4967
4968                         count = rt6_multipath_dead_count(rt, dev);
4969                         if (rt->fib6_nsiblings + 1 == count) {
4970                                 rt6_multipath_flush(rt);
4971                                 return -1;
4972                         }
4973                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4974                                                    RTNH_F_LINKDOWN);
4975                         fib6_update_sernum(net, rt);
4976                         rt6_multipath_rebalance(rt);
4977                 }
4978                 return -2;
4979         case NETDEV_CHANGE:
4980                 if (rt->fib6_nh->fib_nh_dev != dev ||
4981                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4982                         break;
4983                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4984                 rt6_multipath_rebalance(rt);
4985                 break;
4986         }
4987
4988         return 0;
4989 }
4990
4991 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4992 {
4993         struct arg_netdev_event arg = {
4994                 .dev = dev,
4995                 {
4996                         .event = event,
4997                 },
4998         };
4999         struct net *net = dev_net(dev);
5000
5001         if (net->ipv6.sysctl.skip_notify_on_dev_down)
5002                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
5003         else
5004                 fib6_clean_all(net, fib6_ifdown, &arg);
5005 }
5006
5007 void rt6_disable_ip(struct net_device *dev, unsigned long event)
5008 {
5009         rt6_sync_down_dev(dev, event);
5010         rt6_uncached_list_flush_dev(dev);
5011         neigh_ifdown(&nd_tbl, dev);
5012 }
5013
5014 struct rt6_mtu_change_arg {
5015         struct net_device *dev;
5016         unsigned int mtu;
5017         struct fib6_info *f6i;
5018 };
5019
5020 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
5021 {
5022         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
5023         struct fib6_info *f6i = arg->f6i;
5024
5025         /* For administrative MTU increase, there is no way to discover
5026          * IPv6 PMTU increase, so PMTU increase should be updated here.
5027          * Since RFC 1981 doesn't include administrative MTU increase
5028          * update PMTU increase is a MUST. (i.e. jumbo frame)
5029          */
5030         if (nh->fib_nh_dev == arg->dev) {
5031                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
5032                 u32 mtu = f6i->fib6_pmtu;
5033
5034                 if (mtu >= arg->mtu ||
5035                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
5036                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
5037
5038                 spin_lock_bh(&rt6_exception_lock);
5039                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
5040                 spin_unlock_bh(&rt6_exception_lock);
5041         }
5042
5043         return 0;
5044 }
5045
5046 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
5047 {
5048         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
5049         struct inet6_dev *idev;
5050
5051         /* In IPv6 pmtu discovery is not optional,
5052            so that RTAX_MTU lock cannot disable it.
5053            We still use this lock to block changes
5054            caused by addrconf/ndisc.
5055         */
5056
5057         idev = __in6_dev_get(arg->dev);
5058         if (!idev)
5059                 return 0;
5060
5061         if (fib6_metric_locked(f6i, RTAX_MTU))
5062                 return 0;
5063
5064         arg->f6i = f6i;
5065         if (f6i->nh) {
5066                 /* fib6_nh_mtu_change only returns 0, so this is safe */
5067                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
5068                                                 arg);
5069         }
5070
5071         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
5072 }
5073
5074 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
5075 {
5076         struct rt6_mtu_change_arg arg = {
5077                 .dev = dev,
5078                 .mtu = mtu,
5079         };
5080
5081         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
5082 }
5083
5084 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
5085         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
5086         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
5087         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
5088         [RTA_OIF]               = { .type = NLA_U32 },
5089         [RTA_IIF]               = { .type = NLA_U32 },
5090         [RTA_PRIORITY]          = { .type = NLA_U32 },
5091         [RTA_METRICS]           = { .type = NLA_NESTED },
5092         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
5093         [RTA_PREF]              = { .type = NLA_U8 },
5094         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
5095         [RTA_ENCAP]             = { .type = NLA_NESTED },
5096         [RTA_EXPIRES]           = { .type = NLA_U32 },
5097         [RTA_UID]               = { .type = NLA_U32 },
5098         [RTA_MARK]              = { .type = NLA_U32 },
5099         [RTA_TABLE]             = { .type = NLA_U32 },
5100         [RTA_IP_PROTO]          = { .type = NLA_U8 },
5101         [RTA_SPORT]             = { .type = NLA_U16 },
5102         [RTA_DPORT]             = { .type = NLA_U16 },
5103         [RTA_NH_ID]             = { .type = NLA_U32 },
5104         [RTA_FLOWLABEL]         = { .type = NLA_BE32 },
5105 };
5106
5107 static int rtm_to_fib6_multipath_config(struct fib6_config *cfg,
5108                                         struct netlink_ext_ack *extack,
5109                                         bool newroute)
5110 {
5111         struct rtnexthop *rtnh;
5112         int remaining;
5113
5114         remaining = cfg->fc_mp_len;
5115         rtnh = (struct rtnexthop *)cfg->fc_mp;
5116
5117         if (!rtnh_ok(rtnh, remaining)) {
5118                 NL_SET_ERR_MSG(extack, "Invalid nexthop configuration - no valid nexthops");
5119                 return -EINVAL;
5120         }
5121
5122         do {
5123                 bool has_gateway = cfg->fc_flags & RTF_GATEWAY;
5124                 int attrlen = rtnh_attrlen(rtnh);
5125
5126                 if (attrlen > 0) {
5127                         struct nlattr *nla, *attrs;
5128
5129                         attrs = rtnh_attrs(rtnh);
5130                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5131                         if (nla) {
5132                                 if (nla_len(nla) < sizeof(cfg->fc_gateway)) {
5133                                         NL_SET_ERR_MSG(extack,
5134                                                        "Invalid IPv6 address in RTA_GATEWAY");
5135                                         return -EINVAL;
5136                                 }
5137
5138                                 has_gateway = true;
5139                         }
5140                 }
5141
5142                 if (newroute && (cfg->fc_nh_id || !has_gateway)) {
5143                         NL_SET_ERR_MSG(extack,
5144                                        "Device only routes can not be added for IPv6 using the multipath API.");
5145                         return -EINVAL;
5146                 }
5147
5148                 rtnh = rtnh_next(rtnh, &remaining);
5149         } while (rtnh_ok(rtnh, remaining));
5150
5151         return lwtunnel_valid_encap_type_attr(cfg->fc_mp, cfg->fc_mp_len, extack);
5152 }
5153
5154 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
5155                               struct fib6_config *cfg,
5156                               struct netlink_ext_ack *extack)
5157 {
5158         bool newroute = nlh->nlmsg_type == RTM_NEWROUTE;
5159         struct nlattr *tb[RTA_MAX+1];
5160         struct rtmsg *rtm;
5161         unsigned int pref;
5162         int err;
5163
5164         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5165                                      rtm_ipv6_policy, extack);
5166         if (err < 0)
5167                 goto errout;
5168
5169         err = -EINVAL;
5170         rtm = nlmsg_data(nlh);
5171
5172         if (rtm->rtm_tos) {
5173                 NL_SET_ERR_MSG(extack,
5174                                "Invalid dsfield (tos): option not available for IPv6");
5175                 goto errout;
5176         }
5177
5178         if (tb[RTA_FLOWLABEL]) {
5179                 NL_SET_ERR_MSG_ATTR(extack, tb[RTA_FLOWLABEL],
5180                                     "Flow label cannot be specified for this operation");
5181                 goto errout;
5182         }
5183
5184         *cfg = (struct fib6_config){
5185                 .fc_table = rtm->rtm_table,
5186                 .fc_dst_len = rtm->rtm_dst_len,
5187                 .fc_src_len = rtm->rtm_src_len,
5188                 .fc_flags = RTF_UP,
5189                 .fc_protocol = rtm->rtm_protocol,
5190                 .fc_type = rtm->rtm_type,
5191
5192                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5193                 .fc_nlinfo.nlh = nlh,
5194                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5195         };
5196
5197         if (rtm->rtm_type == RTN_UNREACHABLE ||
5198             rtm->rtm_type == RTN_BLACKHOLE ||
5199             rtm->rtm_type == RTN_PROHIBIT ||
5200             rtm->rtm_type == RTN_THROW)
5201                 cfg->fc_flags |= RTF_REJECT;
5202
5203         if (rtm->rtm_type == RTN_LOCAL)
5204                 cfg->fc_flags |= RTF_LOCAL;
5205
5206         if (rtm->rtm_flags & RTM_F_CLONED)
5207                 cfg->fc_flags |= RTF_CACHE;
5208
5209         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5210
5211         if (tb[RTA_NH_ID]) {
5212                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5213                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5214                         NL_SET_ERR_MSG(extack,
5215                                        "Nexthop specification and nexthop id are mutually exclusive");
5216                         goto errout;
5217                 }
5218                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5219         }
5220
5221         if (tb[RTA_GATEWAY]) {
5222                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5223                 cfg->fc_flags |= RTF_GATEWAY;
5224         }
5225         if (tb[RTA_VIA]) {
5226                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5227                 goto errout;
5228         }
5229
5230         if (tb[RTA_DST]) {
5231                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5232
5233                 if (nla_len(tb[RTA_DST]) < plen)
5234                         goto errout;
5235
5236                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5237         }
5238
5239         if (tb[RTA_SRC]) {
5240                 int plen = (rtm->rtm_src_len + 7) >> 3;
5241
5242                 if (nla_len(tb[RTA_SRC]) < plen)
5243                         goto errout;
5244
5245                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5246         }
5247
5248         if (tb[RTA_PREFSRC])
5249                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5250
5251         if (tb[RTA_OIF])
5252                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5253
5254         if (tb[RTA_PRIORITY])
5255                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5256
5257         if (tb[RTA_METRICS]) {
5258                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5259                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5260         }
5261
5262         if (tb[RTA_TABLE])
5263                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5264
5265         if (tb[RTA_MULTIPATH]) {
5266                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5267                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5268
5269                 err = rtm_to_fib6_multipath_config(cfg, extack, newroute);
5270                 if (err < 0)
5271                         goto errout;
5272         }
5273
5274         if (tb[RTA_PREF]) {
5275                 pref = nla_get_u8(tb[RTA_PREF]);
5276                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5277                     pref != ICMPV6_ROUTER_PREF_HIGH)
5278                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5279                 cfg->fc_flags |= RTF_PREF(pref);
5280         }
5281
5282         if (tb[RTA_ENCAP])
5283                 cfg->fc_encap = tb[RTA_ENCAP];
5284
5285         if (tb[RTA_ENCAP_TYPE]) {
5286                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5287
5288                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5289                 if (err < 0)
5290                         goto errout;
5291         }
5292
5293         if (tb[RTA_EXPIRES]) {
5294                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5295
5296                 if (addrconf_finite_timeout(timeout)) {
5297                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5298                         cfg->fc_flags |= RTF_EXPIRES;
5299                 }
5300         }
5301
5302         err = 0;
5303 errout:
5304         return err;
5305 }
5306
5307 struct rt6_nh {
5308         struct fib6_info *fib6_info;
5309         struct fib6_config r_cfg;
5310         struct list_head list;
5311 };
5312
5313 static int ip6_route_info_append(struct list_head *rt6_nh_list,
5314                                  struct fib6_info *rt,
5315                                  struct fib6_config *r_cfg)
5316 {
5317         struct rt6_nh *nh;
5318
5319         list_for_each_entry(nh, rt6_nh_list, list) {
5320                 /* check if fib6_info already exists */
5321                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5322                         return -EEXIST;
5323         }
5324
5325         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5326         if (!nh)
5327                 return -ENOMEM;
5328
5329         nh->fib6_info = rt;
5330         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5331         list_add_tail(&nh->list, rt6_nh_list);
5332
5333         return 0;
5334 }
5335
5336 static void ip6_route_mpath_notify(struct fib6_info *rt,
5337                                    struct fib6_info *rt_last,
5338                                    struct nl_info *info,
5339                                    __u16 nlflags)
5340 {
5341         /* if this is an APPEND route, then rt points to the first route
5342          * inserted and rt_last points to last route inserted. Userspace
5343          * wants a consistent dump of the route which starts at the first
5344          * nexthop. Since sibling routes are always added at the end of
5345          * the list, find the first sibling of the last route appended
5346          */
5347         rcu_read_lock();
5348
5349         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5350                 rt = list_first_or_null_rcu(&rt_last->fib6_siblings,
5351                                             struct fib6_info,
5352                                             fib6_siblings);
5353         }
5354
5355         if (rt)
5356                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5357
5358         rcu_read_unlock();
5359 }
5360
5361 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5362 {
5363         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5364         bool should_notify = false;
5365         struct fib6_info *leaf;
5366         struct fib6_node *fn;
5367
5368         rcu_read_lock();
5369         fn = rcu_dereference(rt->fib6_node);
5370         if (!fn)
5371                 goto out;
5372
5373         leaf = rcu_dereference(fn->leaf);
5374         if (!leaf)
5375                 goto out;
5376
5377         if (rt == leaf ||
5378             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5379              rt6_qualify_for_ecmp(leaf)))
5380                 should_notify = true;
5381 out:
5382         rcu_read_unlock();
5383
5384         return should_notify;
5385 }
5386
5387 static int ip6_route_multipath_add(struct fib6_config *cfg,
5388                                    struct netlink_ext_ack *extack)
5389 {
5390         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5391         struct nl_info *info = &cfg->fc_nlinfo;
5392         struct rt6_nh *nh, *nh_safe;
5393         struct fib6_config r_cfg;
5394         struct rtnexthop *rtnh;
5395         LIST_HEAD(rt6_nh_list);
5396         struct rt6_nh *err_nh;
5397         struct fib6_info *rt;
5398         __u16 nlflags;
5399         int remaining;
5400         int attrlen;
5401         int replace;
5402         int nhn = 0;
5403         int err;
5404
5405         err = fib6_config_validate(cfg, extack);
5406         if (err)
5407                 return err;
5408
5409         replace = (cfg->fc_nlinfo.nlh &&
5410                    (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5411
5412         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5413         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5414                 nlflags |= NLM_F_APPEND;
5415
5416         remaining = cfg->fc_mp_len;
5417         rtnh = (struct rtnexthop *)cfg->fc_mp;
5418
5419         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5420          * fib6_info structs per nexthop
5421          */
5422         while (rtnh_ok(rtnh, remaining)) {
5423                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5424                 if (rtnh->rtnh_ifindex)
5425                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5426
5427                 attrlen = rtnh_attrlen(rtnh);
5428                 if (attrlen > 0) {
5429                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5430
5431                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5432                         if (nla) {
5433                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5434                                 r_cfg.fc_flags |= RTF_GATEWAY;
5435                         }
5436
5437                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5438                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5439                         if (nla)
5440                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5441                 }
5442
5443                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5444                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5445                 if (IS_ERR(rt)) {
5446                         err = PTR_ERR(rt);
5447                         rt = NULL;
5448                         goto cleanup;
5449                 }
5450
5451                 err = ip6_route_info_create_nh(rt, &r_cfg, GFP_KERNEL, extack);
5452                 if (err) {
5453                         rt = NULL;
5454                         goto cleanup;
5455                 }
5456
5457                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5458
5459                 err = ip6_route_info_append(&rt6_nh_list, rt, &r_cfg);
5460                 if (err) {
5461                         fib6_info_release(rt);
5462                         goto cleanup;
5463                 }
5464
5465                 rtnh = rtnh_next(rtnh, &remaining);
5466         }
5467
5468         /* for add and replace send one notification with all nexthops.
5469          * Skip the notification in fib6_add_rt2node and send one with
5470          * the full route when done
5471          */
5472         info->skip_notify = 1;
5473
5474         /* For add and replace, send one notification with all nexthops. For
5475          * append, send one notification with all appended nexthops.
5476          */
5477         info->skip_notify_kernel = 1;
5478
5479         err_nh = NULL;
5480         list_for_each_entry(nh, &rt6_nh_list, list) {
5481                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5482
5483                 if (err) {
5484                         if (replace && nhn)
5485                                 NL_SET_ERR_MSG_MOD(extack,
5486                                                    "multipath route replace failed (check consistency of installed routes)");
5487                         err_nh = nh;
5488                         goto add_errout;
5489                 }
5490                 /* save reference to last route successfully inserted */
5491                 rt_last = nh->fib6_info;
5492
5493                 /* save reference to first route for notification */
5494                 if (!rt_notif)
5495                         rt_notif = nh->fib6_info;
5496
5497                 /* Because each route is added like a single route we remove
5498                  * these flags after the first nexthop: if there is a collision,
5499                  * we have already failed to add the first nexthop:
5500                  * fib6_add_rt2node() has rejected it; when replacing, old
5501                  * nexthops have been replaced by first new, the rest should
5502                  * be added to it.
5503                  */
5504                 if (cfg->fc_nlinfo.nlh) {
5505                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5506                                                              NLM_F_REPLACE);
5507                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5508                 }
5509                 nhn++;
5510         }
5511
5512         /* An in-kernel notification should only be sent in case the new
5513          * multipath route is added as the first route in the node, or if
5514          * it was appended to it. We pass 'rt_notif' since it is the first
5515          * sibling and might allow us to skip some checks in the replace case.
5516          */
5517         if (ip6_route_mpath_should_notify(rt_notif)) {
5518                 enum fib_event_type fib_event;
5519
5520                 if (rt_notif->fib6_nsiblings != nhn - 1)
5521                         fib_event = FIB_EVENT_ENTRY_APPEND;
5522                 else
5523                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5524
5525                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5526                                                           fib_event, rt_notif,
5527                                                           nhn - 1, extack);
5528                 if (err) {
5529                         /* Delete all the siblings that were just added */
5530                         err_nh = NULL;
5531                         goto add_errout;
5532                 }
5533         }
5534
5535         /* success ... tell user about new route */
5536         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5537         goto cleanup;
5538
5539 add_errout:
5540         /* send notification for routes that were added so that
5541          * the delete notifications sent by ip6_route_del are
5542          * coherent
5543          */
5544         if (rt_notif)
5545                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5546
5547         /* Delete routes that were already added */
5548         list_for_each_entry(nh, &rt6_nh_list, list) {
5549                 if (err_nh == nh)
5550                         break;
5551                 ip6_route_del(&nh->r_cfg, extack);
5552         }
5553
5554 cleanup:
5555         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, list) {
5556                 fib6_info_release(nh->fib6_info);
5557                 list_del(&nh->list);
5558                 kfree(nh);
5559         }
5560
5561         return err;
5562 }
5563
5564 static int ip6_route_multipath_del(struct fib6_config *cfg,
5565                                    struct netlink_ext_ack *extack)
5566 {
5567         struct fib6_config r_cfg;
5568         struct rtnexthop *rtnh;
5569         int last_err = 0;
5570         int remaining;
5571         int attrlen;
5572         int err;
5573
5574         remaining = cfg->fc_mp_len;
5575         rtnh = (struct rtnexthop *)cfg->fc_mp;
5576
5577         /* Parse a Multipath Entry */
5578         while (rtnh_ok(rtnh, remaining)) {
5579                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5580                 if (rtnh->rtnh_ifindex)
5581                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5582
5583                 attrlen = rtnh_attrlen(rtnh);
5584                 if (attrlen > 0) {
5585                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5586
5587                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5588                         if (nla) {
5589                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5590                                 r_cfg.fc_flags |= RTF_GATEWAY;
5591                         }
5592                 }
5593
5594                 err = ip6_route_del(&r_cfg, extack);
5595                 if (err)
5596                         last_err = err;
5597
5598                 rtnh = rtnh_next(rtnh, &remaining);
5599         }
5600
5601         return last_err;
5602 }
5603
5604 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5605                               struct netlink_ext_ack *extack)
5606 {
5607         struct fib6_config cfg;
5608         int err;
5609
5610         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5611         if (err < 0)
5612                 return err;
5613
5614         if (cfg.fc_nh_id) {
5615                 rcu_read_lock();
5616                 err = !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id);
5617                 rcu_read_unlock();
5618
5619                 if (err) {
5620                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5621                         return -EINVAL;
5622                 }
5623         }
5624
5625         if (cfg.fc_mp) {
5626                 return ip6_route_multipath_del(&cfg, extack);
5627         } else {
5628                 cfg.fc_delete_all_nh = 1;
5629                 return ip6_route_del(&cfg, extack);
5630         }
5631 }
5632
5633 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5634                               struct netlink_ext_ack *extack)
5635 {
5636         struct fib6_config cfg;
5637         int err;
5638
5639         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5640         if (err < 0)
5641                 return err;
5642
5643         if (cfg.fc_metric == 0)
5644                 cfg.fc_metric = IP6_RT_PRIO_USER;
5645
5646         if (cfg.fc_mp)
5647                 return ip6_route_multipath_add(&cfg, extack);
5648         else
5649                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5650 }
5651
5652 /* add the overhead of this fib6_nh to nexthop_len */
5653 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5654 {
5655         int *nexthop_len = arg;
5656
5657         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5658                      + NLA_ALIGN(sizeof(struct rtnexthop))
5659                      + nla_total_size(16); /* RTA_GATEWAY */
5660
5661         if (nh->fib_nh_lws) {
5662                 /* RTA_ENCAP_TYPE */
5663                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5664                 /* RTA_ENCAP */
5665                 *nexthop_len += nla_total_size(2);
5666         }
5667
5668         return 0;
5669 }
5670
5671 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5672 {
5673         int nexthop_len;
5674
5675         if (f6i->nh) {
5676                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5677                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5678                                          &nexthop_len);
5679         } else {
5680                 struct fib6_nh *nh = f6i->fib6_nh;
5681                 struct fib6_info *sibling;
5682
5683                 nexthop_len = 0;
5684                 if (f6i->fib6_nsiblings) {
5685                         rt6_nh_nlmsg_size(nh, &nexthop_len);
5686
5687                         rcu_read_lock();
5688
5689                         list_for_each_entry_rcu(sibling, &f6i->fib6_siblings,
5690                                                 fib6_siblings) {
5691                                 rt6_nh_nlmsg_size(sibling->fib6_nh, &nexthop_len);
5692                         }
5693
5694                         rcu_read_unlock();
5695                 }
5696                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5697         }
5698
5699         return NLMSG_ALIGN(sizeof(struct rtmsg))
5700                + nla_total_size(16) /* RTA_SRC */
5701                + nla_total_size(16) /* RTA_DST */
5702                + nla_total_size(16) /* RTA_GATEWAY */
5703                + nla_total_size(16) /* RTA_PREFSRC */
5704                + nla_total_size(4) /* RTA_TABLE */
5705                + nla_total_size(4) /* RTA_IIF */
5706                + nla_total_size(4) /* RTA_OIF */
5707                + nla_total_size(4) /* RTA_PRIORITY */
5708                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5709                + nla_total_size(sizeof(struct rta_cacheinfo))
5710                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5711                + nla_total_size(1) /* RTA_PREF */
5712                + nexthop_len;
5713 }
5714
5715 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5716                                  unsigned char *flags)
5717 {
5718         if (nexthop_is_multipath(nh)) {
5719                 struct nlattr *mp;
5720
5721                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5722                 if (!mp)
5723                         goto nla_put_failure;
5724
5725                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5726                         goto nla_put_failure;
5727
5728                 nla_nest_end(skb, mp);
5729         } else {
5730                 struct fib6_nh *fib6_nh;
5731
5732                 fib6_nh = nexthop_fib6_nh(nh);
5733                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5734                                      flags, false) < 0)
5735                         goto nla_put_failure;
5736         }
5737
5738         return 0;
5739
5740 nla_put_failure:
5741         return -EMSGSIZE;
5742 }
5743
5744 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5745                          struct fib6_info *rt, struct dst_entry *dst,
5746                          struct in6_addr *dest, struct in6_addr *src,
5747                          int iif, int type, u32 portid, u32 seq,
5748                          unsigned int flags)
5749 {
5750         struct rt6_info *rt6 = dst_rt6_info(dst);
5751         struct rt6key *rt6_dst, *rt6_src;
5752         u32 *pmetrics, table, rt6_flags;
5753         unsigned char nh_flags = 0;
5754         struct nlmsghdr *nlh;
5755         struct rtmsg *rtm;
5756         long expires = 0;
5757
5758         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5759         if (!nlh)
5760                 return -EMSGSIZE;
5761
5762         if (rt6) {
5763                 rt6_dst = &rt6->rt6i_dst;
5764                 rt6_src = &rt6->rt6i_src;
5765                 rt6_flags = rt6->rt6i_flags;
5766         } else {
5767                 rt6_dst = &rt->fib6_dst;
5768                 rt6_src = &rt->fib6_src;
5769                 rt6_flags = rt->fib6_flags;
5770         }
5771
5772         rtm = nlmsg_data(nlh);
5773         rtm->rtm_family = AF_INET6;
5774         rtm->rtm_dst_len = rt6_dst->plen;
5775         rtm->rtm_src_len = rt6_src->plen;
5776         rtm->rtm_tos = 0;
5777         if (rt->fib6_table)
5778                 table = rt->fib6_table->tb6_id;
5779         else
5780                 table = RT6_TABLE_UNSPEC;
5781         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5782         if (nla_put_u32(skb, RTA_TABLE, table))
5783                 goto nla_put_failure;
5784
5785         rtm->rtm_type = rt->fib6_type;
5786         rtm->rtm_flags = 0;
5787         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5788         rtm->rtm_protocol = rt->fib6_protocol;
5789
5790         if (rt6_flags & RTF_CACHE)
5791                 rtm->rtm_flags |= RTM_F_CLONED;
5792
5793         if (dest) {
5794                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5795                         goto nla_put_failure;
5796                 rtm->rtm_dst_len = 128;
5797         } else if (rtm->rtm_dst_len)
5798                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5799                         goto nla_put_failure;
5800 #ifdef CONFIG_IPV6_SUBTREES
5801         if (src) {
5802                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5803                         goto nla_put_failure;
5804                 rtm->rtm_src_len = 128;
5805         } else if (rtm->rtm_src_len &&
5806                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5807                 goto nla_put_failure;
5808 #endif
5809         if (iif) {
5810 #ifdef CONFIG_IPV6_MROUTE
5811                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5812                         int err = ip6mr_get_route(net, skb, rtm, portid);
5813
5814                         if (err == 0)
5815                                 return 0;
5816                         if (err < 0)
5817                                 goto nla_put_failure;
5818                 } else
5819 #endif
5820                         if (nla_put_u32(skb, RTA_IIF, iif))
5821                                 goto nla_put_failure;
5822         } else if (dest) {
5823                 struct in6_addr saddr_buf;
5824                 if (ip6_route_get_saddr(net, rt, dest, 0, 0, &saddr_buf) == 0 &&
5825                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5826                         goto nla_put_failure;
5827         }
5828
5829         if (rt->fib6_prefsrc.plen) {
5830                 struct in6_addr saddr_buf;
5831                 saddr_buf = rt->fib6_prefsrc.addr;
5832                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5833                         goto nla_put_failure;
5834         }
5835
5836         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5837         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5838                 goto nla_put_failure;
5839
5840         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5841                 goto nla_put_failure;
5842
5843         /* For multipath routes, walk the siblings list and add
5844          * each as a nexthop within RTA_MULTIPATH.
5845          */
5846         if (rt6) {
5847                 if (rt6_flags & RTF_GATEWAY &&
5848                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5849                         goto nla_put_failure;
5850
5851                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5852                         goto nla_put_failure;
5853
5854                 if (dst->lwtstate &&
5855                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5856                         goto nla_put_failure;
5857         } else if (rt->fib6_nsiblings) {
5858                 struct fib6_info *sibling;
5859                 struct nlattr *mp;
5860
5861                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5862                 if (!mp)
5863                         goto nla_put_failure;
5864
5865                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5866                                     rt->fib6_nh->fib_nh_weight, AF_INET6,
5867                                     0) < 0)
5868                         goto nla_put_failure;
5869
5870                 rcu_read_lock();
5871
5872                 list_for_each_entry_rcu(sibling, &rt->fib6_siblings,
5873                                         fib6_siblings) {
5874                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5875                                             sibling->fib6_nh->fib_nh_weight,
5876                                             AF_INET6, 0) < 0) {
5877                                 rcu_read_unlock();
5878
5879                                 goto nla_put_failure;
5880                         }
5881                 }
5882
5883                 rcu_read_unlock();
5884
5885                 nla_nest_end(skb, mp);
5886         } else if (rt->nh) {
5887                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5888                         goto nla_put_failure;
5889
5890                 if (nexthop_is_blackhole(rt->nh))
5891                         rtm->rtm_type = RTN_BLACKHOLE;
5892
5893                 if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5894                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5895                         goto nla_put_failure;
5896
5897                 rtm->rtm_flags |= nh_flags;
5898         } else {
5899                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5900                                      &nh_flags, false) < 0)
5901                         goto nla_put_failure;
5902
5903                 rtm->rtm_flags |= nh_flags;
5904         }
5905
5906         if (rt6_flags & RTF_EXPIRES) {
5907                 expires = dst ? dst->expires : rt->expires;
5908                 expires -= jiffies;
5909         }
5910
5911         if (!dst) {
5912                 if (READ_ONCE(rt->offload))
5913                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5914                 if (READ_ONCE(rt->trap))
5915                         rtm->rtm_flags |= RTM_F_TRAP;
5916                 if (READ_ONCE(rt->offload_failed))
5917                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5918         }
5919
5920         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5921                 goto nla_put_failure;
5922
5923         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5924                 goto nla_put_failure;
5925
5926
5927         nlmsg_end(skb, nlh);
5928         return 0;
5929
5930 nla_put_failure:
5931         nlmsg_cancel(skb, nlh);
5932         return -EMSGSIZE;
5933 }
5934
5935 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5936 {
5937         const struct net_device *dev = arg;
5938
5939         if (nh->fib_nh_dev == dev)
5940                 return 1;
5941
5942         return 0;
5943 }
5944
5945 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5946                                const struct net_device *dev)
5947 {
5948         if (f6i->nh) {
5949                 struct net_device *_dev = (struct net_device *)dev;
5950
5951                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5952                                                   fib6_info_nh_uses_dev,
5953                                                   _dev);
5954         }
5955
5956         if (f6i->fib6_nh->fib_nh_dev == dev)
5957                 return true;
5958
5959         if (f6i->fib6_nsiblings) {
5960                 struct fib6_info *sibling, *next_sibling;
5961
5962                 list_for_each_entry_safe(sibling, next_sibling,
5963                                          &f6i->fib6_siblings, fib6_siblings) {
5964                         if (sibling->fib6_nh->fib_nh_dev == dev)
5965                                 return true;
5966                 }
5967         }
5968
5969         return false;
5970 }
5971
5972 struct fib6_nh_exception_dump_walker {
5973         struct rt6_rtnl_dump_arg *dump;
5974         struct fib6_info *rt;
5975         unsigned int flags;
5976         unsigned int skip;
5977         unsigned int count;
5978 };
5979
5980 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5981 {
5982         struct fib6_nh_exception_dump_walker *w = arg;
5983         struct rt6_rtnl_dump_arg *dump = w->dump;
5984         struct rt6_exception_bucket *bucket;
5985         struct rt6_exception *rt6_ex;
5986         int i, err;
5987
5988         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5989         if (!bucket)
5990                 return 0;
5991
5992         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5993                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5994                         if (w->skip) {
5995                                 w->skip--;
5996                                 continue;
5997                         }
5998
5999                         /* Expiration of entries doesn't bump sernum, insertion
6000                          * does. Removal is triggered by insertion, so we can
6001                          * rely on the fact that if entries change between two
6002                          * partial dumps, this node is scanned again completely,
6003                          * see rt6_insert_exception() and fib6_dump_table().
6004                          *
6005                          * Count expired entries we go through as handled
6006                          * entries that we'll skip next time, in case of partial
6007                          * node dump. Otherwise, if entries expire meanwhile,
6008                          * we'll skip the wrong amount.
6009                          */
6010                         if (rt6_check_expired(rt6_ex->rt6i)) {
6011                                 w->count++;
6012                                 continue;
6013                         }
6014
6015                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
6016                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
6017                                             RTM_NEWROUTE,
6018                                             NETLINK_CB(dump->cb->skb).portid,
6019                                             dump->cb->nlh->nlmsg_seq, w->flags);
6020                         if (err)
6021                                 return err;
6022
6023                         w->count++;
6024                 }
6025                 bucket++;
6026         }
6027
6028         return 0;
6029 }
6030
6031 /* Return -1 if done with node, number of handled routes on partial dump */
6032 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
6033 {
6034         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
6035         struct fib_dump_filter *filter = &arg->filter;
6036         unsigned int flags = NLM_F_MULTI;
6037         struct net *net = arg->net;
6038         int count = 0;
6039
6040         if (rt == net->ipv6.fib6_null_entry)
6041                 return -1;
6042
6043         if ((filter->flags & RTM_F_PREFIX) &&
6044             !(rt->fib6_flags & RTF_PREFIX_RT)) {
6045                 /* success since this is not a prefix route */
6046                 return -1;
6047         }
6048         if (filter->filter_set &&
6049             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
6050              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
6051              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
6052                 return -1;
6053         }
6054
6055         if (filter->filter_set ||
6056             !filter->dump_routes || !filter->dump_exceptions) {
6057                 flags |= NLM_F_DUMP_FILTERED;
6058         }
6059
6060         if (filter->dump_routes) {
6061                 if (skip) {
6062                         skip--;
6063                 } else {
6064                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
6065                                           0, RTM_NEWROUTE,
6066                                           NETLINK_CB(arg->cb->skb).portid,
6067                                           arg->cb->nlh->nlmsg_seq, flags)) {
6068                                 return 0;
6069                         }
6070                         count++;
6071                 }
6072         }
6073
6074         if (filter->dump_exceptions) {
6075                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
6076                                                            .rt = rt,
6077                                                            .flags = flags,
6078                                                            .skip = skip,
6079                                                            .count = 0 };
6080                 int err;
6081
6082                 rcu_read_lock();
6083                 if (rt->nh) {
6084                         err = nexthop_for_each_fib6_nh(rt->nh,
6085                                                        rt6_nh_dump_exceptions,
6086                                                        &w);
6087                 } else {
6088                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
6089                 }
6090                 rcu_read_unlock();
6091
6092                 if (err)
6093                         return count + w.count;
6094         }
6095
6096         return -1;
6097 }
6098
6099 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
6100                                         const struct nlmsghdr *nlh,
6101                                         struct nlattr **tb,
6102                                         struct netlink_ext_ack *extack)
6103 {
6104         struct rtmsg *rtm;
6105         int i, err;
6106
6107         rtm = nlmsg_payload(nlh, sizeof(*rtm));
6108         if (!rtm) {
6109                 NL_SET_ERR_MSG_MOD(extack,
6110                                    "Invalid header for get route request");
6111                 return -EINVAL;
6112         }
6113
6114         if (!netlink_strict_get_check(skb))
6115                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
6116                                               rtm_ipv6_policy, extack);
6117
6118         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
6119             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
6120             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
6121             rtm->rtm_type) {
6122                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
6123                 return -EINVAL;
6124         }
6125         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
6126                 NL_SET_ERR_MSG_MOD(extack,
6127                                    "Invalid flags for get route request");
6128                 return -EINVAL;
6129         }
6130
6131         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
6132                                             rtm_ipv6_policy, extack);
6133         if (err)
6134                 return err;
6135
6136         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
6137             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
6138                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
6139                 return -EINVAL;
6140         }
6141
6142         if (tb[RTA_FLOWLABEL] &&
6143             (nla_get_be32(tb[RTA_FLOWLABEL]) & ~IPV6_FLOWLABEL_MASK)) {
6144                 NL_SET_ERR_MSG_ATTR(extack, tb[RTA_FLOWLABEL],
6145                                     "Invalid flow label");
6146                 return -EINVAL;
6147         }
6148
6149         for (i = 0; i <= RTA_MAX; i++) {
6150                 if (!tb[i])
6151                         continue;
6152
6153                 switch (i) {
6154                 case RTA_SRC:
6155                 case RTA_DST:
6156                 case RTA_IIF:
6157                 case RTA_OIF:
6158                 case RTA_MARK:
6159                 case RTA_UID:
6160                 case RTA_SPORT:
6161                 case RTA_DPORT:
6162                 case RTA_IP_PROTO:
6163                 case RTA_FLOWLABEL:
6164                         break;
6165                 default:
6166                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6167                         return -EINVAL;
6168                 }
6169         }
6170
6171         return 0;
6172 }
6173
6174 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6175                               struct netlink_ext_ack *extack)
6176 {
6177         struct net *net = sock_net(in_skb->sk);
6178         struct nlattr *tb[RTA_MAX+1];
6179         int err, iif = 0, oif = 0;
6180         struct fib6_info *from;
6181         struct dst_entry *dst;
6182         struct rt6_info *rt;
6183         struct sk_buff *skb;
6184         struct rtmsg *rtm;
6185         struct flowi6 fl6 = {};
6186         __be32 flowlabel;
6187         bool fibmatch;
6188
6189         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6190         if (err < 0)
6191                 goto errout;
6192
6193         err = -EINVAL;
6194         rtm = nlmsg_data(nlh);
6195         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6196
6197         if (tb[RTA_SRC]) {
6198                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6199                         goto errout;
6200
6201                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6202         }
6203
6204         if (tb[RTA_DST]) {
6205                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6206                         goto errout;
6207
6208                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6209         }
6210
6211         if (tb[RTA_IIF])
6212                 iif = nla_get_u32(tb[RTA_IIF]);
6213
6214         if (tb[RTA_OIF])
6215                 oif = nla_get_u32(tb[RTA_OIF]);
6216
6217         if (tb[RTA_MARK])
6218                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6219
6220         if (tb[RTA_UID])
6221                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6222                                            nla_get_u32(tb[RTA_UID]));
6223         else
6224                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6225
6226         if (tb[RTA_SPORT])
6227                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6228
6229         if (tb[RTA_DPORT])
6230                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6231
6232         if (tb[RTA_IP_PROTO]) {
6233                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6234                                                   &fl6.flowi6_proto, AF_INET6,
6235                                                   extack);
6236                 if (err)
6237                         goto errout;
6238         }
6239
6240         flowlabel = nla_get_be32_default(tb[RTA_FLOWLABEL], 0);
6241         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, flowlabel);
6242
6243         if (iif) {
6244                 struct net_device *dev;
6245                 int flags = 0;
6246
6247                 rcu_read_lock();
6248
6249                 dev = dev_get_by_index_rcu(net, iif);
6250                 if (!dev) {
6251                         rcu_read_unlock();
6252                         err = -ENODEV;
6253                         goto errout;
6254                 }
6255
6256                 fl6.flowi6_iif = iif;
6257
6258                 if (!ipv6_addr_any(&fl6.saddr))
6259                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6260
6261                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6262
6263                 rcu_read_unlock();
6264         } else {
6265                 fl6.flowi6_oif = oif;
6266
6267                 dst = ip6_route_output(net, NULL, &fl6);
6268         }
6269
6270
6271         rt = dst_rt6_info(dst);
6272         if (rt->dst.error) {
6273                 err = rt->dst.error;
6274                 ip6_rt_put(rt);
6275                 goto errout;
6276         }
6277
6278         if (rt == net->ipv6.ip6_null_entry) {
6279                 err = rt->dst.error;
6280                 ip6_rt_put(rt);
6281                 goto errout;
6282         }
6283
6284         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6285         if (!skb) {
6286                 ip6_rt_put(rt);
6287                 err = -ENOBUFS;
6288                 goto errout;
6289         }
6290
6291         skb_dst_set(skb, &rt->dst);
6292
6293         rcu_read_lock();
6294         from = rcu_dereference(rt->from);
6295         if (from) {
6296                 if (fibmatch)
6297                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6298                                             iif, RTM_NEWROUTE,
6299                                             NETLINK_CB(in_skb).portid,
6300                                             nlh->nlmsg_seq, 0);
6301                 else
6302                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6303                                             &fl6.saddr, iif, RTM_NEWROUTE,
6304                                             NETLINK_CB(in_skb).portid,
6305                                             nlh->nlmsg_seq, 0);
6306         } else {
6307                 err = -ENETUNREACH;
6308         }
6309         rcu_read_unlock();
6310
6311         if (err < 0) {
6312                 kfree_skb(skb);
6313                 goto errout;
6314         }
6315
6316         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6317 errout:
6318         return err;
6319 }
6320
6321 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6322                      unsigned int nlm_flags)
6323 {
6324         struct sk_buff *skb;
6325         struct net *net = info->nl_net;
6326         u32 seq;
6327         int err;
6328
6329         err = -ENOBUFS;
6330         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6331
6332         rcu_read_lock();
6333
6334         skb = nlmsg_new(rt6_nlmsg_size(rt), GFP_ATOMIC);
6335         if (!skb)
6336                 goto errout;
6337
6338         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6339                             event, info->portid, seq, nlm_flags);
6340         if (err < 0) {
6341                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6342                 WARN_ON(err == -EMSGSIZE);
6343                 kfree_skb(skb);
6344                 goto errout;
6345         }
6346
6347         rcu_read_unlock();
6348
6349         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6350                     info->nlh, GFP_ATOMIC);
6351         return;
6352 errout:
6353         rcu_read_unlock();
6354         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6355 }
6356
6357 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6358                     struct nl_info *info)
6359 {
6360         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6361         struct sk_buff *skb;
6362         int err = -ENOBUFS;
6363
6364         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6365         if (!skb)
6366                 goto errout;
6367
6368         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6369                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6370         if (err < 0) {
6371                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6372                 WARN_ON(err == -EMSGSIZE);
6373                 kfree_skb(skb);
6374                 goto errout;
6375         }
6376         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6377                     info->nlh, gfp_any());
6378         return;
6379 errout:
6380         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6381 }
6382
6383 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6384                             bool offload, bool trap, bool offload_failed)
6385 {
6386         struct sk_buff *skb;
6387         int err;
6388
6389         if (READ_ONCE(f6i->offload) == offload &&
6390             READ_ONCE(f6i->trap) == trap &&
6391             READ_ONCE(f6i->offload_failed) == offload_failed)
6392                 return;
6393
6394         WRITE_ONCE(f6i->offload, offload);
6395         WRITE_ONCE(f6i->trap, trap);
6396
6397         /* 2 means send notifications only if offload_failed was changed. */
6398         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6399             READ_ONCE(f6i->offload_failed) == offload_failed)
6400                 return;
6401
6402         WRITE_ONCE(f6i->offload_failed, offload_failed);
6403
6404         if (!rcu_access_pointer(f6i->fib6_node))
6405                 /* The route was removed from the tree, do not send
6406                  * notification.
6407                  */
6408                 return;
6409
6410         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6411                 return;
6412
6413         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6414         if (!skb) {
6415                 err = -ENOBUFS;
6416                 goto errout;
6417         }
6418
6419         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6420                             0, 0);
6421         if (err < 0) {
6422                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6423                 WARN_ON(err == -EMSGSIZE);
6424                 kfree_skb(skb);
6425                 goto errout;
6426         }
6427
6428         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6429         return;
6430
6431 errout:
6432         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6433 }
6434 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6435
6436 static int ip6_route_dev_notify(struct notifier_block *this,
6437                                 unsigned long event, void *ptr)
6438 {
6439         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6440         struct net *net = dev_net(dev);
6441
6442         if (!(dev->flags & IFF_LOOPBACK))
6443                 return NOTIFY_OK;
6444
6445         if (event == NETDEV_REGISTER) {
6446                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6447                 net->ipv6.ip6_null_entry->dst.dev = dev;
6448                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6449 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6450                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6451                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6452                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6453                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6454 #endif
6455          } else if (event == NETDEV_UNREGISTER &&
6456                     dev->reg_state != NETREG_UNREGISTERED) {
6457                 /* NETDEV_UNREGISTER could be fired for multiple times by
6458                  * netdev_wait_allrefs(). Make sure we only call this once.
6459                  */
6460                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6461 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6462                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6463                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6464 #endif
6465         }
6466
6467         return NOTIFY_OK;
6468 }
6469
6470 /*
6471  *      /proc
6472  */
6473
6474 #ifdef CONFIG_PROC_FS
6475 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6476 {
6477         struct net *net = (struct net *)seq->private;
6478         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6479                    net->ipv6.rt6_stats->fib_nodes,
6480                    net->ipv6.rt6_stats->fib_route_nodes,
6481                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6482                    net->ipv6.rt6_stats->fib_rt_entries,
6483                    net->ipv6.rt6_stats->fib_rt_cache,
6484                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6485                    net->ipv6.rt6_stats->fib_discarded_routes);
6486
6487         return 0;
6488 }
6489 #endif  /* CONFIG_PROC_FS */
6490
6491 #ifdef CONFIG_SYSCTL
6492
6493 static int ipv6_sysctl_rtcache_flush(const struct ctl_table *ctl, int write,
6494                               void *buffer, size_t *lenp, loff_t *ppos)
6495 {
6496         struct net *net;
6497         int delay;
6498         int ret;
6499         if (!write)
6500                 return -EINVAL;
6501
6502         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6503         if (ret)
6504                 return ret;
6505
6506         net = (struct net *)ctl->extra1;
6507         delay = net->ipv6.sysctl.flush_delay;
6508         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6509         return 0;
6510 }
6511
6512 static struct ctl_table ipv6_route_table_template[] = {
6513         {
6514                 .procname       =       "max_size",
6515                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6516                 .maxlen         =       sizeof(int),
6517                 .mode           =       0644,
6518                 .proc_handler   =       proc_dointvec,
6519         },
6520         {
6521                 .procname       =       "gc_thresh",
6522                 .data           =       &ip6_dst_ops_template.gc_thresh,
6523                 .maxlen         =       sizeof(int),
6524                 .mode           =       0644,
6525                 .proc_handler   =       proc_dointvec,
6526         },
6527         {
6528                 .procname       =       "flush",
6529                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6530                 .maxlen         =       sizeof(int),
6531                 .mode           =       0200,
6532                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6533         },
6534         {
6535                 .procname       =       "gc_min_interval",
6536                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6537                 .maxlen         =       sizeof(int),
6538                 .mode           =       0644,
6539                 .proc_handler   =       proc_dointvec_jiffies,
6540         },
6541         {
6542                 .procname       =       "gc_timeout",
6543                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6544                 .maxlen         =       sizeof(int),
6545                 .mode           =       0644,
6546                 .proc_handler   =       proc_dointvec_jiffies,
6547         },
6548         {
6549                 .procname       =       "gc_interval",
6550                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6551                 .maxlen         =       sizeof(int),
6552                 .mode           =       0644,
6553                 .proc_handler   =       proc_dointvec_jiffies,
6554         },
6555         {
6556                 .procname       =       "gc_elasticity",
6557                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6558                 .maxlen         =       sizeof(int),
6559                 .mode           =       0644,
6560                 .proc_handler   =       proc_dointvec,
6561         },
6562         {
6563                 .procname       =       "mtu_expires",
6564                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6565                 .maxlen         =       sizeof(int),
6566                 .mode           =       0644,
6567                 .proc_handler   =       proc_dointvec_jiffies,
6568         },
6569         {
6570                 .procname       =       "min_adv_mss",
6571                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6572                 .maxlen         =       sizeof(int),
6573                 .mode           =       0644,
6574                 .proc_handler   =       proc_dointvec,
6575         },
6576         {
6577                 .procname       =       "gc_min_interval_ms",
6578                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6579                 .maxlen         =       sizeof(int),
6580                 .mode           =       0644,
6581                 .proc_handler   =       proc_dointvec_ms_jiffies,
6582         },
6583         {
6584                 .procname       =       "skip_notify_on_dev_down",
6585                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6586                 .maxlen         =       sizeof(u8),
6587                 .mode           =       0644,
6588                 .proc_handler   =       proc_dou8vec_minmax,
6589                 .extra1         =       SYSCTL_ZERO,
6590                 .extra2         =       SYSCTL_ONE,
6591         },
6592 };
6593
6594 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6595 {
6596         struct ctl_table *table;
6597
6598         table = kmemdup(ipv6_route_table_template,
6599                         sizeof(ipv6_route_table_template),
6600                         GFP_KERNEL);
6601
6602         if (table) {
6603                 table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6604                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6605                 table[2].data = &net->ipv6.sysctl.flush_delay;
6606                 table[2].extra1 = net;
6607                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6608                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6609                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6610                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6611                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6612                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6613                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6614                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6615         }
6616
6617         return table;
6618 }
6619
6620 size_t ipv6_route_sysctl_table_size(struct net *net)
6621 {
6622         /* Don't export sysctls to unprivileged users */
6623         if (net->user_ns != &init_user_ns)
6624                 return 1;
6625
6626         return ARRAY_SIZE(ipv6_route_table_template);
6627 }
6628 #endif
6629
6630 static int __net_init ip6_route_net_init(struct net *net)
6631 {
6632         int ret = -ENOMEM;
6633
6634         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6635                sizeof(net->ipv6.ip6_dst_ops));
6636
6637         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6638                 goto out_ip6_dst_ops;
6639
6640         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6641         if (!net->ipv6.fib6_null_entry)
6642                 goto out_ip6_dst_entries;
6643         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6644                sizeof(*net->ipv6.fib6_null_entry));
6645
6646         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6647                                            sizeof(*net->ipv6.ip6_null_entry),
6648                                            GFP_KERNEL);
6649         if (!net->ipv6.ip6_null_entry)
6650                 goto out_fib6_null_entry;
6651         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6652         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6653                          ip6_template_metrics, true);
6654         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->dst.rt_uncached);
6655
6656 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6657         net->ipv6.fib6_has_custom_rules = false;
6658         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6659                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6660                                                GFP_KERNEL);
6661         if (!net->ipv6.ip6_prohibit_entry)
6662                 goto out_ip6_null_entry;
6663         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6664         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6665                          ip6_template_metrics, true);
6666         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->dst.rt_uncached);
6667
6668         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6669                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6670                                                GFP_KERNEL);
6671         if (!net->ipv6.ip6_blk_hole_entry)
6672                 goto out_ip6_prohibit_entry;
6673         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6674         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6675                          ip6_template_metrics, true);
6676         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->dst.rt_uncached);
6677 #ifdef CONFIG_IPV6_SUBTREES
6678         net->ipv6.fib6_routes_require_src = 0;
6679 #endif
6680 #endif
6681
6682         net->ipv6.sysctl.flush_delay = 0;
6683         net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
6684         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6685         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6686         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6687         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6688         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6689         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6690         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6691
6692         atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6693
6694         ret = 0;
6695 out:
6696         return ret;
6697
6698 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6699 out_ip6_prohibit_entry:
6700         kfree(net->ipv6.ip6_prohibit_entry);
6701 out_ip6_null_entry:
6702         kfree(net->ipv6.ip6_null_entry);
6703 #endif
6704 out_fib6_null_entry:
6705         kfree(net->ipv6.fib6_null_entry);
6706 out_ip6_dst_entries:
6707         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6708 out_ip6_dst_ops:
6709         goto out;
6710 }
6711
6712 static void __net_exit ip6_route_net_exit(struct net *net)
6713 {
6714         kfree(net->ipv6.fib6_null_entry);
6715         kfree(net->ipv6.ip6_null_entry);
6716 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6717         kfree(net->ipv6.ip6_prohibit_entry);
6718         kfree(net->ipv6.ip6_blk_hole_entry);
6719 #endif
6720         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6721 }
6722
6723 static int __net_init ip6_route_net_init_late(struct net *net)
6724 {
6725 #ifdef CONFIG_PROC_FS
6726         if (!proc_create_net("ipv6_route", 0, net->proc_net,
6727                              &ipv6_route_seq_ops,
6728                              sizeof(struct ipv6_route_iter)))
6729                 return -ENOMEM;
6730
6731         if (!proc_create_net_single("rt6_stats", 0444, net->proc_net,
6732                                     rt6_stats_seq_show, NULL)) {
6733                 remove_proc_entry("ipv6_route", net->proc_net);
6734                 return -ENOMEM;
6735         }
6736 #endif
6737         return 0;
6738 }
6739
6740 static void __net_exit ip6_route_net_exit_late(struct net *net)
6741 {
6742 #ifdef CONFIG_PROC_FS
6743         remove_proc_entry("ipv6_route", net->proc_net);
6744         remove_proc_entry("rt6_stats", net->proc_net);
6745 #endif
6746 }
6747
6748 static struct pernet_operations ip6_route_net_ops = {
6749         .init = ip6_route_net_init,
6750         .exit = ip6_route_net_exit,
6751 };
6752
6753 static int __net_init ipv6_inetpeer_init(struct net *net)
6754 {
6755         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6756
6757         if (!bp)
6758                 return -ENOMEM;
6759         inet_peer_base_init(bp);
6760         net->ipv6.peers = bp;
6761         return 0;
6762 }
6763
6764 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6765 {
6766         struct inet_peer_base *bp = net->ipv6.peers;
6767
6768         net->ipv6.peers = NULL;
6769         inetpeer_invalidate_tree(bp);
6770         kfree(bp);
6771 }
6772
6773 static struct pernet_operations ipv6_inetpeer_ops = {
6774         .init   =       ipv6_inetpeer_init,
6775         .exit   =       ipv6_inetpeer_exit,
6776 };
6777
6778 static struct pernet_operations ip6_route_net_late_ops = {
6779         .init = ip6_route_net_init_late,
6780         .exit = ip6_route_net_exit_late,
6781 };
6782
6783 static struct notifier_block ip6_route_dev_notifier = {
6784         .notifier_call = ip6_route_dev_notify,
6785         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6786 };
6787
6788 void __init ip6_route_init_special_entries(void)
6789 {
6790         /* Registering of the loopback is done before this portion of code,
6791          * the loopback reference in rt6_info will not be taken, do it
6792          * manually for init_net */
6793         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6794         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6795         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6796   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6797         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6798         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6799         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6800         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6801   #endif
6802 }
6803
6804 #if IS_BUILTIN(CONFIG_IPV6)
6805 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6806 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6807
6808 BTF_ID_LIST(btf_fib6_info_id)
6809 BTF_ID(struct, fib6_info)
6810
6811 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6812         .seq_ops                = &ipv6_route_seq_ops,
6813         .init_seq_private       = bpf_iter_init_seq_net,
6814         .fini_seq_private       = bpf_iter_fini_seq_net,
6815         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6816 };
6817
6818 static struct bpf_iter_reg ipv6_route_reg_info = {
6819         .target                 = "ipv6_route",
6820         .ctx_arg_info_size      = 1,
6821         .ctx_arg_info           = {
6822                 { offsetof(struct bpf_iter__ipv6_route, rt),
6823                   PTR_TO_BTF_ID_OR_NULL },
6824         },
6825         .seq_info               = &ipv6_route_seq_info,
6826 };
6827
6828 static int __init bpf_iter_register(void)
6829 {
6830         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6831         return bpf_iter_reg_target(&ipv6_route_reg_info);
6832 }
6833
6834 static void bpf_iter_unregister(void)
6835 {
6836         bpf_iter_unreg_target(&ipv6_route_reg_info);
6837 }
6838 #endif
6839 #endif
6840
6841 static const struct rtnl_msg_handler ip6_route_rtnl_msg_handlers[] __initconst_or_module = {
6842         {.owner = THIS_MODULE, .protocol = PF_INET6, .msgtype = RTM_NEWROUTE,
6843          .doit = inet6_rtm_newroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
6844         {.owner = THIS_MODULE, .protocol = PF_INET6, .msgtype = RTM_DELROUTE,
6845          .doit = inet6_rtm_delroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
6846         {.owner = THIS_MODULE, .protocol = PF_INET6, .msgtype = RTM_GETROUTE,
6847          .doit = inet6_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED},
6848 };
6849
6850 int __init ip6_route_init(void)
6851 {
6852         int ret;
6853         int cpu;
6854
6855         ret = -ENOMEM;
6856         ip6_dst_ops_template.kmem_cachep =
6857                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6858                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6859         if (!ip6_dst_ops_template.kmem_cachep)
6860                 goto out;
6861
6862         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6863         if (ret)
6864                 goto out_kmem_cache;
6865
6866         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6867         if (ret)
6868                 goto out_dst_entries;
6869
6870         ret = register_pernet_subsys(&ip6_route_net_ops);
6871         if (ret)
6872                 goto out_register_inetpeer;
6873
6874         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6875
6876         ret = fib6_init();
6877         if (ret)
6878                 goto out_register_subsys;
6879
6880         ret = xfrm6_init();
6881         if (ret)
6882                 goto out_fib6_init;
6883
6884         ret = fib6_rules_init();
6885         if (ret)
6886                 goto xfrm6_init;
6887
6888         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6889         if (ret)
6890                 goto fib6_rules_init;
6891
6892         ret = rtnl_register_many(ip6_route_rtnl_msg_handlers);
6893         if (ret < 0)
6894                 goto out_register_late_subsys;
6895
6896         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6897         if (ret)
6898                 goto out_register_late_subsys;
6899
6900 #if IS_BUILTIN(CONFIG_IPV6)
6901 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6902         ret = bpf_iter_register();
6903         if (ret)
6904                 goto out_register_late_subsys;
6905 #endif
6906 #endif
6907
6908         for_each_possible_cpu(cpu) {
6909                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6910
6911                 INIT_LIST_HEAD(&ul->head);
6912                 spin_lock_init(&ul->lock);
6913         }
6914
6915 out:
6916         return ret;
6917
6918 out_register_late_subsys:
6919         rtnl_unregister_all(PF_INET6);
6920         unregister_pernet_subsys(&ip6_route_net_late_ops);
6921 fib6_rules_init:
6922         fib6_rules_cleanup();
6923 xfrm6_init:
6924         xfrm6_fini();
6925 out_fib6_init:
6926         fib6_gc_cleanup();
6927 out_register_subsys:
6928         unregister_pernet_subsys(&ip6_route_net_ops);
6929 out_register_inetpeer:
6930         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6931 out_dst_entries:
6932         dst_entries_destroy(&ip6_dst_blackhole_ops);
6933 out_kmem_cache:
6934         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6935         goto out;
6936 }
6937
6938 void ip6_route_cleanup(void)
6939 {
6940 #if IS_BUILTIN(CONFIG_IPV6)
6941 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6942         bpf_iter_unregister();
6943 #endif
6944 #endif
6945         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6946         unregister_pernet_subsys(&ip6_route_net_late_ops);
6947         fib6_rules_cleanup();
6948         xfrm6_fini();
6949         fib6_gc_cleanup();
6950         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6951         unregister_pernet_subsys(&ip6_route_net_ops);
6952         dst_entries_destroy(&ip6_dst_blackhole_ops);
6953         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6954 }