net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <[email protected]>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <linux/siphash.h>
  45 #include <net/net_namespace.h>
  46 #include <net/snmp.h>
  47 #include <net/ipv6.h>
  48 #include <net/ip6_fib.h>
  49 #include <net/ip6_route.h>
  50 #include <net/ndisc.h>
  51 #include <net/addrconf.h>
  52 #include <net/tcp.h>
  53 #include <linux/rtnetlink.h>
  54 #include <net/dst.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/xfrm.h>
  57 #include <net/netevent.h>
  58 #include <net/netlink.h>
  59 #include <net/rtnh.h>
  60 #include <net/lwtunnel.h>
  61 #include <net/ip_tunnels.h>
  62 #include <net/l3mdev.h>
  63 #include <net/ip.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/btf_ids.h>
  66
  67 #ifdef CONFIG_SYSCTL
  68 #include <linux/sysctl.h>
  69 #endif
  70
  71 static int ip6_rt_type_to_error(u8 fib6_type);
  72
  73 #define CREATE_TRACE_POINTS
  74 #include <trace/events/fib6.h>
  75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  76 #undef CREATE_TRACE_POINTS
  77
  78 enum rt6_nud_state {
  79         RT6_NUD_FAIL_HARD = -3,
  80         RT6_NUD_FAIL_PROBE = -2,
  81         RT6_NUD_FAIL_DO_RR = -1,
  82         RT6_NUD_SUCCEED = 1
  83 };
  84
  85 INDIRECT_CALLABLE_SCOPE
  86 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  87 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  88 INDIRECT_CALLABLE_SCOPE
  89 unsigned int            ip6_mtu(const struct dst_entry *dst);
  90 static 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         struct list_head        quarantine;
 135 };
 136
 137 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 138
 139 void rt6_uncached_list_add(struct rt6_info *rt)
 140 {
 141         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 142
 143         rt->dst.rt_uncached_list = ul;
 144
 145         spin_lock_bh(&ul->lock);
 146         list_add_tail(&rt->dst.rt_uncached, &ul->head);
 147         spin_unlock_bh(&ul->lock);
 148 }
 149
 150 void rt6_uncached_list_del(struct rt6_info *rt)
 151 {
 152         if (!list_empty(&rt->dst.rt_uncached)) {
 153                 struct uncached_list *ul = rt->dst.rt_uncached_list;
 154
 155                 spin_lock_bh(&ul->lock);
 156                 list_del_init(&rt->dst.rt_uncached);
 157                 spin_unlock_bh(&ul->lock);
 158         }
 159 }
 160
 161 static void rt6_uncached_list_flush_dev(struct net_device *dev)
 162 {
 163         int cpu;
 164
 165         for_each_possible_cpu(cpu) {
 166                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 167                 struct rt6_info *rt, *safe;
 168
 169                 if (list_empty(&ul->head))
 170                         continue;
 171
 172                 spin_lock_bh(&ul->lock);
 173                 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) {
 174                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 175                         struct net_device *rt_dev = rt->dst.dev;
 176                         bool handled = false;
 177
 178                         if (rt_idev->dev == dev) {
 179                                 rt->rt6i_idev = in6_dev_get(blackhole_netdev);
 180                                 in6_dev_put(rt_idev);
 181                                 handled = true;
 182                         }
 183
 184                         if (rt_dev == dev) {
 185                                 rt->dst.dev = blackhole_netdev;
 186                                 netdev_ref_replace(rt_dev, blackhole_netdev,
 187                                                    &rt->dst.dev_tracker,
 188                                                    GFP_ATOMIC);
 189                                 handled = true;
 190                         }
 191                         if (handled)
 192                                 list_move(&rt->dst.rt_uncached,
 193                                           &ul->quarantine);
 194                 }
 195                 spin_unlock_bh(&ul->lock);
 196         }
 197 }
 198
 199 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 200                                              struct sk_buff *skb,
 201                                              const void *daddr)
 202 {
 203         if (!ipv6_addr_any(p))
 204                 return (const void *) p;
 205         else if (skb)
 206                 return &ipv6_hdr(skb)->daddr;
 207         return daddr;
 208 }
 209
 210 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 211                                    struct net_device *dev,
 212                                    struct sk_buff *skb,
 213                                    const void *daddr)
 214 {
 215         struct neighbour *n;
 216
 217         daddr = choose_neigh_daddr(gw, skb, daddr);
 218         n = __ipv6_neigh_lookup(dev, daddr);
 219         if (n)
 220                 return n;
 221
 222         n = neigh_create(&nd_tbl, daddr, dev);
 223         return IS_ERR(n) ? NULL : n;
 224 }
 225
 226 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 227                                               struct sk_buff *skb,
 228                                               const void *daddr)
 229 {
 230         const struct rt6_info *rt = dst_rt6_info(dst);
 231
 232         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 233                                 dst->dev, skb, daddr);
 234 }
 235
 236 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 237 {
 238         const struct rt6_info *rt = dst_rt6_info(dst);
 239         struct net_device *dev = dst->dev;
 240
 241         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 242         if (!daddr)
 243                 return;
 244         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 245                 return;
 246         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 247                 return;
 248         __ipv6_confirm_neigh(dev, daddr);
 249 }
 250
 251 static struct dst_ops ip6_dst_ops_template = {
 252         .family                 =       AF_INET6,
 253         .gc                     =       ip6_dst_gc,
 254         .gc_thresh              =       1024,
 255         .check                  =       ip6_dst_check,
 256         .default_advmss         =       ip6_default_advmss,
 257         .mtu                    =       ip6_mtu,
 258         .cow_metrics            =       dst_cow_metrics_generic,
 259         .destroy                =       ip6_dst_destroy,
 260         .ifdown                 =       ip6_dst_ifdown,
 261         .negative_advice        =       ip6_negative_advice,
 262         .link_failure           =       ip6_link_failure,
 263         .update_pmtu            =       ip6_rt_update_pmtu,
 264         .redirect               =       rt6_do_redirect,
 265         .local_out              =       __ip6_local_out,
 266         .neigh_lookup           =       ip6_dst_neigh_lookup,
 267         .confirm_neigh          =       ip6_confirm_neigh,
 268 };
 269
 270 static struct dst_ops ip6_dst_blackhole_ops = {
 271         .family                 = AF_INET6,
 272         .default_advmss         = ip6_default_advmss,
 273         .neigh_lookup           = ip6_dst_neigh_lookup,
 274         .check                  = ip6_dst_check,
 275         .destroy                = ip6_dst_destroy,
 276         .cow_metrics            = dst_cow_metrics_generic,
 277         .update_pmtu            = dst_blackhole_update_pmtu,
 278         .redirect               = dst_blackhole_redirect,
 279         .mtu                    = dst_blackhole_mtu,
 280 };
 281
 282 static const u32 ip6_template_metrics[RTAX_MAX] = {
 283         [RTAX_HOPLIMIT - 1] = 0,
 284 };
 285
 286 static const struct fib6_info fib6_null_entry_template = {
 287         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 288         .fib6_protocol  = RTPROT_KERNEL,
 289         .fib6_metric    = ~(u32)0,
 290         .fib6_ref       = REFCOUNT_INIT(1),
 291         .fib6_type      = RTN_UNREACHABLE,
 292         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 293 };
 294
 295 static const struct rt6_info ip6_null_entry_template = {
 296         .dst = {
 297                 .__rcuref       = RCUREF_INIT(1),
 298                 .__use          = 1,
 299                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 300                 .error          = -ENETUNREACH,
 301                 .input          = ip6_pkt_discard,
 302                 .output         = ip6_pkt_discard_out,
 303         },
 304         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 305 };
 306
 307 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 308
 309 static const struct rt6_info ip6_prohibit_entry_template = {
 310         .dst = {
 311                 .__rcuref       = RCUREF_INIT(1),
 312                 .__use          = 1,
 313                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 314                 .error          = -EACCES,
 315                 .input          = ip6_pkt_prohibit,
 316                 .output         = ip6_pkt_prohibit_out,
 317         },
 318         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 319 };
 320
 321 static const struct rt6_info ip6_blk_hole_entry_template = {
 322         .dst = {
 323                 .__rcuref       = RCUREF_INIT(1),
 324                 .__use          = 1,
 325                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 326                 .error          = -EINVAL,
 327                 .input          = dst_discard,
 328                 .output         = dst_discard_out,
 329         },
 330         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 331 };
 332
 333 #endif
 334
 335 static void rt6_info_init(struct rt6_info *rt)
 336 {
 337         memset_after(rt, 0, dst);
 338 }
 339
 340 /* allocate dst with ip6_dst_ops */
 341 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 342                                int flags)
 343 {
 344         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 345                                         DST_OBSOLETE_FORCE_CHK, flags);
 346
 347         if (rt) {
 348                 rt6_info_init(rt);
 349                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 350         }
 351
 352         return rt;
 353 }
 354 EXPORT_SYMBOL(ip6_dst_alloc);
 355
 356 static void ip6_dst_destroy(struct dst_entry *dst)
 357 {
 358         struct rt6_info *rt = dst_rt6_info(dst);
 359         struct fib6_info *from;
 360         struct inet6_dev *idev;
 361
 362         ip_dst_metrics_put(dst);
 363         rt6_uncached_list_del(rt);
 364
 365         idev = rt->rt6i_idev;
 366         if (idev) {
 367                 rt->rt6i_idev = NULL;
 368                 in6_dev_put(idev);
 369         }
 370
 371         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 372         fib6_info_release(from);
 373 }
 374
 375 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
 376 {
 377         struct rt6_info *rt = dst_rt6_info(dst);
 378         struct inet6_dev *idev = rt->rt6i_idev;
 379
 380         if (idev && idev->dev != blackhole_netdev) {
 381                 struct inet6_dev *blackhole_idev = in6_dev_get(blackhole_netdev);
 382
 383                 if (blackhole_idev) {
 384                         rt->rt6i_idev = blackhole_idev;
 385                         in6_dev_put(idev);
 386                 }
 387         }
 388 }
 389
 390 static bool __rt6_check_expired(const struct rt6_info *rt)
 391 {
 392         if (rt->rt6i_flags & RTF_EXPIRES)
 393                 return time_after(jiffies, rt->dst.expires);
 394         else
 395                 return false;
 396 }
 397
 398 static bool rt6_check_expired(const struct rt6_info *rt)
 399 {
 400         struct fib6_info *from;
 401
 402         from = rcu_dereference(rt->from);
 403
 404         if (rt->rt6i_flags & RTF_EXPIRES) {
 405                 if (time_after(jiffies, rt->dst.expires))
 406                         return true;
 407         } else if (from) {
 408                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 409                         fib6_check_expired(from);
 410         }
 411         return false;
 412 }
 413
 414 void fib6_select_path(const struct net *net, struct fib6_result *res,
 415                       struct flowi6 *fl6, int oif, bool have_oif_match,
 416                       const struct sk_buff *skb, int strict)
 417 {
 418         struct fib6_info *sibling, *next_sibling;
 419         struct fib6_info *match = res->f6i;
 420
 421         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 422                 goto out;
 423
 424         if (match->nh && have_oif_match && res->nh)
 425                 return;
 426
 427         if (skb)
 428                 IP6CB(skb)->flags |= IP6SKB_MULTIPATH;
 429
 430         /* We might have already computed the hash for ICMPv6 errors. In such
 431          * case it will always be non-zero. Otherwise now is the time to do it.
 432          */
 433         if (!fl6->mp_hash &&
 434             (!match->nh || nexthop_is_multipath(match->nh)))
 435                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 436
 437         if (unlikely(match->nh)) {
 438                 nexthop_path_fib6_result(res, fl6->mp_hash);
 439                 return;
 440         }
 441
 442         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 443                 goto out;
 444
 445         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 446                                  fib6_siblings) {
 447                 const struct fib6_nh *nh = sibling->fib6_nh;
 448                 int nh_upper_bound;
 449
 450                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 451                 if (fl6->mp_hash > nh_upper_bound)
 452                         continue;
 453                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 454                         break;
 455                 match = sibling;
 456                 break;
 457         }
 458
 459 out:
 460         res->f6i = match;
 461         res->nh = match->fib6_nh;
 462 }
 463
 464 /*
 465  *      Route lookup. rcu_read_lock() should be held.
 466  */
 467
 468 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 469                                const struct in6_addr *saddr, int oif, int flags)
 470 {
 471         const struct net_device *dev;
 472
 473         if (nh->fib_nh_flags & RTNH_F_DEAD)
 474                 return false;
 475
 476         dev = nh->fib_nh_dev;
 477         if (oif) {
 478                 if (dev->ifindex == oif)
 479                         return true;
 480         } else {
 481                 if (ipv6_chk_addr(net, saddr, dev,
 482                                   flags & RT6_LOOKUP_F_IFACE))
 483                         return true;
 484         }
 485
 486         return false;
 487 }
 488
 489 struct fib6_nh_dm_arg {
 490         struct net              *net;
 491         const struct in6_addr   *saddr;
 492         int                     oif;
 493         int                     flags;
 494         struct fib6_nh          *nh;
 495 };
 496
 497 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 498 {
 499         struct fib6_nh_dm_arg *arg = _arg;
 500
 501         arg->nh = nh;
 502         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 503                                   arg->flags);
 504 }
 505
 506 /* returns fib6_nh from nexthop or NULL */
 507 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 508                                         struct fib6_result *res,
 509                                         const struct in6_addr *saddr,
 510                                         int oif, int flags)
 511 {
 512         struct fib6_nh_dm_arg arg = {
 513                 .net   = net,
 514                 .saddr = saddr,
 515                 .oif   = oif,
 516                 .flags = flags,
 517         };
 518
 519         if (nexthop_is_blackhole(nh))
 520                 return NULL;
 521
 522         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 523                 return arg.nh;
 524
 525         return NULL;
 526 }
 527
 528 static void rt6_device_match(struct net *net, struct fib6_result *res,
 529                              const struct in6_addr *saddr, int oif, int flags)
 530 {
 531         struct fib6_info *f6i = res->f6i;
 532         struct fib6_info *spf6i;
 533         struct fib6_nh *nh;
 534
 535         if (!oif && ipv6_addr_any(saddr)) {
 536                 if (unlikely(f6i->nh)) {
 537                         nh = nexthop_fib6_nh(f6i->nh);
 538                         if (nexthop_is_blackhole(f6i->nh))
 539                                 goto out_blackhole;
 540                 } else {
 541                         nh = f6i->fib6_nh;
 542                 }
 543                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 544                         goto out;
 545         }
 546
 547         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 548                 bool matched = false;
 549
 550                 if (unlikely(spf6i->nh)) {
 551                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 552                                               oif, flags);
 553                         if (nh)
 554                                 matched = true;
 555                 } else {
 556                         nh = spf6i->fib6_nh;
 557                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 558                                 matched = true;
 559                 }
 560                 if (matched) {
 561                         res->f6i = spf6i;
 562                         goto out;
 563                 }
 564         }
 565
 566         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 567                 res->f6i = net->ipv6.fib6_null_entry;
 568                 nh = res->f6i->fib6_nh;
 569                 goto out;
 570         }
 571
 572         if (unlikely(f6i->nh)) {
 573                 nh = nexthop_fib6_nh(f6i->nh);
 574                 if (nexthop_is_blackhole(f6i->nh))
 575                         goto out_blackhole;
 576         } else {
 577                 nh = f6i->fib6_nh;
 578         }
 579
 580         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 581                 res->f6i = net->ipv6.fib6_null_entry;
 582                 nh = res->f6i->fib6_nh;
 583         }
 584 out:
 585         res->nh = nh;
 586         res->fib6_type = res->f6i->fib6_type;
 587         res->fib6_flags = res->f6i->fib6_flags;
 588         return;
 589
 590 out_blackhole:
 591         res->fib6_flags |= RTF_REJECT;
 592         res->fib6_type = RTN_BLACKHOLE;
 593         res->nh = nh;
 594 }
 595
 596 #ifdef CONFIG_IPV6_ROUTER_PREF
 597 struct __rt6_probe_work {
 598         struct work_struct work;
 599         struct in6_addr target;
 600         struct net_device *dev;
 601         netdevice_tracker dev_tracker;
 602 };
 603
 604 static void rt6_probe_deferred(struct work_struct *w)
 605 {
 606         struct in6_addr mcaddr;
 607         struct __rt6_probe_work *work =
 608                 container_of(w, struct __rt6_probe_work, work);
 609
 610         addrconf_addr_solict_mult(&work->target, &mcaddr);
 611         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 612         netdev_put(work->dev, &work->dev_tracker);
 613         kfree(work);
 614 }
 615
 616 static void rt6_probe(struct fib6_nh *fib6_nh)
 617 {
 618         struct __rt6_probe_work *work = NULL;
 619         const struct in6_addr *nh_gw;
 620         unsigned long last_probe;
 621         struct neighbour *neigh;
 622         struct net_device *dev;
 623         struct inet6_dev *idev;
 624
 625         /*
 626          * Okay, this does not seem to be appropriate
 627          * for now, however, we need to check if it
 628          * is really so; aka Router Reachability Probing.
 629          *
 630          * Router Reachability Probe MUST be rate-limited
 631          * to no more than one per minute.
 632          */
 633         if (!fib6_nh->fib_nh_gw_family)
 634                 return;
 635
 636         nh_gw = &fib6_nh->fib_nh_gw6;
 637         dev = fib6_nh->fib_nh_dev;
 638         rcu_read_lock();
 639         last_probe = READ_ONCE(fib6_nh->last_probe);
 640         idev = __in6_dev_get(dev);
 641         if (!idev)
 642                 goto out;
 643         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 644         if (neigh) {
 645                 if (READ_ONCE(neigh->nud_state) & NUD_VALID)
 646                         goto out;
 647
 648                 write_lock_bh(&neigh->lock);
 649                 if (!(neigh->nud_state & NUD_VALID) &&
 650                     time_after(jiffies,
 651                                neigh->updated +
 652                                READ_ONCE(idev->cnf.rtr_probe_interval))) {
 653                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 654                         if (work)
 655                                 __neigh_set_probe_once(neigh);
 656                 }
 657                 write_unlock_bh(&neigh->lock);
 658         } else if (time_after(jiffies, last_probe +
 659                                        READ_ONCE(idev->cnf.rtr_probe_interval))) {
 660                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 661         }
 662
 663         if (!work || cmpxchg(&fib6_nh->last_probe,
 664                              last_probe, jiffies) != last_probe) {
 665                 kfree(work);
 666         } else {
 667                 INIT_WORK(&work->work, rt6_probe_deferred);
 668                 work->target = *nh_gw;
 669                 netdev_hold(dev, &work->dev_tracker, GFP_ATOMIC);
 670                 work->dev = dev;
 671                 schedule_work(&work->work);
 672         }
 673
 674 out:
 675         rcu_read_unlock();
 676 }
 677 #else
 678 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 679 {
 680 }
 681 #endif
 682
 683 /*
 684  * Default Router Selection (RFC 2461 6.3.6)
 685  */
 686 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 687 {
 688         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 689         struct neighbour *neigh;
 690
 691         rcu_read_lock();
 692         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 693                                           &fib6_nh->fib_nh_gw6);
 694         if (neigh) {
 695                 u8 nud_state = READ_ONCE(neigh->nud_state);
 696
 697                 if (nud_state & NUD_VALID)
 698                         ret = RT6_NUD_SUCCEED;
 699 #ifdef CONFIG_IPV6_ROUTER_PREF
 700                 else if (!(nud_state & NUD_FAILED))
 701                         ret = RT6_NUD_SUCCEED;
 702                 else
 703                         ret = RT6_NUD_FAIL_PROBE;
 704 #endif
 705         } else {
 706                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 707                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 708         }
 709         rcu_read_unlock();
 710
 711         return ret;
 712 }
 713
 714 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 715                            int strict)
 716 {
 717         int m = 0;
 718
 719         if (!oif || nh->fib_nh_dev->ifindex == oif)
 720                 m = 2;
 721
 722         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 723                 return RT6_NUD_FAIL_HARD;
 724 #ifdef CONFIG_IPV6_ROUTER_PREF
 725         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 726 #endif
 727         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 728             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 729                 int n = rt6_check_neigh(nh);
 730                 if (n < 0)
 731                         return n;
 732         }
 733         return m;
 734 }
 735
 736 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 737                        int oif, int strict, int *mpri, bool *do_rr)
 738 {
 739         bool match_do_rr = false;
 740         bool rc = false;
 741         int m;
 742
 743         if (nh->fib_nh_flags & RTNH_F_DEAD)
 744                 goto out;
 745
 746         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 747             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 748             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 749                 goto out;
 750
 751         m = rt6_score_route(nh, fib6_flags, oif, strict);
 752         if (m == RT6_NUD_FAIL_DO_RR) {
 753                 match_do_rr = true;
 754                 m = 0; /* lowest valid score */
 755         } else if (m == RT6_NUD_FAIL_HARD) {
 756                 goto out;
 757         }
 758
 759         if (strict & RT6_LOOKUP_F_REACHABLE)
 760                 rt6_probe(nh);
 761
 762         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 763         if (m > *mpri) {
 764                 *do_rr = match_do_rr;
 765                 *mpri = m;
 766                 rc = true;
 767         }
 768 out:
 769         return rc;
 770 }
 771
 772 struct fib6_nh_frl_arg {
 773         u32             flags;
 774         int             oif;
 775         int             strict;
 776         int             *mpri;
 777         bool            *do_rr;
 778         struct fib6_nh  *nh;
 779 };
 780
 781 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 782 {
 783         struct fib6_nh_frl_arg *arg = _arg;
 784
 785         arg->nh = nh;
 786         return find_match(nh, arg->flags, arg->oif, arg->strict,
 787                           arg->mpri, arg->do_rr);
 788 }
 789
 790 static void __find_rr_leaf(struct fib6_info *f6i_start,
 791                            struct fib6_info *nomatch, u32 metric,
 792                            struct fib6_result *res, struct fib6_info **cont,
 793                            int oif, int strict, bool *do_rr, int *mpri)
 794 {
 795         struct fib6_info *f6i;
 796
 797         for (f6i = f6i_start;
 798              f6i && f6i != nomatch;
 799              f6i = rcu_dereference(f6i->fib6_next)) {
 800                 bool matched = false;
 801                 struct fib6_nh *nh;
 802
 803                 if (cont && f6i->fib6_metric != metric) {
 804                         *cont = f6i;
 805                         return;
 806                 }
 807
 808                 if (fib6_check_expired(f6i))
 809                         continue;
 810
 811                 if (unlikely(f6i->nh)) {
 812                         struct fib6_nh_frl_arg arg = {
 813                                 .flags  = f6i->fib6_flags,
 814                                 .oif    = oif,
 815                                 .strict = strict,
 816                                 .mpri   = mpri,
 817                                 .do_rr  = do_rr
 818                         };
 819
 820                         if (nexthop_is_blackhole(f6i->nh)) {
 821                                 res->fib6_flags = RTF_REJECT;
 822                                 res->fib6_type = RTN_BLACKHOLE;
 823                                 res->f6i = f6i;
 824                                 res->nh = nexthop_fib6_nh(f6i->nh);
 825                                 return;
 826                         }
 827                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 828                                                      &arg)) {
 829                                 matched = true;
 830                                 nh = arg.nh;
 831                         }
 832                 } else {
 833                         nh = f6i->fib6_nh;
 834                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 835                                        mpri, do_rr))
 836                                 matched = true;
 837                 }
 838                 if (matched) {
 839                         res->f6i = f6i;
 840                         res->nh = nh;
 841                         res->fib6_flags = f6i->fib6_flags;
 842                         res->fib6_type = f6i->fib6_type;
 843                 }
 844         }
 845 }
 846
 847 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 848                          struct fib6_info *rr_head, int oif, int strict,
 849                          bool *do_rr, struct fib6_result *res)
 850 {
 851         u32 metric = rr_head->fib6_metric;
 852         struct fib6_info *cont = NULL;
 853         int mpri = -1;
 854
 855         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 856                        oif, strict, do_rr, &mpri);
 857
 858         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 859                        oif, strict, do_rr, &mpri);
 860
 861         if (res->f6i || !cont)
 862                 return;
 863
 864         __find_rr_leaf(cont, NULL, metric, res, NULL,
 865                        oif, strict, do_rr, &mpri);
 866 }
 867
 868 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 869                        struct fib6_result *res, int strict)
 870 {
 871         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 872         struct fib6_info *rt0;
 873         bool do_rr = false;
 874         int key_plen;
 875
 876         /* make sure this function or its helpers sets f6i */
 877         res->f6i = NULL;
 878
 879         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 880                 goto out;
 881
 882         rt0 = rcu_dereference(fn->rr_ptr);
 883         if (!rt0)
 884                 rt0 = leaf;
 885
 886         /* Double check to make sure fn is not an intermediate node
 887          * and fn->leaf does not points to its child's leaf
 888          * (This might happen if all routes under fn are deleted from
 889          * the tree and fib6_repair_tree() is called on the node.)
 890          */
 891         key_plen = rt0->fib6_dst.plen;
 892 #ifdef CONFIG_IPV6_SUBTREES
 893         if (rt0->fib6_src.plen)
 894                 key_plen = rt0->fib6_src.plen;
 895 #endif
 896         if (fn->fn_bit != key_plen)
 897                 goto out;
 898
 899         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 900         if (do_rr) {
 901                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 902
 903                 /* no entries matched; do round-robin */
 904                 if (!next || next->fib6_metric != rt0->fib6_metric)
 905                         next = leaf;
 906
 907                 if (next != rt0) {
 908                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 909                         /* make sure next is not being deleted from the tree */
 910                         if (next->fib6_node)
 911                                 rcu_assign_pointer(fn->rr_ptr, next);
 912                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 913                 }
 914         }
 915
 916 out:
 917         if (!res->f6i) {
 918                 res->f6i = net->ipv6.fib6_null_entry;
 919                 res->nh = res->f6i->fib6_nh;
 920                 res->fib6_flags = res->f6i->fib6_flags;
 921                 res->fib6_type = res->f6i->fib6_type;
 922         }
 923 }
 924
 925 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 926 {
 927         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 928                res->nh->fib_nh_gw_family;
 929 }
 930
 931 #ifdef CONFIG_IPV6_ROUTE_INFO
 932 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 933                   const struct in6_addr *gwaddr)
 934 {
 935         struct net *net = dev_net(dev);
 936         struct route_info *rinfo = (struct route_info *) opt;
 937         struct in6_addr prefix_buf, *prefix;
 938         struct fib6_table *table;
 939         unsigned int pref;
 940         unsigned long lifetime;
 941         struct fib6_info *rt;
 942
 943         if (len < sizeof(struct route_info)) {
 944                 return -EINVAL;
 945         }
 946
 947         /* Sanity check for prefix_len and length */
 948         if (rinfo->length > 3) {
 949                 return -EINVAL;
 950         } else if (rinfo->prefix_len > 128) {
 951                 return -EINVAL;
 952         } else if (rinfo->prefix_len > 64) {
 953                 if (rinfo->length < 2) {
 954                         return -EINVAL;
 955                 }
 956         } else if (rinfo->prefix_len > 0) {
 957                 if (rinfo->length < 1) {
 958                         return -EINVAL;
 959                 }
 960         }
 961
 962         pref = rinfo->route_pref;
 963         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 964                 return -EINVAL;
 965
 966         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 967
 968         if (rinfo->length == 3)
 969                 prefix = (struct in6_addr *)rinfo->prefix;
 970         else {
 971                 /* this function is safe */
 972                 ipv6_addr_prefix(&prefix_buf,
 973                                  (struct in6_addr *)rinfo->prefix,
 974                                  rinfo->prefix_len);
 975                 prefix = &prefix_buf;
 976         }
 977
 978         if (rinfo->prefix_len == 0)
 979                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 980         else
 981                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 982                                         gwaddr, dev);
 983
 984         if (rt && !lifetime) {
 985                 ip6_del_rt(net, rt, false);
 986                 rt = NULL;
 987         }
 988
 989         if (!rt && lifetime)
 990                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 991                                         dev, pref);
 992         else if (rt)
 993                 rt->fib6_flags = RTF_ROUTEINFO |
 994                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 995
 996         if (rt) {
 997                 table = rt->fib6_table;
 998                 spin_lock_bh(&table->tb6_lock);
 999
1000                 if (!addrconf_finite_timeout(lifetime)) {
1001                         fib6_clean_expires(rt);
1002                         fib6_remove_gc_list(rt);
1003                 } else {
1004                         fib6_set_expires(rt, jiffies + HZ * lifetime);
1005                         fib6_add_gc_list(rt);
1006                 }
1007
1008                 spin_unlock_bh(&table->tb6_lock);
1009
1010                 fib6_info_release(rt);
1011         }
1012         return 0;
1013 }
1014 #endif
1015
1016 /*
1017  *      Misc support functions
1018  */
1019
1020 /* called with rcu_lock held */
1021 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1022 {
1023         struct net_device *dev = res->nh->fib_nh_dev;
1024
1025         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1026                 /* for copies of local routes, dst->dev needs to be the
1027                  * device if it is a master device, the master device if
1028                  * device is enslaved, and the loopback as the default
1029                  */
1030                 if (netif_is_l3_slave(dev) &&
1031                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1032                         dev = l3mdev_master_dev_rcu(dev);
1033                 else if (!netif_is_l3_master(dev))
1034                         dev = dev_net(dev)->loopback_dev;
1035                 /* last case is netif_is_l3_master(dev) is true in which
1036                  * case we want dev returned to be dev
1037                  */
1038         }
1039
1040         return dev;
1041 }
1042
1043 static const int fib6_prop[RTN_MAX + 1] = {
1044         [RTN_UNSPEC]    = 0,
1045         [RTN_UNICAST]   = 0,
1046         [RTN_LOCAL]     = 0,
1047         [RTN_BROADCAST] = 0,
1048         [RTN_ANYCAST]   = 0,
1049         [RTN_MULTICAST] = 0,
1050         [RTN_BLACKHOLE] = -EINVAL,
1051         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1052         [RTN_PROHIBIT]  = -EACCES,
1053         [RTN_THROW]     = -EAGAIN,
1054         [RTN_NAT]       = -EINVAL,
1055         [RTN_XRESOLVE]  = -EINVAL,
1056 };
1057
1058 static int ip6_rt_type_to_error(u8 fib6_type)
1059 {
1060         return fib6_prop[fib6_type];
1061 }
1062
1063 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1064 {
1065         unsigned short flags = 0;
1066
1067         if (rt->dst_nocount)
1068                 flags |= DST_NOCOUNT;
1069         if (rt->dst_nopolicy)
1070                 flags |= DST_NOPOLICY;
1071
1072         return flags;
1073 }
1074
1075 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1076 {
1077         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1078
1079         switch (fib6_type) {
1080         case RTN_BLACKHOLE:
1081                 rt->dst.output = dst_discard_out;
1082                 rt->dst.input = dst_discard;
1083                 break;
1084         case RTN_PROHIBIT:
1085                 rt->dst.output = ip6_pkt_prohibit_out;
1086                 rt->dst.input = ip6_pkt_prohibit;
1087                 break;
1088         case RTN_THROW:
1089         case RTN_UNREACHABLE:
1090         default:
1091                 rt->dst.output = ip6_pkt_discard_out;
1092                 rt->dst.input = ip6_pkt_discard;
1093                 break;
1094         }
1095 }
1096
1097 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1098 {
1099         struct fib6_info *f6i = res->f6i;
1100
1101         if (res->fib6_flags & RTF_REJECT) {
1102                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1103                 return;
1104         }
1105
1106         rt->dst.error = 0;
1107         rt->dst.output = ip6_output;
1108
1109         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1110                 rt->dst.input = ip6_input;
1111         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1112                 rt->dst.input = ip6_mc_input;
1113         } else {
1114                 rt->dst.input = ip6_forward;
1115         }
1116
1117         if (res->nh->fib_nh_lws) {
1118                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1119                 lwtunnel_set_redirect(&rt->dst);
1120         }
1121
1122         rt->dst.lastuse = jiffies;
1123 }
1124
1125 /* Caller must already hold reference to @from */
1126 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1127 {
1128         rt->rt6i_flags &= ~RTF_EXPIRES;
1129         rcu_assign_pointer(rt->from, from);
1130         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1131 }
1132
1133 /* Caller must already hold reference to f6i in result */
1134 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1135 {
1136         const struct fib6_nh *nh = res->nh;
1137         const struct net_device *dev = nh->fib_nh_dev;
1138         struct fib6_info *f6i = res->f6i;
1139
1140         ip6_rt_init_dst(rt, res);
1141
1142         rt->rt6i_dst = f6i->fib6_dst;
1143         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1144         rt->rt6i_flags = res->fib6_flags;
1145         if (nh->fib_nh_gw_family) {
1146                 rt->rt6i_gateway = nh->fib_nh_gw6;
1147                 rt->rt6i_flags |= RTF_GATEWAY;
1148         }
1149         rt6_set_from(rt, f6i);
1150 #ifdef CONFIG_IPV6_SUBTREES
1151         rt->rt6i_src = f6i->fib6_src;
1152 #endif
1153 }
1154
1155 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1156                                         struct in6_addr *saddr)
1157 {
1158         struct fib6_node *pn, *sn;
1159         while (1) {
1160                 if (fn->fn_flags & RTN_TL_ROOT)
1161                         return NULL;
1162                 pn = rcu_dereference(fn->parent);
1163                 sn = FIB6_SUBTREE(pn);
1164                 if (sn && sn != fn)
1165                         fn = fib6_node_lookup(sn, NULL, saddr);
1166                 else
1167                         fn = pn;
1168                 if (fn->fn_flags & RTN_RTINFO)
1169                         return fn;
1170         }
1171 }
1172
1173 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1174 {
1175         struct rt6_info *rt = *prt;
1176
1177         if (dst_hold_safe(&rt->dst))
1178                 return true;
1179         if (net) {
1180                 rt = net->ipv6.ip6_null_entry;
1181                 dst_hold(&rt->dst);
1182         } else {
1183                 rt = NULL;
1184         }
1185         *prt = rt;
1186         return false;
1187 }
1188
1189 /* called with rcu_lock held */
1190 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1191 {
1192         struct net_device *dev = res->nh->fib_nh_dev;
1193         struct fib6_info *f6i = res->f6i;
1194         unsigned short flags;
1195         struct rt6_info *nrt;
1196
1197         if (!fib6_info_hold_safe(f6i))
1198                 goto fallback;
1199
1200         flags = fib6_info_dst_flags(f6i);
1201         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1202         if (!nrt) {
1203                 fib6_info_release(f6i);
1204                 goto fallback;
1205         }
1206
1207         ip6_rt_copy_init(nrt, res);
1208         return nrt;
1209
1210 fallback:
1211         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1212         dst_hold(&nrt->dst);
1213         return nrt;
1214 }
1215
1216 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1217                                              struct fib6_table *table,
1218                                              struct flowi6 *fl6,
1219                                              const struct sk_buff *skb,
1220                                              int flags)
1221 {
1222         struct fib6_result res = {};
1223         struct fib6_node *fn;
1224         struct rt6_info *rt;
1225
1226         rcu_read_lock();
1227         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1228 restart:
1229         res.f6i = rcu_dereference(fn->leaf);
1230         if (!res.f6i)
1231                 res.f6i = net->ipv6.fib6_null_entry;
1232         else
1233                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1234                                  flags);
1235
1236         if (res.f6i == net->ipv6.fib6_null_entry) {
1237                 fn = fib6_backtrack(fn, &fl6->saddr);
1238                 if (fn)
1239                         goto restart;
1240
1241                 rt = net->ipv6.ip6_null_entry;
1242                 dst_hold(&rt->dst);
1243                 goto out;
1244         } else if (res.fib6_flags & RTF_REJECT) {
1245                 goto do_create;
1246         }
1247
1248         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1249                          fl6->flowi6_oif != 0, skb, flags);
1250
1251         /* Search through exception table */
1252         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1253         if (rt) {
1254                 if (ip6_hold_safe(net, &rt))
1255                         dst_use_noref(&rt->dst, jiffies);
1256         } else {
1257 do_create:
1258                 rt = ip6_create_rt_rcu(&res);
1259         }
1260
1261 out:
1262         trace_fib6_table_lookup(net, &res, table, fl6);
1263
1264         rcu_read_unlock();
1265
1266         return rt;
1267 }
1268
1269 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1270                                    const struct sk_buff *skb, int flags)
1271 {
1272         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1273 }
1274 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1275
1276 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1277                             const struct in6_addr *saddr, int oif,
1278                             const struct sk_buff *skb, int strict)
1279 {
1280         struct flowi6 fl6 = {
1281                 .flowi6_oif = oif,
1282                 .daddr = *daddr,
1283         };
1284         struct dst_entry *dst;
1285         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1286
1287         if (saddr) {
1288                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1289                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1290         }
1291
1292         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1293         if (dst->error == 0)
1294                 return dst_rt6_info(dst);
1295
1296         dst_release(dst);
1297
1298         return NULL;
1299 }
1300 EXPORT_SYMBOL(rt6_lookup);
1301
1302 /* ip6_ins_rt is called with FREE table->tb6_lock.
1303  * It takes new route entry, the addition fails by any reason the
1304  * route is released.
1305  * Caller must hold dst before calling it.
1306  */
1307
1308 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1309                         struct netlink_ext_ack *extack)
1310 {
1311         int err;
1312         struct fib6_table *table;
1313
1314         table = rt->fib6_table;
1315         spin_lock_bh(&table->tb6_lock);
1316         err = fib6_add(&table->tb6_root, rt, info, extack);
1317         spin_unlock_bh(&table->tb6_lock);
1318
1319         return err;
1320 }
1321
1322 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1323 {
1324         struct nl_info info = { .nl_net = net, };
1325
1326         return __ip6_ins_rt(rt, &info, NULL);
1327 }
1328
1329 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1330                                            const struct in6_addr *daddr,
1331                                            const struct in6_addr *saddr)
1332 {
1333         struct fib6_info *f6i = res->f6i;
1334         struct net_device *dev;
1335         struct rt6_info *rt;
1336
1337         /*
1338          *      Clone the route.
1339          */
1340
1341         if (!fib6_info_hold_safe(f6i))
1342                 return NULL;
1343
1344         dev = ip6_rt_get_dev_rcu(res);
1345         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1346         if (!rt) {
1347                 fib6_info_release(f6i);
1348                 return NULL;
1349         }
1350
1351         ip6_rt_copy_init(rt, res);
1352         rt->rt6i_flags |= RTF_CACHE;
1353         rt->rt6i_dst.addr = *daddr;
1354         rt->rt6i_dst.plen = 128;
1355
1356         if (!rt6_is_gw_or_nonexthop(res)) {
1357                 if (f6i->fib6_dst.plen != 128 &&
1358                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1359                         rt->rt6i_flags |= RTF_ANYCAST;
1360 #ifdef CONFIG_IPV6_SUBTREES
1361                 if (rt->rt6i_src.plen && saddr) {
1362                         rt->rt6i_src.addr = *saddr;
1363                         rt->rt6i_src.plen = 128;
1364                 }
1365 #endif
1366         }
1367
1368         return rt;
1369 }
1370
1371 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1372 {
1373         struct fib6_info *f6i = res->f6i;
1374         unsigned short flags = fib6_info_dst_flags(f6i);
1375         struct net_device *dev;
1376         struct rt6_info *pcpu_rt;
1377
1378         if (!fib6_info_hold_safe(f6i))
1379                 return NULL;
1380
1381         rcu_read_lock();
1382         dev = ip6_rt_get_dev_rcu(res);
1383         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1384         rcu_read_unlock();
1385         if (!pcpu_rt) {
1386                 fib6_info_release(f6i);
1387                 return NULL;
1388         }
1389         ip6_rt_copy_init(pcpu_rt, res);
1390         pcpu_rt->rt6i_flags |= RTF_PCPU;
1391
1392         if (f6i->nh)
1393                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1394
1395         return pcpu_rt;
1396 }
1397
1398 static bool rt6_is_valid(const struct rt6_info *rt6)
1399 {
1400         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1401 }
1402
1403 /* It should be called with rcu_read_lock() acquired */
1404 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1405 {
1406         struct rt6_info *pcpu_rt;
1407
1408         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1409
1410         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1411                 struct rt6_info *prev, **p;
1412
1413                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1414                 /* Paired with READ_ONCE() in __fib6_drop_pcpu_from() */
1415                 prev = xchg(p, NULL);
1416                 if (prev) {
1417                         dst_dev_put(&prev->dst);
1418                         dst_release(&prev->dst);
1419                 }
1420
1421                 pcpu_rt = NULL;
1422         }
1423
1424         return pcpu_rt;
1425 }
1426
1427 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1428                                             const struct fib6_result *res)
1429 {
1430         struct rt6_info *pcpu_rt, *prev, **p;
1431
1432         pcpu_rt = ip6_rt_pcpu_alloc(res);
1433         if (!pcpu_rt)
1434                 return NULL;
1435
1436         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1437         prev = cmpxchg(p, NULL, pcpu_rt);
1438         BUG_ON(prev);
1439
1440         if (res->f6i->fib6_destroying) {
1441                 struct fib6_info *from;
1442
1443                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1444                 fib6_info_release(from);
1445         }
1446
1447         return pcpu_rt;
1448 }
1449
1450 /* exception hash table implementation
1451  */
1452 static DEFINE_SPINLOCK(rt6_exception_lock);
1453
1454 /* Remove rt6_ex from hash table and free the memory
1455  * Caller must hold rt6_exception_lock
1456  */
1457 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1458                                  struct rt6_exception *rt6_ex)
1459 {
1460         struct fib6_info *from;
1461         struct net *net;
1462
1463         if (!bucket || !rt6_ex)
1464                 return;
1465
1466         net = dev_net(rt6_ex->rt6i->dst.dev);
1467         net->ipv6.rt6_stats->fib_rt_cache--;
1468
1469         /* purge completely the exception to allow releasing the held resources:
1470          * some [sk] cache may keep the dst around for unlimited time
1471          */
1472         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1473         fib6_info_release(from);
1474         dst_dev_put(&rt6_ex->rt6i->dst);
1475
1476         hlist_del_rcu(&rt6_ex->hlist);
1477         dst_release(&rt6_ex->rt6i->dst);
1478         kfree_rcu(rt6_ex, rcu);
1479         WARN_ON_ONCE(!bucket->depth);
1480         bucket->depth--;
1481 }
1482
1483 /* Remove oldest rt6_ex in bucket and free the memory
1484  * Caller must hold rt6_exception_lock
1485  */
1486 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1487 {
1488         struct rt6_exception *rt6_ex, *oldest = NULL;
1489
1490         if (!bucket)
1491                 return;
1492
1493         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1494                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1495                         oldest = rt6_ex;
1496         }
1497         rt6_remove_exception(bucket, oldest);
1498 }
1499
1500 static u32 rt6_exception_hash(const struct in6_addr *dst,
1501                               const struct in6_addr *src)
1502 {
1503         static siphash_aligned_key_t rt6_exception_key;
1504         struct {
1505                 struct in6_addr dst;
1506                 struct in6_addr src;
1507         } __aligned(SIPHASH_ALIGNMENT) combined = {
1508                 .dst = *dst,
1509         };
1510         u64 val;
1511
1512         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1513
1514 #ifdef CONFIG_IPV6_SUBTREES
1515         if (src)
1516                 combined.src = *src;
1517 #endif
1518         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1519
1520         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1521 }
1522
1523 /* Helper function to find the cached rt in the hash table
1524  * and update bucket pointer to point to the bucket for this
1525  * (daddr, saddr) pair
1526  * Caller must hold rt6_exception_lock
1527  */
1528 static struct rt6_exception *
1529 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1530                               const struct in6_addr *daddr,
1531                               const struct in6_addr *saddr)
1532 {
1533         struct rt6_exception *rt6_ex;
1534         u32 hval;
1535
1536         if (!(*bucket) || !daddr)
1537                 return NULL;
1538
1539         hval = rt6_exception_hash(daddr, saddr);
1540         *bucket += hval;
1541
1542         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1543                 struct rt6_info *rt6 = rt6_ex->rt6i;
1544                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1545
1546 #ifdef CONFIG_IPV6_SUBTREES
1547                 if (matched && saddr)
1548                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1549 #endif
1550                 if (matched)
1551                         return rt6_ex;
1552         }
1553         return NULL;
1554 }
1555
1556 /* Helper function to find the cached rt in the hash table
1557  * and update bucket pointer to point to the bucket for this
1558  * (daddr, saddr) pair
1559  * Caller must hold rcu_read_lock()
1560  */
1561 static struct rt6_exception *
1562 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1563                          const struct in6_addr *daddr,
1564                          const struct in6_addr *saddr)
1565 {
1566         struct rt6_exception *rt6_ex;
1567         u32 hval;
1568
1569         WARN_ON_ONCE(!rcu_read_lock_held());
1570
1571         if (!(*bucket) || !daddr)
1572                 return NULL;
1573
1574         hval = rt6_exception_hash(daddr, saddr);
1575         *bucket += hval;
1576
1577         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1578                 struct rt6_info *rt6 = rt6_ex->rt6i;
1579                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1580
1581 #ifdef CONFIG_IPV6_SUBTREES
1582                 if (matched && saddr)
1583                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1584 #endif
1585                 if (matched)
1586                         return rt6_ex;
1587         }
1588         return NULL;
1589 }
1590
1591 static unsigned int fib6_mtu(const struct fib6_result *res)
1592 {
1593         const struct fib6_nh *nh = res->nh;
1594         unsigned int mtu;
1595
1596         if (res->f6i->fib6_pmtu) {
1597                 mtu = res->f6i->fib6_pmtu;
1598         } else {
1599                 struct net_device *dev = nh->fib_nh_dev;
1600                 struct inet6_dev *idev;
1601
1602                 rcu_read_lock();
1603                 idev = __in6_dev_get(dev);
1604                 mtu = READ_ONCE(idev->cnf.mtu6);
1605                 rcu_read_unlock();
1606         }
1607
1608         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1609
1610         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1611 }
1612
1613 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1614
1615 /* used when the flushed bit is not relevant, only access to the bucket
1616  * (ie., all bucket users except rt6_insert_exception);
1617  *
1618  * called under rcu lock; sometimes called with rt6_exception_lock held
1619  */
1620 static
1621 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1622                                                        spinlock_t *lock)
1623 {
1624         struct rt6_exception_bucket *bucket;
1625
1626         if (lock)
1627                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1628                                                    lockdep_is_held(lock));
1629         else
1630                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1631
1632         /* remove bucket flushed bit if set */
1633         if (bucket) {
1634                 unsigned long p = (unsigned long)bucket;
1635
1636                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1637                 bucket = (struct rt6_exception_bucket *)p;
1638         }
1639
1640         return bucket;
1641 }
1642
1643 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1644 {
1645         unsigned long p = (unsigned long)bucket;
1646
1647         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1648 }
1649
1650 /* called with rt6_exception_lock held */
1651 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1652                                               spinlock_t *lock)
1653 {
1654         struct rt6_exception_bucket *bucket;
1655         unsigned long p;
1656
1657         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1658                                            lockdep_is_held(lock));
1659
1660         p = (unsigned long)bucket;
1661         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1662         bucket = (struct rt6_exception_bucket *)p;
1663         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1664 }
1665
1666 static int rt6_insert_exception(struct rt6_info *nrt,
1667                                 const struct fib6_result *res)
1668 {
1669         struct net *net = dev_net(nrt->dst.dev);
1670         struct rt6_exception_bucket *bucket;
1671         struct fib6_info *f6i = res->f6i;
1672         struct in6_addr *src_key = NULL;
1673         struct rt6_exception *rt6_ex;
1674         struct fib6_nh *nh = res->nh;
1675         int max_depth;
1676         int err = 0;
1677
1678         spin_lock_bh(&rt6_exception_lock);
1679
1680         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1681                                           lockdep_is_held(&rt6_exception_lock));
1682         if (!bucket) {
1683                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1684                                  GFP_ATOMIC);
1685                 if (!bucket) {
1686                         err = -ENOMEM;
1687                         goto out;
1688                 }
1689                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1690         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1691                 err = -EINVAL;
1692                 goto out;
1693         }
1694
1695 #ifdef CONFIG_IPV6_SUBTREES
1696         /* fib6_src.plen != 0 indicates f6i is in subtree
1697          * and exception table is indexed by a hash of
1698          * both fib6_dst and fib6_src.
1699          * Otherwise, the exception table is indexed by
1700          * a hash of only fib6_dst.
1701          */
1702         if (f6i->fib6_src.plen)
1703                 src_key = &nrt->rt6i_src.addr;
1704 #endif
1705         /* rt6_mtu_change() might lower mtu on f6i.
1706          * Only insert this exception route if its mtu
1707          * is less than f6i's mtu value.
1708          */
1709         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1710                 err = -EINVAL;
1711                 goto out;
1712         }
1713
1714         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1715                                                src_key);
1716         if (rt6_ex)
1717                 rt6_remove_exception(bucket, rt6_ex);
1718
1719         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1720         if (!rt6_ex) {
1721                 err = -ENOMEM;
1722                 goto out;
1723         }
1724         rt6_ex->rt6i = nrt;
1725         rt6_ex->stamp = jiffies;
1726         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1727         bucket->depth++;
1728         net->ipv6.rt6_stats->fib_rt_cache++;
1729
1730         /* Randomize max depth to avoid some side channels attacks. */
1731         max_depth = FIB6_MAX_DEPTH + get_random_u32_below(FIB6_MAX_DEPTH);
1732         while (bucket->depth > max_depth)
1733                 rt6_exception_remove_oldest(bucket);
1734
1735 out:
1736         spin_unlock_bh(&rt6_exception_lock);
1737
1738         /* Update fn->fn_sernum to invalidate all cached dst */
1739         if (!err) {
1740                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1741                 fib6_update_sernum(net, f6i);
1742                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1743                 fib6_force_start_gc(net);
1744         }
1745
1746         return err;
1747 }
1748
1749 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1750 {
1751         struct rt6_exception_bucket *bucket;
1752         struct rt6_exception *rt6_ex;
1753         struct hlist_node *tmp;
1754         int i;
1755
1756         spin_lock_bh(&rt6_exception_lock);
1757
1758         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1759         if (!bucket)
1760                 goto out;
1761
1762         /* Prevent rt6_insert_exception() to recreate the bucket list */
1763         if (!from)
1764                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1765
1766         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1767                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1768                         if (!from ||
1769                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1770                                 rt6_remove_exception(bucket, rt6_ex);
1771                 }
1772                 WARN_ON_ONCE(!from && bucket->depth);
1773                 bucket++;
1774         }
1775 out:
1776         spin_unlock_bh(&rt6_exception_lock);
1777 }
1778
1779 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1780 {
1781         struct fib6_info *f6i = arg;
1782
1783         fib6_nh_flush_exceptions(nh, f6i);
1784
1785         return 0;
1786 }
1787
1788 void rt6_flush_exceptions(struct fib6_info *f6i)
1789 {
1790         if (f6i->nh)
1791                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1792                                          f6i);
1793         else
1794                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1795 }
1796
1797 /* Find cached rt in the hash table inside passed in rt
1798  * Caller has to hold rcu_read_lock()
1799  */
1800 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1801                                            const struct in6_addr *daddr,
1802                                            const struct in6_addr *saddr)
1803 {
1804         const struct in6_addr *src_key = NULL;
1805         struct rt6_exception_bucket *bucket;
1806         struct rt6_exception *rt6_ex;
1807         struct rt6_info *ret = NULL;
1808
1809 #ifdef CONFIG_IPV6_SUBTREES
1810         /* fib6i_src.plen != 0 indicates f6i is in subtree
1811          * and exception table is indexed by a hash of
1812          * both fib6_dst and fib6_src.
1813          * However, the src addr used to create the hash
1814          * might not be exactly the passed in saddr which
1815          * is a /128 addr from the flow.
1816          * So we need to use f6i->fib6_src to redo lookup
1817          * if the passed in saddr does not find anything.
1818          * (See the logic in ip6_rt_cache_alloc() on how
1819          * rt->rt6i_src is updated.)
1820          */
1821         if (res->f6i->fib6_src.plen)
1822                 src_key = saddr;
1823 find_ex:
1824 #endif
1825         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1826         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1827
1828         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1829                 ret = rt6_ex->rt6i;
1830
1831 #ifdef CONFIG_IPV6_SUBTREES
1832         /* Use fib6_src as src_key and redo lookup */
1833         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1834                 src_key = &res->f6i->fib6_src.addr;
1835                 goto find_ex;
1836         }
1837 #endif
1838
1839         return ret;
1840 }
1841
1842 /* Remove the passed in cached rt from the hash table that contains it */
1843 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1844                                     const struct rt6_info *rt)
1845 {
1846         const struct in6_addr *src_key = NULL;
1847         struct rt6_exception_bucket *bucket;
1848         struct rt6_exception *rt6_ex;
1849         int err;
1850
1851         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1852                 return -ENOENT;
1853
1854         spin_lock_bh(&rt6_exception_lock);
1855         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1856
1857 #ifdef CONFIG_IPV6_SUBTREES
1858         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1859          * and exception table is indexed by a hash of
1860          * both rt6i_dst and rt6i_src.
1861          * Otherwise, the exception table is indexed by
1862          * a hash of only rt6i_dst.
1863          */
1864         if (plen)
1865                 src_key = &rt->rt6i_src.addr;
1866 #endif
1867         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1868                                                &rt->rt6i_dst.addr,
1869                                                src_key);
1870         if (rt6_ex) {
1871                 rt6_remove_exception(bucket, rt6_ex);
1872                 err = 0;
1873         } else {
1874                 err = -ENOENT;
1875         }
1876
1877         spin_unlock_bh(&rt6_exception_lock);
1878         return err;
1879 }
1880
1881 struct fib6_nh_excptn_arg {
1882         struct rt6_info *rt;
1883         int             plen;
1884 };
1885
1886 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1887 {
1888         struct fib6_nh_excptn_arg *arg = _arg;
1889         int err;
1890
1891         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1892         if (err == 0)
1893                 return 1;
1894
1895         return 0;
1896 }
1897
1898 static int rt6_remove_exception_rt(struct rt6_info *rt)
1899 {
1900         struct fib6_info *from;
1901
1902         from = rcu_dereference(rt->from);
1903         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1904                 return -EINVAL;
1905
1906         if (from->nh) {
1907                 struct fib6_nh_excptn_arg arg = {
1908                         .rt = rt,
1909                         .plen = from->fib6_src.plen
1910                 };
1911                 int rc;
1912
1913                 /* rc = 1 means an entry was found */
1914                 rc = nexthop_for_each_fib6_nh(from->nh,
1915                                               rt6_nh_remove_exception_rt,
1916                                               &arg);
1917                 return rc ? 0 : -ENOENT;
1918         }
1919
1920         return fib6_nh_remove_exception(from->fib6_nh,
1921                                         from->fib6_src.plen, rt);
1922 }
1923
1924 /* Find rt6_ex which contains the passed in rt cache and
1925  * refresh its stamp
1926  */
1927 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1928                                      const struct rt6_info *rt)
1929 {
1930         const struct in6_addr *src_key = NULL;
1931         struct rt6_exception_bucket *bucket;
1932         struct rt6_exception *rt6_ex;
1933
1934         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1935 #ifdef CONFIG_IPV6_SUBTREES
1936         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1937          * and exception table is indexed by a hash of
1938          * both rt6i_dst and rt6i_src.
1939          * Otherwise, the exception table is indexed by
1940          * a hash of only rt6i_dst.
1941          */
1942         if (plen)
1943                 src_key = &rt->rt6i_src.addr;
1944 #endif
1945         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1946         if (rt6_ex)
1947                 rt6_ex->stamp = jiffies;
1948 }
1949
1950 struct fib6_nh_match_arg {
1951         const struct net_device *dev;
1952         const struct in6_addr   *gw;
1953         struct fib6_nh          *match;
1954 };
1955
1956 /* determine if fib6_nh has given device and gateway */
1957 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1958 {
1959         struct fib6_nh_match_arg *arg = _arg;
1960
1961         if (arg->dev != nh->fib_nh_dev ||
1962             (arg->gw && !nh->fib_nh_gw_family) ||
1963             (!arg->gw && nh->fib_nh_gw_family) ||
1964             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1965                 return 0;
1966
1967         arg->match = nh;
1968
1969         /* found a match, break the loop */
1970         return 1;
1971 }
1972
1973 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1974 {
1975         struct fib6_info *from;
1976         struct fib6_nh *fib6_nh;
1977
1978         rcu_read_lock();
1979
1980         from = rcu_dereference(rt->from);
1981         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1982                 goto unlock;
1983
1984         if (from->nh) {
1985                 struct fib6_nh_match_arg arg = {
1986                         .dev = rt->dst.dev,
1987                         .gw = &rt->rt6i_gateway,
1988                 };
1989
1990                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1991
1992                 if (!arg.match)
1993                         goto unlock;
1994                 fib6_nh = arg.match;
1995         } else {
1996                 fib6_nh = from->fib6_nh;
1997         }
1998         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1999 unlock:
2000         rcu_read_unlock();
2001 }
2002
2003 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
2004                                          struct rt6_info *rt, int mtu)
2005 {
2006         /* If the new MTU is lower than the route PMTU, this new MTU will be the
2007          * lowest MTU in the path: always allow updating the route PMTU to
2008          * reflect PMTU decreases.
2009          *
2010          * If the new MTU is higher, and the route PMTU is equal to the local
2011          * MTU, this means the old MTU is the lowest in the path, so allow
2012          * updating it: if other nodes now have lower MTUs, PMTU discovery will
2013          * handle this.
2014          */
2015
2016         if (dst_mtu(&rt->dst) >= mtu)
2017                 return true;
2018
2019         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2020                 return true;
2021
2022         return false;
2023 }
2024
2025 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2026                                        const struct fib6_nh *nh, int mtu)
2027 {
2028         struct rt6_exception_bucket *bucket;
2029         struct rt6_exception *rt6_ex;
2030         int i;
2031
2032         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2033         if (!bucket)
2034                 return;
2035
2036         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2037                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2038                         struct rt6_info *entry = rt6_ex->rt6i;
2039
2040                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2041                          * route), the metrics of its rt->from have already
2042                          * been updated.
2043                          */
2044                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2045                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2046                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2047                 }
2048                 bucket++;
2049         }
2050 }
2051
2052 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2053
2054 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2055                                             const struct in6_addr *gateway)
2056 {
2057         struct rt6_exception_bucket *bucket;
2058         struct rt6_exception *rt6_ex;
2059         struct hlist_node *tmp;
2060         int i;
2061
2062         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2063                 return;
2064
2065         spin_lock_bh(&rt6_exception_lock);
2066         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2067         if (bucket) {
2068                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2069                         hlist_for_each_entry_safe(rt6_ex, tmp,
2070                                                   &bucket->chain, hlist) {
2071                                 struct rt6_info *entry = rt6_ex->rt6i;
2072
2073                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2074                                     RTF_CACHE_GATEWAY &&
2075                                     ipv6_addr_equal(gateway,
2076                                                     &entry->rt6i_gateway)) {
2077                                         rt6_remove_exception(bucket, rt6_ex);
2078                                 }
2079                         }
2080                         bucket++;
2081                 }
2082         }
2083
2084         spin_unlock_bh(&rt6_exception_lock);
2085 }
2086
2087 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2088                                       struct rt6_exception *rt6_ex,
2089                                       struct fib6_gc_args *gc_args,
2090                                       unsigned long now)
2091 {
2092         struct rt6_info *rt = rt6_ex->rt6i;
2093
2094         /* we are pruning and obsoleting aged-out and non gateway exceptions
2095          * even if others have still references to them, so that on next
2096          * dst_check() such references can be dropped.
2097          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2098          * expired, independently from their aging, as per RFC 8201 section 4
2099          */
2100         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2101                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2102                         pr_debug("aging clone %p\n", rt);
2103                         rt6_remove_exception(bucket, rt6_ex);
2104                         return;
2105                 }
2106         } else if (time_after(jiffies, rt->dst.expires)) {
2107                 pr_debug("purging expired route %p\n", rt);
2108                 rt6_remove_exception(bucket, rt6_ex);
2109                 return;
2110         }
2111
2112         if (rt->rt6i_flags & RTF_GATEWAY) {
2113                 struct neighbour *neigh;
2114
2115                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2116
2117                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2118                         pr_debug("purging route %p via non-router but gateway\n",
2119                                  rt);
2120                         rt6_remove_exception(bucket, rt6_ex);
2121                         return;
2122                 }
2123         }
2124
2125         gc_args->more++;
2126 }
2127
2128 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2129                                    struct fib6_gc_args *gc_args,
2130                                    unsigned long now)
2131 {
2132         struct rt6_exception_bucket *bucket;
2133         struct rt6_exception *rt6_ex;
2134         struct hlist_node *tmp;
2135         int i;
2136
2137         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2138                 return;
2139
2140         rcu_read_lock_bh();
2141         spin_lock(&rt6_exception_lock);
2142         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2143         if (bucket) {
2144                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2145                         hlist_for_each_entry_safe(rt6_ex, tmp,
2146                                                   &bucket->chain, hlist) {
2147                                 rt6_age_examine_exception(bucket, rt6_ex,
2148                                                           gc_args, now);
2149                         }
2150                         bucket++;
2151                 }
2152         }
2153         spin_unlock(&rt6_exception_lock);
2154         rcu_read_unlock_bh();
2155 }
2156
2157 struct fib6_nh_age_excptn_arg {
2158         struct fib6_gc_args     *gc_args;
2159         unsigned long           now;
2160 };
2161
2162 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2163 {
2164         struct fib6_nh_age_excptn_arg *arg = _arg;
2165
2166         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2167         return 0;
2168 }
2169
2170 void rt6_age_exceptions(struct fib6_info *f6i,
2171                         struct fib6_gc_args *gc_args,
2172                         unsigned long now)
2173 {
2174         if (f6i->nh) {
2175                 struct fib6_nh_age_excptn_arg arg = {
2176                         .gc_args = gc_args,
2177                         .now = now
2178                 };
2179
2180                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2181                                          &arg);
2182         } else {
2183                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2184         }
2185 }
2186
2187 /* must be called with rcu lock held */
2188 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2189                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2190 {
2191         struct fib6_node *fn, *saved_fn;
2192
2193         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2194         saved_fn = fn;
2195
2196 redo_rt6_select:
2197         rt6_select(net, fn, oif, res, strict);
2198         if (res->f6i == net->ipv6.fib6_null_entry) {
2199                 fn = fib6_backtrack(fn, &fl6->saddr);
2200                 if (fn)
2201                         goto redo_rt6_select;
2202                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2203                         /* also consider unreachable route */
2204                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2205                         fn = saved_fn;
2206                         goto redo_rt6_select;
2207                 }
2208         }
2209
2210         trace_fib6_table_lookup(net, res, table, fl6);
2211
2212         return 0;
2213 }
2214
2215 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2216                                int oif, struct flowi6 *fl6,
2217                                const struct sk_buff *skb, int flags)
2218 {
2219         struct fib6_result res = {};
2220         struct rt6_info *rt = NULL;
2221         int strict = 0;
2222
2223         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2224                      !rcu_read_lock_held());
2225
2226         strict |= flags & RT6_LOOKUP_F_IFACE;
2227         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2228         if (READ_ONCE(net->ipv6.devconf_all->forwarding) == 0)
2229                 strict |= RT6_LOOKUP_F_REACHABLE;
2230
2231         rcu_read_lock();
2232
2233         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2234         if (res.f6i == net->ipv6.fib6_null_entry)
2235                 goto out;
2236
2237         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2238
2239         /*Search through exception table */
2240         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2241         if (rt) {
2242                 goto out;
2243         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2244                             !res.nh->fib_nh_gw_family)) {
2245                 /* Create a RTF_CACHE clone which will not be
2246                  * owned by the fib6 tree.  It is for the special case where
2247                  * the daddr in the skb during the neighbor look-up is different
2248                  * from the fl6->daddr used to look-up route here.
2249                  */
2250                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2251
2252                 if (rt) {
2253                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2254                          * As rt6_uncached_list_add() does not consume refcnt,
2255                          * this refcnt is always returned to the caller even
2256                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2257                          */
2258                         rt6_uncached_list_add(rt);
2259                         rcu_read_unlock();
2260
2261                         return rt;
2262                 }
2263         } else {
2264                 /* Get a percpu copy */
2265                 local_bh_disable();
2266                 rt = rt6_get_pcpu_route(&res);
2267
2268                 if (!rt)
2269                         rt = rt6_make_pcpu_route(net, &res);
2270
2271                 local_bh_enable();
2272         }
2273 out:
2274         if (!rt)
2275                 rt = net->ipv6.ip6_null_entry;
2276         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2277                 ip6_hold_safe(net, &rt);
2278         rcu_read_unlock();
2279
2280         return rt;
2281 }
2282 EXPORT_SYMBOL_GPL(ip6_pol_route);
2283
2284 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2285                                             struct fib6_table *table,
2286                                             struct flowi6 *fl6,
2287                                             const struct sk_buff *skb,
2288                                             int flags)
2289 {
2290         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2291 }
2292
2293 struct dst_entry *ip6_route_input_lookup(struct net *net,
2294                                          struct net_device *dev,
2295                                          struct flowi6 *fl6,
2296                                          const struct sk_buff *skb,
2297                                          int flags)
2298 {
2299         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2300                 flags |= RT6_LOOKUP_F_IFACE;
2301
2302         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2303 }
2304 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2305
2306 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2307                                   struct flow_keys *keys,
2308                                   struct flow_keys *flkeys)
2309 {
2310         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2311         const struct ipv6hdr *key_iph = outer_iph;
2312         struct flow_keys *_flkeys = flkeys;
2313         const struct ipv6hdr *inner_iph;
2314         const struct icmp6hdr *icmph;
2315         struct ipv6hdr _inner_iph;
2316         struct icmp6hdr _icmph;
2317
2318         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2319                 goto out;
2320
2321         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2322                                    sizeof(_icmph), &_icmph);
2323         if (!icmph)
2324                 goto out;
2325
2326         if (!icmpv6_is_err(icmph->icmp6_type))
2327                 goto out;
2328
2329         inner_iph = skb_header_pointer(skb,
2330                                        skb_transport_offset(skb) + sizeof(*icmph),
2331                                        sizeof(_inner_iph), &_inner_iph);
2332         if (!inner_iph)
2333                 goto out;
2334
2335         key_iph = inner_iph;
2336         _flkeys = NULL;
2337 out:
2338         if (_flkeys) {
2339                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2340                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2341                 keys->tags.flow_label = _flkeys->tags.flow_label;
2342                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2343         } else {
2344                 keys->addrs.v6addrs.src = key_iph->saddr;
2345                 keys->addrs.v6addrs.dst = key_iph->daddr;
2346                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2347                 keys->basic.ip_proto = key_iph->nexthdr;
2348         }
2349 }
2350
2351 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2352                                            const struct sk_buff *skb,
2353                                            bool *p_has_inner)
2354 {
2355         u32 hash_fields = ip6_multipath_hash_fields(net);
2356         struct flow_keys keys, hash_keys;
2357
2358         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2359                 return 0;
2360
2361         memset(&hash_keys, 0, sizeof(hash_keys));
2362         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2363
2364         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2365         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2366                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2367         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2368                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2369         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2370                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2371         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2372                 hash_keys.tags.flow_label = keys.tags.flow_label;
2373         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2374                 hash_keys.ports.src = keys.ports.src;
2375         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2376                 hash_keys.ports.dst = keys.ports.dst;
2377
2378         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2379         return flow_hash_from_keys(&hash_keys);
2380 }
2381
2382 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2383                                            const struct sk_buff *skb,
2384                                            bool has_inner)
2385 {
2386         u32 hash_fields = ip6_multipath_hash_fields(net);
2387         struct flow_keys keys, hash_keys;
2388
2389         /* We assume the packet carries an encapsulation, but if none was
2390          * encountered during dissection of the outer flow, then there is no
2391          * point in calling the flow dissector again.
2392          */
2393         if (!has_inner)
2394                 return 0;
2395
2396         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2397                 return 0;
2398
2399         memset(&hash_keys, 0, sizeof(hash_keys));
2400         skb_flow_dissect_flow_keys(skb, &keys, 0);
2401
2402         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2403                 return 0;
2404
2405         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2406                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2407                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2408                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2409                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2410                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2411         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2412                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2413                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2414                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2415                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2416                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2417                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2418                         hash_keys.tags.flow_label = keys.tags.flow_label;
2419         }
2420
2421         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2422                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2423         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2424                 hash_keys.ports.src = keys.ports.src;
2425         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2426                 hash_keys.ports.dst = keys.ports.dst;
2427
2428         return flow_hash_from_keys(&hash_keys);
2429 }
2430
2431 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2432                                          const struct sk_buff *skb)
2433 {
2434         u32 mhash, mhash_inner;
2435         bool has_inner = true;
2436
2437         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2438         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2439
2440         return jhash_2words(mhash, mhash_inner, 0);
2441 }
2442
2443 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2444                                          const struct flowi6 *fl6)
2445 {
2446         u32 hash_fields = ip6_multipath_hash_fields(net);
2447         struct flow_keys hash_keys;
2448
2449         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2450                 return 0;
2451
2452         memset(&hash_keys, 0, sizeof(hash_keys));
2453         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2454         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2455                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2456         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2457                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2458         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2459                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2460         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2461                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2462         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2463                 hash_keys.ports.src = fl6->fl6_sport;
2464         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2465                 hash_keys.ports.dst = fl6->fl6_dport;
2466
2467         return flow_hash_from_keys(&hash_keys);
2468 }
2469
2470 /* if skb is set it will be used and fl6 can be NULL */
2471 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2472                        const struct sk_buff *skb, struct flow_keys *flkeys)
2473 {
2474         struct flow_keys hash_keys;
2475         u32 mhash = 0;
2476
2477         switch (ip6_multipath_hash_policy(net)) {
2478         case 0:
2479                 memset(&hash_keys, 0, sizeof(hash_keys));
2480                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2481                 if (skb) {
2482                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2483                 } else {
2484                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2485                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2486                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2487                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2488                 }
2489                 mhash = flow_hash_from_keys(&hash_keys);
2490                 break;
2491         case 1:
2492                 if (skb) {
2493                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2494                         struct flow_keys keys;
2495
2496                         /* short-circuit if we already have L4 hash present */
2497                         if (skb->l4_hash)
2498                                 return skb_get_hash_raw(skb) >> 1;
2499
2500                         memset(&hash_keys, 0, sizeof(hash_keys));
2501
2502                         if (!flkeys) {
2503                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2504                                 flkeys = &keys;
2505                         }
2506                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2507                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2508                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2509                         hash_keys.ports.src = flkeys->ports.src;
2510                         hash_keys.ports.dst = flkeys->ports.dst;
2511                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2512                 } else {
2513                         memset(&hash_keys, 0, sizeof(hash_keys));
2514                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2515                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2516                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2517                         hash_keys.ports.src = fl6->fl6_sport;
2518                         hash_keys.ports.dst = fl6->fl6_dport;
2519                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2520                 }
2521                 mhash = flow_hash_from_keys(&hash_keys);
2522                 break;
2523         case 2:
2524                 memset(&hash_keys, 0, sizeof(hash_keys));
2525                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2526                 if (skb) {
2527                         struct flow_keys keys;
2528
2529                         if (!flkeys) {
2530                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2531                                 flkeys = &keys;
2532                         }
2533
2534                         /* Inner can be v4 or v6 */
2535                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2536                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2537                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2538                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2539                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2540                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2541                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2542                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2543                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2544                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2545                         } else {
2546                                 /* Same as case 0 */
2547                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2548                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2549                         }
2550                 } else {
2551                         /* Same as case 0 */
2552                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2553                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2554                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2555                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2556                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2557                 }
2558                 mhash = flow_hash_from_keys(&hash_keys);
2559                 break;
2560         case 3:
2561                 if (skb)
2562                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2563                 else
2564                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2565                 break;
2566         }
2567
2568         return mhash >> 1;
2569 }
2570
2571 /* Called with rcu held */
2572 void ip6_route_input(struct sk_buff *skb)
2573 {
2574         const struct ipv6hdr *iph = ipv6_hdr(skb);
2575         struct net *net = dev_net(skb->dev);
2576         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2577         struct ip_tunnel_info *tun_info;
2578         struct flowi6 fl6 = {
2579                 .flowi6_iif = skb->dev->ifindex,
2580                 .daddr = iph->daddr,
2581                 .saddr = iph->saddr,
2582                 .flowlabel = ip6_flowinfo(iph),
2583                 .flowi6_mark = skb->mark,
2584                 .flowi6_proto = iph->nexthdr,
2585         };
2586         struct flow_keys *flkeys = NULL, _flkeys;
2587
2588         tun_info = skb_tunnel_info(skb);
2589         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2590                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2591
2592         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2593                 flkeys = &_flkeys;
2594
2595         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2596                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2597         skb_dst_drop(skb);
2598         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2599                                                       &fl6, skb, flags));
2600 }
2601
2602 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2603                                              struct fib6_table *table,
2604                                              struct flowi6 *fl6,
2605                                              const struct sk_buff *skb,
2606                                              int flags)
2607 {
2608         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2609 }
2610
2611 static struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2612                                                       const struct sock *sk,
2613                                                       struct flowi6 *fl6,
2614                                                       int flags)
2615 {
2616         bool any_src;
2617
2618         if (ipv6_addr_type(&fl6->daddr) &
2619             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2620                 struct dst_entry *dst;
2621
2622                 /* This function does not take refcnt on the dst */
2623                 dst = l3mdev_link_scope_lookup(net, fl6);
2624                 if (dst)
2625                         return dst;
2626         }
2627
2628         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2629
2630         flags |= RT6_LOOKUP_F_DST_NOREF;
2631         any_src = ipv6_addr_any(&fl6->saddr);
2632         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2633             (fl6->flowi6_oif && any_src))
2634                 flags |= RT6_LOOKUP_F_IFACE;
2635
2636         if (!any_src)
2637                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2638         else if (sk)
2639                 flags |= rt6_srcprefs2flags(READ_ONCE(inet6_sk(sk)->srcprefs));
2640
2641         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2642 }
2643
2644 struct dst_entry *ip6_route_output_flags(struct net *net,
2645                                          const struct sock *sk,
2646                                          struct flowi6 *fl6,
2647                                          int flags)
2648 {
2649         struct dst_entry *dst;
2650         struct rt6_info *rt6;
2651
2652         rcu_read_lock();
2653         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2654         rt6 = dst_rt6_info(dst);
2655         /* For dst cached in uncached_list, refcnt is already taken. */
2656         if (list_empty(&rt6->dst.rt_uncached) && !dst_hold_safe(dst)) {
2657                 dst = &net->ipv6.ip6_null_entry->dst;
2658                 dst_hold(dst);
2659         }
2660         rcu_read_unlock();
2661
2662         return dst;
2663 }
2664 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2665
2666 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2667 {
2668         struct rt6_info *rt, *ort = dst_rt6_info(dst_orig);
2669         struct net_device *loopback_dev = net->loopback_dev;
2670         struct dst_entry *new = NULL;
2671
2672         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev,
2673                        DST_OBSOLETE_DEAD, 0);
2674         if (rt) {
2675                 rt6_info_init(rt);
2676                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2677
2678                 new = &rt->dst;
2679                 new->__use = 1;
2680                 new->input = dst_discard;
2681                 new->output = dst_discard_out;
2682
2683                 dst_copy_metrics(new, &ort->dst);
2684
2685                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2686                 rt->rt6i_gateway = ort->rt6i_gateway;
2687                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2688
2689                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2690 #ifdef CONFIG_IPV6_SUBTREES
2691                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2692 #endif
2693         }
2694
2695         dst_release(dst_orig);
2696         return new ? new : ERR_PTR(-ENOMEM);
2697 }
2698
2699 /*
2700  *      Destination cache support functions
2701  */
2702
2703 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2704 {
2705         u32 rt_cookie = 0;
2706
2707         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2708                 return false;
2709
2710         if (fib6_check_expired(f6i))
2711                 return false;
2712
2713         return true;
2714 }
2715
2716 static struct dst_entry *rt6_check(struct rt6_info *rt,
2717                                    struct fib6_info *from,
2718                                    u32 cookie)
2719 {
2720         u32 rt_cookie = 0;
2721
2722         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2723             rt_cookie != cookie)
2724                 return NULL;
2725
2726         if (rt6_check_expired(rt))
2727                 return NULL;
2728
2729         return &rt->dst;
2730 }
2731
2732 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2733                                             struct fib6_info *from,
2734                                             u32 cookie)
2735 {
2736         if (!__rt6_check_expired(rt) &&
2737             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2738             fib6_check(from, cookie))
2739                 return &rt->dst;
2740         else
2741                 return NULL;
2742 }
2743
2744 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2745                                                         u32 cookie)
2746 {
2747         struct dst_entry *dst_ret;
2748         struct fib6_info *from;
2749         struct rt6_info *rt;
2750
2751         rt = dst_rt6_info(dst);
2752
2753         if (rt->sernum)
2754                 return rt6_is_valid(rt) ? dst : NULL;
2755
2756         rcu_read_lock();
2757
2758         /* All IPV6 dsts are created with ->obsolete set to the value
2759          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2760          * into this function always.
2761          */
2762
2763         from = rcu_dereference(rt->from);
2764
2765         if (from && (rt->rt6i_flags & RTF_PCPU ||
2766             unlikely(!list_empty(&rt->dst.rt_uncached))))
2767                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2768         else
2769                 dst_ret = rt6_check(rt, from, cookie);
2770
2771         rcu_read_unlock();
2772
2773         return dst_ret;
2774 }
2775 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2776
2777 static void ip6_negative_advice(struct sock *sk,
2778                                 struct dst_entry *dst)
2779 {
2780         struct rt6_info *rt = dst_rt6_info(dst);
2781
2782         if (rt->rt6i_flags & RTF_CACHE) {
2783                 rcu_read_lock();
2784                 if (rt6_check_expired(rt)) {
2785                         /* counteract the dst_release() in sk_dst_reset() */
2786                         dst_hold(dst);
2787                         sk_dst_reset(sk);
2788
2789                         rt6_remove_exception_rt(rt);
2790                 }
2791                 rcu_read_unlock();
2792                 return;
2793         }
2794         sk_dst_reset(sk);
2795 }
2796
2797 static void ip6_link_failure(struct sk_buff *skb)
2798 {
2799         struct rt6_info *rt;
2800
2801         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2802
2803         rt = dst_rt6_info(skb_dst(skb));
2804         if (rt) {
2805                 rcu_read_lock();
2806                 if (rt->rt6i_flags & RTF_CACHE) {
2807                         rt6_remove_exception_rt(rt);
2808                 } else {
2809                         struct fib6_info *from;
2810                         struct fib6_node *fn;
2811
2812                         from = rcu_dereference(rt->from);
2813                         if (from) {
2814                                 fn = rcu_dereference(from->fib6_node);
2815                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2816                                         WRITE_ONCE(fn->fn_sernum, -1);
2817                         }
2818                 }
2819                 rcu_read_unlock();
2820         }
2821 }
2822
2823 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2824 {
2825         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2826                 struct fib6_info *from;
2827
2828                 rcu_read_lock();
2829                 from = rcu_dereference(rt0->from);
2830                 if (from)
2831                         rt0->dst.expires = from->expires;
2832                 rcu_read_unlock();
2833         }
2834
2835         dst_set_expires(&rt0->dst, timeout);
2836         rt0->rt6i_flags |= RTF_EXPIRES;
2837 }
2838
2839 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2840 {
2841         struct net *net = dev_net(rt->dst.dev);
2842
2843         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2844         rt->rt6i_flags |= RTF_MODIFIED;
2845         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2846 }
2847
2848 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2849 {
2850         return !(rt->rt6i_flags & RTF_CACHE) &&
2851                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2852 }
2853
2854 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2855                                  const struct ipv6hdr *iph, u32 mtu,
2856                                  bool confirm_neigh)
2857 {
2858         const struct in6_addr *daddr, *saddr;
2859         struct rt6_info *rt6 = dst_rt6_info(dst);
2860
2861         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2862          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2863          * [see also comment in rt6_mtu_change_route()]
2864          */
2865
2866         if (iph) {
2867                 daddr = &iph->daddr;
2868                 saddr = &iph->saddr;
2869         } else if (sk) {
2870                 daddr = &sk->sk_v6_daddr;
2871                 saddr = &inet6_sk(sk)->saddr;
2872         } else {
2873                 daddr = NULL;
2874                 saddr = NULL;
2875         }
2876
2877         if (confirm_neigh)
2878                 dst_confirm_neigh(dst, daddr);
2879
2880         if (mtu < IPV6_MIN_MTU)
2881                 return;
2882         if (mtu >= dst_mtu(dst))
2883                 return;
2884
2885         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2886                 rt6_do_update_pmtu(rt6, mtu);
2887                 /* update rt6_ex->stamp for cache */
2888                 if (rt6->rt6i_flags & RTF_CACHE)
2889                         rt6_update_exception_stamp_rt(rt6);
2890         } else if (daddr) {
2891                 struct fib6_result res = {};
2892                 struct rt6_info *nrt6;
2893
2894                 rcu_read_lock();
2895                 res.f6i = rcu_dereference(rt6->from);
2896                 if (!res.f6i)
2897                         goto out_unlock;
2898
2899                 res.fib6_flags = res.f6i->fib6_flags;
2900                 res.fib6_type = res.f6i->fib6_type;
2901
2902                 if (res.f6i->nh) {
2903                         struct fib6_nh_match_arg arg = {
2904                                 .dev = dst->dev,
2905                                 .gw = &rt6->rt6i_gateway,
2906                         };
2907
2908                         nexthop_for_each_fib6_nh(res.f6i->nh,
2909                                                  fib6_nh_find_match, &arg);
2910
2911                         /* fib6_info uses a nexthop that does not have fib6_nh
2912                          * using the dst->dev + gw. Should be impossible.
2913                          */
2914                         if (!arg.match)
2915                                 goto out_unlock;
2916
2917                         res.nh = arg.match;
2918                 } else {
2919                         res.nh = res.f6i->fib6_nh;
2920                 }
2921
2922                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2923                 if (nrt6) {
2924                         rt6_do_update_pmtu(nrt6, mtu);
2925                         if (rt6_insert_exception(nrt6, &res))
2926                                 dst_release_immediate(&nrt6->dst);
2927                 }
2928 out_unlock:
2929                 rcu_read_unlock();
2930         }
2931 }
2932
2933 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2934                                struct sk_buff *skb, u32 mtu,
2935                                bool confirm_neigh)
2936 {
2937         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2938                              confirm_neigh);
2939 }
2940
2941 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2942                      int oif, u32 mark, kuid_t uid)
2943 {
2944         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2945         struct dst_entry *dst;
2946         struct flowi6 fl6 = {
2947                 .flowi6_oif = oif,
2948                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2949                 .daddr = iph->daddr,
2950                 .saddr = iph->saddr,
2951                 .flowlabel = ip6_flowinfo(iph),
2952                 .flowi6_uid = uid,
2953         };
2954
2955         dst = ip6_route_output(net, NULL, &fl6);
2956         if (!dst->error)
2957                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2958         dst_release(dst);
2959 }
2960 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2961
2962 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2963 {
2964         int oif = sk->sk_bound_dev_if;
2965         struct dst_entry *dst;
2966
2967         if (!oif && skb->dev)
2968                 oif = l3mdev_master_ifindex(skb->dev);
2969
2970         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, READ_ONCE(sk->sk_mark),
2971                         sk->sk_uid);
2972
2973         dst = __sk_dst_get(sk);
2974         if (!dst || !dst->obsolete ||
2975             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2976                 return;
2977
2978         bh_lock_sock(sk);
2979         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2980                 ip6_datagram_dst_update(sk, false);
2981         bh_unlock_sock(sk);
2982 }
2983 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2984
2985 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2986                            const struct flowi6 *fl6)
2987 {
2988 #ifdef CONFIG_IPV6_SUBTREES
2989         struct ipv6_pinfo *np = inet6_sk(sk);
2990 #endif
2991
2992         ip6_dst_store(sk, dst,
2993                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2994                       &sk->sk_v6_daddr : NULL,
2995 #ifdef CONFIG_IPV6_SUBTREES
2996                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2997                       &np->saddr :
2998 #endif
2999                       NULL);
3000 }
3001
3002 static bool ip6_redirect_nh_match(const struct fib6_result *res,
3003                                   struct flowi6 *fl6,
3004                                   const struct in6_addr *gw,
3005                                   struct rt6_info **ret)
3006 {
3007         const struct fib6_nh *nh = res->nh;
3008
3009         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
3010             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
3011                 return false;
3012
3013         /* rt_cache's gateway might be different from its 'parent'
3014          * in the case of an ip redirect.
3015          * So we keep searching in the exception table if the gateway
3016          * is different.
3017          */
3018         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3019                 struct rt6_info *rt_cache;
3020
3021                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3022                 if (rt_cache &&
3023                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3024                         *ret = rt_cache;
3025                         return true;
3026                 }
3027                 return false;
3028         }
3029         return true;
3030 }
3031
3032 struct fib6_nh_rd_arg {
3033         struct fib6_result      *res;
3034         struct flowi6           *fl6;
3035         const struct in6_addr   *gw;
3036         struct rt6_info         **ret;
3037 };
3038
3039 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3040 {
3041         struct fib6_nh_rd_arg *arg = _arg;
3042
3043         arg->res->nh = nh;
3044         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3045 }
3046
3047 /* Handle redirects */
3048 struct ip6rd_flowi {
3049         struct flowi6 fl6;
3050         struct in6_addr gateway;
3051 };
3052
3053 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3054                                              struct fib6_table *table,
3055                                              struct flowi6 *fl6,
3056                                              const struct sk_buff *skb,
3057                                              int flags)
3058 {
3059         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3060         struct rt6_info *ret = NULL;
3061         struct fib6_result res = {};
3062         struct fib6_nh_rd_arg arg = {
3063                 .res = &res,
3064                 .fl6 = fl6,
3065                 .gw  = &rdfl->gateway,
3066                 .ret = &ret
3067         };
3068         struct fib6_info *rt;
3069         struct fib6_node *fn;
3070
3071         /* Get the "current" route for this destination and
3072          * check if the redirect has come from appropriate router.
3073          *
3074          * RFC 4861 specifies that redirects should only be
3075          * accepted if they come from the nexthop to the target.
3076          * Due to the way the routes are chosen, this notion
3077          * is a bit fuzzy and one might need to check all possible
3078          * routes.
3079          */
3080
3081         rcu_read_lock();
3082         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3083 restart:
3084         for_each_fib6_node_rt_rcu(fn) {
3085                 res.f6i = rt;
3086                 if (fib6_check_expired(rt))
3087                         continue;
3088                 if (rt->fib6_flags & RTF_REJECT)
3089                         break;
3090                 if (unlikely(rt->nh)) {
3091                         if (nexthop_is_blackhole(rt->nh))
3092                                 continue;
3093                         /* on match, res->nh is filled in and potentially ret */
3094                         if (nexthop_for_each_fib6_nh(rt->nh,
3095                                                      fib6_nh_redirect_match,
3096                                                      &arg))
3097                                 goto out;
3098                 } else {
3099                         res.nh = rt->fib6_nh;
3100                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3101                                                   &ret))
3102                                 goto out;
3103                 }
3104         }
3105
3106         if (!rt)
3107                 rt = net->ipv6.fib6_null_entry;
3108         else if (rt->fib6_flags & RTF_REJECT) {
3109                 ret = net->ipv6.ip6_null_entry;
3110                 goto out;
3111         }
3112
3113         if (rt == net->ipv6.fib6_null_entry) {
3114                 fn = fib6_backtrack(fn, &fl6->saddr);
3115                 if (fn)
3116                         goto restart;
3117         }
3118
3119         res.f6i = rt;
3120         res.nh = rt->fib6_nh;
3121 out:
3122         if (ret) {
3123                 ip6_hold_safe(net, &ret);
3124         } else {
3125                 res.fib6_flags = res.f6i->fib6_flags;
3126                 res.fib6_type = res.f6i->fib6_type;
3127                 ret = ip6_create_rt_rcu(&res);
3128         }
3129
3130         rcu_read_unlock();
3131
3132         trace_fib6_table_lookup(net, &res, table, fl6);
3133         return ret;
3134 };
3135
3136 static struct dst_entry *ip6_route_redirect(struct net *net,
3137                                             const struct flowi6 *fl6,
3138                                             const struct sk_buff *skb,
3139                                             const struct in6_addr *gateway)
3140 {
3141         int flags = RT6_LOOKUP_F_HAS_SADDR;
3142         struct ip6rd_flowi rdfl;
3143
3144         rdfl.fl6 = *fl6;
3145         rdfl.gateway = *gateway;
3146
3147         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3148                                 flags, __ip6_route_redirect);
3149 }
3150
3151 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3152                   kuid_t uid)
3153 {
3154         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3155         struct dst_entry *dst;
3156         struct flowi6 fl6 = {
3157                 .flowi6_iif = LOOPBACK_IFINDEX,
3158                 .flowi6_oif = oif,
3159                 .flowi6_mark = mark,
3160                 .daddr = iph->daddr,
3161                 .saddr = iph->saddr,
3162                 .flowlabel = ip6_flowinfo(iph),
3163                 .flowi6_uid = uid,
3164         };
3165
3166         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3167         rt6_do_redirect(dst, NULL, skb);
3168         dst_release(dst);
3169 }
3170 EXPORT_SYMBOL_GPL(ip6_redirect);
3171
3172 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3173 {
3174         const struct ipv6hdr *iph = ipv6_hdr(skb);
3175         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3176         struct dst_entry *dst;
3177         struct flowi6 fl6 = {
3178                 .flowi6_iif = LOOPBACK_IFINDEX,
3179                 .flowi6_oif = oif,
3180                 .daddr = msg->dest,
3181                 .saddr = iph->daddr,
3182                 .flowi6_uid = sock_net_uid(net, NULL),
3183         };
3184
3185         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3186         rt6_do_redirect(dst, NULL, skb);
3187         dst_release(dst);
3188 }
3189
3190 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3191 {
3192         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if,
3193                      READ_ONCE(sk->sk_mark), sk->sk_uid);
3194 }
3195 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3196
3197 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3198 {
3199         struct net_device *dev = dst->dev;
3200         unsigned int mtu = dst_mtu(dst);
3201         struct net *net = dev_net(dev);
3202
3203         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3204
3205         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3206                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3207
3208         /*
3209          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3210          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3211          * IPV6_MAXPLEN is also valid and means: "any MSS,
3212          * rely only on pmtu discovery"
3213          */
3214         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3215                 mtu = IPV6_MAXPLEN;
3216         return mtu;
3217 }
3218
3219 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3220 {
3221         return ip6_dst_mtu_maybe_forward(dst, false);
3222 }
3223 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3224
3225 /* MTU selection:
3226  * 1. mtu on route is locked - use it
3227  * 2. mtu from nexthop exception
3228  * 3. mtu from egress device
3229  *
3230  * based on ip6_dst_mtu_forward and exception logic of
3231  * rt6_find_cached_rt; called with rcu_read_lock
3232  */
3233 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3234                       const struct in6_addr *daddr,
3235                       const struct in6_addr *saddr)
3236 {
3237         const struct fib6_nh *nh = res->nh;
3238         struct fib6_info *f6i = res->f6i;
3239         struct inet6_dev *idev;
3240         struct rt6_info *rt;
3241         u32 mtu = 0;
3242
3243         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3244                 mtu = f6i->fib6_pmtu;
3245                 if (mtu)
3246                         goto out;
3247         }
3248
3249         rt = rt6_find_cached_rt(res, daddr, saddr);
3250         if (unlikely(rt)) {
3251                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3252         } else {
3253                 struct net_device *dev = nh->fib_nh_dev;
3254
3255                 mtu = IPV6_MIN_MTU;
3256                 idev = __in6_dev_get(dev);
3257                 if (idev)
3258                         mtu = max_t(u32, mtu, READ_ONCE(idev->cnf.mtu6));
3259         }
3260
3261         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3262 out:
3263         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3264 }
3265
3266 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3267                                   struct flowi6 *fl6)
3268 {
3269         struct dst_entry *dst;
3270         struct rt6_info *rt;
3271         struct inet6_dev *idev = in6_dev_get(dev);
3272         struct net *net = dev_net(dev);
3273
3274         if (unlikely(!idev))
3275                 return ERR_PTR(-ENODEV);
3276
3277         rt = ip6_dst_alloc(net, dev, 0);
3278         if (unlikely(!rt)) {
3279                 in6_dev_put(idev);
3280                 dst = ERR_PTR(-ENOMEM);
3281                 goto out;
3282         }
3283
3284         rt->dst.input = ip6_input;
3285         rt->dst.output  = ip6_output;
3286         rt->rt6i_gateway  = fl6->daddr;
3287         rt->rt6i_dst.addr = fl6->daddr;
3288         rt->rt6i_dst.plen = 128;
3289         rt->rt6i_idev     = idev;
3290         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3291
3292         /* Add this dst into uncached_list so that rt6_disable_ip() can
3293          * do proper release of the net_device
3294          */
3295         rt6_uncached_list_add(rt);
3296
3297         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3298
3299 out:
3300         return dst;
3301 }
3302
3303 static void ip6_dst_gc(struct dst_ops *ops)
3304 {
3305         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3306         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3307         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3308         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3309         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3310         unsigned int val;
3311         int entries;
3312
3313         if (time_after(rt_last_gc + rt_min_interval, jiffies))
3314                 goto out;
3315
3316         fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3317         entries = dst_entries_get_slow(ops);
3318         if (entries < ops->gc_thresh)
3319                 atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3320 out:
3321         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3322         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3323 }
3324
3325 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3326                                const struct in6_addr *gw_addr, u32 tbid,
3327                                int flags, struct fib6_result *res)
3328 {
3329         struct flowi6 fl6 = {
3330                 .flowi6_oif = cfg->fc_ifindex,
3331                 .daddr = *gw_addr,
3332                 .saddr = cfg->fc_prefsrc,
3333         };
3334         struct fib6_table *table;
3335         int err;
3336
3337         table = fib6_get_table(net, tbid);
3338         if (!table)
3339                 return -EINVAL;
3340
3341         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3342                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3343
3344         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3345
3346         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3347         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3348                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3349                                  cfg->fc_ifindex != 0, NULL, flags);
3350
3351         return err;
3352 }
3353
3354 static int ip6_route_check_nh_onlink(struct net *net,
3355                                      struct fib6_config *cfg,
3356                                      const struct net_device *dev,
3357                                      struct netlink_ext_ack *extack)
3358 {
3359         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3360         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3361         struct fib6_result res = {};
3362         int err;
3363
3364         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3365         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3366             /* ignore match if it is the default route */
3367             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3368             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3369                 NL_SET_ERR_MSG(extack,
3370                                "Nexthop has invalid gateway or device mismatch");
3371                 err = -EINVAL;
3372         }
3373
3374         return err;
3375 }
3376
3377 static int ip6_route_check_nh(struct net *net,
3378                               struct fib6_config *cfg,
3379                               struct net_device **_dev,
3380                               netdevice_tracker *dev_tracker,
3381                               struct inet6_dev **idev)
3382 {
3383         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3384         struct net_device *dev = _dev ? *_dev : NULL;
3385         int flags = RT6_LOOKUP_F_IFACE;
3386         struct fib6_result res = {};
3387         int err = -EHOSTUNREACH;
3388
3389         if (cfg->fc_table) {
3390                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3391                                           cfg->fc_table, flags, &res);
3392                 /* gw_addr can not require a gateway or resolve to a reject
3393                  * route. If a device is given, it must match the result.
3394                  */
3395                 if (err || res.fib6_flags & RTF_REJECT ||
3396                     res.nh->fib_nh_gw_family ||
3397                     (dev && dev != res.nh->fib_nh_dev))
3398                         err = -EHOSTUNREACH;
3399         }
3400
3401         if (err < 0) {
3402                 struct flowi6 fl6 = {
3403                         .flowi6_oif = cfg->fc_ifindex,
3404                         .daddr = *gw_addr,
3405                 };
3406
3407                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3408                 if (err || res.fib6_flags & RTF_REJECT ||
3409                     res.nh->fib_nh_gw_family)
3410                         err = -EHOSTUNREACH;
3411
3412                 if (err)
3413                         return err;
3414
3415                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3416                                  cfg->fc_ifindex != 0, NULL, flags);
3417         }
3418
3419         err = 0;
3420         if (dev) {
3421                 if (dev != res.nh->fib_nh_dev)
3422                         err = -EHOSTUNREACH;
3423         } else {
3424                 *_dev = dev = res.nh->fib_nh_dev;
3425                 netdev_hold(dev, dev_tracker, GFP_ATOMIC);
3426                 *idev = in6_dev_get(dev);
3427         }
3428
3429         return err;
3430 }
3431
3432 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3433                            struct net_device **_dev,
3434                            netdevice_tracker *dev_tracker,
3435                            struct inet6_dev **idev,
3436                            struct netlink_ext_ack *extack)
3437 {
3438         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3439         int gwa_type = ipv6_addr_type(gw_addr);
3440         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3441         const struct net_device *dev = *_dev;
3442         bool need_addr_check = !dev;
3443         int err = -EINVAL;
3444
3445         /* if gw_addr is local we will fail to detect this in case
3446          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3447          * will return already-added prefix route via interface that
3448          * prefix route was assigned to, which might be non-loopback.
3449          */
3450         if (dev &&
3451             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3452                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3453                 goto out;
3454         }
3455
3456         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3457                 /* IPv6 strictly inhibits using not link-local
3458                  * addresses as nexthop address.
3459                  * Otherwise, router will not able to send redirects.
3460                  * It is very good, but in some (rare!) circumstances
3461                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3462                  * some exceptions. --ANK
3463                  * We allow IPv4-mapped nexthops to support RFC4798-type
3464                  * addressing
3465                  */
3466                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3467                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3468                         goto out;
3469                 }
3470
3471                 rcu_read_lock();
3472
3473                 if (cfg->fc_flags & RTNH_F_ONLINK)
3474                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3475                 else
3476                         err = ip6_route_check_nh(net, cfg, _dev, dev_tracker,
3477                                                  idev);
3478
3479                 rcu_read_unlock();
3480
3481                 if (err)
3482                         goto out;
3483         }
3484
3485         /* reload in case device was changed */
3486         dev = *_dev;
3487
3488         err = -EINVAL;
3489         if (!dev) {
3490                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3491                 goto out;
3492         } else if (dev->flags & IFF_LOOPBACK) {
3493                 NL_SET_ERR_MSG(extack,
3494                                "Egress device can not be loopback device for this route");
3495                 goto out;
3496         }
3497
3498         /* if we did not check gw_addr above, do so now that the
3499          * egress device has been resolved.
3500          */
3501         if (need_addr_check &&
3502             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3503                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3504                 goto out;
3505         }
3506
3507         err = 0;
3508 out:
3509         return err;
3510 }
3511
3512 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3513 {
3514         if ((flags & RTF_REJECT) ||
3515             (dev && (dev->flags & IFF_LOOPBACK) &&
3516              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3517              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3518                 return true;
3519
3520         return false;
3521 }
3522
3523 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3524                  struct fib6_config *cfg, gfp_t gfp_flags,
3525                  struct netlink_ext_ack *extack)
3526 {
3527         netdevice_tracker *dev_tracker = &fib6_nh->fib_nh_dev_tracker;
3528         struct net_device *dev = NULL;
3529         struct inet6_dev *idev = NULL;
3530         int addr_type;
3531         int err;
3532
3533         fib6_nh->fib_nh_family = AF_INET6;
3534 #ifdef CONFIG_IPV6_ROUTER_PREF
3535         fib6_nh->last_probe = jiffies;
3536 #endif
3537         if (cfg->fc_is_fdb) {
3538                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3539                 fib6_nh->fib_nh_gw_family = AF_INET6;
3540                 return 0;
3541         }
3542
3543         err = -ENODEV;
3544         if (cfg->fc_ifindex) {
3545                 dev = netdev_get_by_index(net, cfg->fc_ifindex,
3546                                           dev_tracker, gfp_flags);
3547                 if (!dev)
3548                         goto out;
3549                 idev = in6_dev_get(dev);
3550                 if (!idev)
3551                         goto out;
3552         }
3553
3554         if (cfg->fc_flags & RTNH_F_ONLINK) {
3555                 if (!dev) {
3556                         NL_SET_ERR_MSG(extack,
3557                                        "Nexthop device required for onlink");
3558                         goto out;
3559                 }
3560
3561                 if (!(dev->flags & IFF_UP)) {
3562                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3563                         err = -ENETDOWN;
3564                         goto out;
3565                 }
3566
3567                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3568         }
3569
3570         fib6_nh->fib_nh_weight = 1;
3571
3572         /* We cannot add true routes via loopback here,
3573          * they would result in kernel looping; promote them to reject routes
3574          */
3575         addr_type = ipv6_addr_type(&cfg->fc_dst);
3576         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3577                 /* hold loopback dev/idev if we haven't done so. */
3578                 if (dev != net->loopback_dev) {
3579                         if (dev) {
3580                                 netdev_put(dev, dev_tracker);
3581                                 in6_dev_put(idev);
3582                         }
3583                         dev = net->loopback_dev;
3584                         netdev_hold(dev, dev_tracker, gfp_flags);
3585                         idev = in6_dev_get(dev);
3586                         if (!idev) {
3587                                 err = -ENODEV;
3588                                 goto out;
3589                         }
3590                 }
3591                 goto pcpu_alloc;
3592         }
3593
3594         if (cfg->fc_flags & RTF_GATEWAY) {
3595                 err = ip6_validate_gw(net, cfg, &dev, dev_tracker,
3596                                       &idev, extack);
3597                 if (err)
3598                         goto out;
3599
3600                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3601                 fib6_nh->fib_nh_gw_family = AF_INET6;
3602         }
3603
3604         err = -ENODEV;
3605         if (!dev)
3606                 goto out;
3607
3608         if (!idev || idev->cnf.disable_ipv6) {
3609                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3610                 err = -EACCES;
3611                 goto out;
3612         }
3613
3614         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3615                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3616                 err = -ENETDOWN;
3617                 goto out;
3618         }
3619
3620         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3621             !netif_carrier_ok(dev))
3622                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3623
3624         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3625                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3626         if (err)
3627                 goto out;
3628
3629 pcpu_alloc:
3630         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3631         if (!fib6_nh->rt6i_pcpu) {
3632                 err = -ENOMEM;
3633                 goto out;
3634         }
3635
3636         fib6_nh->fib_nh_dev = dev;
3637         fib6_nh->fib_nh_oif = dev->ifindex;
3638         err = 0;
3639 out:
3640         if (idev)
3641                 in6_dev_put(idev);
3642
3643         if (err) {
3644                 lwtstate_put(fib6_nh->fib_nh_lws);
3645                 fib6_nh->fib_nh_lws = NULL;
3646                 netdev_put(dev, dev_tracker);
3647         }
3648
3649         return err;
3650 }
3651
3652 void fib6_nh_release(struct fib6_nh *fib6_nh)
3653 {
3654         struct rt6_exception_bucket *bucket;
3655
3656         rcu_read_lock();
3657
3658         fib6_nh_flush_exceptions(fib6_nh, NULL);
3659         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3660         if (bucket) {
3661                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3662                 kfree(bucket);
3663         }
3664
3665         rcu_read_unlock();
3666
3667         fib6_nh_release_dsts(fib6_nh);
3668         free_percpu(fib6_nh->rt6i_pcpu);
3669
3670         fib_nh_common_release(&fib6_nh->nh_common);
3671 }
3672
3673 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3674 {
3675         int cpu;
3676
3677         if (!fib6_nh->rt6i_pcpu)
3678                 return;
3679
3680         for_each_possible_cpu(cpu) {
3681                 struct rt6_info *pcpu_rt, **ppcpu_rt;
3682
3683                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3684                 pcpu_rt = xchg(ppcpu_rt, NULL);
3685                 if (pcpu_rt) {
3686                         dst_dev_put(&pcpu_rt->dst);
3687                         dst_release(&pcpu_rt->dst);
3688                 }
3689         }
3690 }
3691
3692 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3693                                               gfp_t gfp_flags,
3694                                               struct netlink_ext_ack *extack)
3695 {
3696         struct net *net = cfg->fc_nlinfo.nl_net;
3697         struct fib6_info *rt = NULL;
3698         struct nexthop *nh = NULL;
3699         struct fib6_table *table;
3700         struct fib6_nh *fib6_nh;
3701         int err = -EINVAL;
3702         int addr_type;
3703
3704         /* RTF_PCPU is an internal flag; can not be set by userspace */
3705         if (cfg->fc_flags & RTF_PCPU) {
3706                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3707                 goto out;
3708         }
3709
3710         /* RTF_CACHE is an internal flag; can not be set by userspace */
3711         if (cfg->fc_flags & RTF_CACHE) {
3712                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3713                 goto out;
3714         }
3715
3716         if (cfg->fc_type > RTN_MAX) {
3717                 NL_SET_ERR_MSG(extack, "Invalid route type");
3718                 goto out;
3719         }
3720
3721         if (cfg->fc_dst_len > 128) {
3722                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3723                 goto out;
3724         }
3725         if (cfg->fc_src_len > 128) {
3726                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3727                 goto out;
3728         }
3729 #ifndef CONFIG_IPV6_SUBTREES
3730         if (cfg->fc_src_len) {
3731                 NL_SET_ERR_MSG(extack,
3732                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3733                 goto out;
3734         }
3735 #endif
3736         if (cfg->fc_nh_id) {
3737                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3738                 if (!nh) {
3739                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3740                         goto out;
3741                 }
3742                 err = fib6_check_nexthop(nh, cfg, extack);
3743                 if (err)
3744                         goto out;
3745         }
3746
3747         err = -ENOBUFS;
3748         if (cfg->fc_nlinfo.nlh &&
3749             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3750                 table = fib6_get_table(net, cfg->fc_table);
3751                 if (!table) {
3752                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3753                         table = fib6_new_table(net, cfg->fc_table);
3754                 }
3755         } else {
3756                 table = fib6_new_table(net, cfg->fc_table);
3757         }
3758
3759         if (!table)
3760                 goto out;
3761
3762         err = -ENOMEM;
3763         rt = fib6_info_alloc(gfp_flags, !nh);
3764         if (!rt)
3765                 goto out;
3766
3767         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3768                                                extack);
3769         if (IS_ERR(rt->fib6_metrics)) {
3770                 err = PTR_ERR(rt->fib6_metrics);
3771                 /* Do not leave garbage there. */
3772                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3773                 goto out_free;
3774         }
3775
3776         if (cfg->fc_flags & RTF_ADDRCONF)
3777                 rt->dst_nocount = true;
3778
3779         if (cfg->fc_flags & RTF_EXPIRES)
3780                 fib6_set_expires(rt, jiffies +
3781                                 clock_t_to_jiffies(cfg->fc_expires));
3782
3783         if (cfg->fc_protocol == RTPROT_UNSPEC)
3784                 cfg->fc_protocol = RTPROT_BOOT;
3785         rt->fib6_protocol = cfg->fc_protocol;
3786
3787         rt->fib6_table = table;
3788         rt->fib6_metric = cfg->fc_metric;
3789         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3790         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3791
3792         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3793         rt->fib6_dst.plen = cfg->fc_dst_len;
3794
3795 #ifdef CONFIG_IPV6_SUBTREES
3796         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3797         rt->fib6_src.plen = cfg->fc_src_len;
3798 #endif
3799         if (nh) {
3800                 if (rt->fib6_src.plen) {
3801                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3802                         goto out_free;
3803                 }
3804                 if (!nexthop_get(nh)) {
3805                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3806                         goto out_free;
3807                 }
3808                 rt->nh = nh;
3809                 fib6_nh = nexthop_fib6_nh(rt->nh);
3810         } else {
3811                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3812                 if (err)
3813                         goto out;
3814
3815                 fib6_nh = rt->fib6_nh;
3816
3817                 /* We cannot add true routes via loopback here, they would
3818                  * result in kernel looping; promote them to reject routes
3819                  */
3820                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3821                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3822                                    addr_type))
3823                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3824         }
3825
3826         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3827                 struct net_device *dev = fib6_nh->fib_nh_dev;
3828
3829                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3830                         NL_SET_ERR_MSG(extack, "Invalid source address");
3831                         err = -EINVAL;
3832                         goto out;
3833                 }
3834                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3835                 rt->fib6_prefsrc.plen = 128;
3836         } else
3837                 rt->fib6_prefsrc.plen = 0;
3838
3839         return rt;
3840 out:
3841         fib6_info_release(rt);
3842         return ERR_PTR(err);
3843 out_free:
3844         ip_fib_metrics_put(rt->fib6_metrics);
3845         kfree(rt);
3846         return ERR_PTR(err);
3847 }
3848
3849 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3850                   struct netlink_ext_ack *extack)
3851 {
3852         struct fib6_info *rt;
3853         int err;
3854
3855         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3856         if (IS_ERR(rt))
3857                 return PTR_ERR(rt);
3858
3859         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3860         fib6_info_release(rt);
3861
3862         return err;
3863 }
3864
3865 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3866 {
3867         struct net *net = info->nl_net;
3868         struct fib6_table *table;
3869         int err;
3870
3871         if (rt == net->ipv6.fib6_null_entry) {
3872                 err = -ENOENT;
3873                 goto out;
3874         }
3875
3876         table = rt->fib6_table;
3877         spin_lock_bh(&table->tb6_lock);
3878         err = fib6_del(rt, info);
3879         spin_unlock_bh(&table->tb6_lock);
3880
3881 out:
3882         fib6_info_release(rt);
3883         return err;
3884 }
3885
3886 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3887 {
3888         struct nl_info info = {
3889                 .nl_net = net,
3890                 .skip_notify = skip_notify
3891         };
3892
3893         return __ip6_del_rt(rt, &info);
3894 }
3895
3896 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3897 {
3898         struct nl_info *info = &cfg->fc_nlinfo;
3899         struct net *net = info->nl_net;
3900         struct sk_buff *skb = NULL;
3901         struct fib6_table *table;
3902         int err = -ENOENT;
3903
3904         if (rt == net->ipv6.fib6_null_entry)
3905                 goto out_put;
3906         table = rt->fib6_table;
3907         spin_lock_bh(&table->tb6_lock);
3908
3909         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3910                 struct fib6_info *sibling, *next_sibling;
3911                 struct fib6_node *fn;
3912
3913                 /* prefer to send a single notification with all hops */
3914                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3915                 if (skb) {
3916                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3917
3918                         if (rt6_fill_node(net, skb, rt, NULL,
3919                                           NULL, NULL, 0, RTM_DELROUTE,
3920                                           info->portid, seq, 0) < 0) {
3921                                 kfree_skb(skb);
3922                                 skb = NULL;
3923                         } else
3924                                 info->skip_notify = 1;
3925                 }
3926
3927                 /* 'rt' points to the first sibling route. If it is not the
3928                  * leaf, then we do not need to send a notification. Otherwise,
3929                  * we need to check if the last sibling has a next route or not
3930                  * and emit a replace or delete notification, respectively.
3931                  */
3932                 info->skip_notify_kernel = 1;
3933                 fn = rcu_dereference_protected(rt->fib6_node,
3934                                             lockdep_is_held(&table->tb6_lock));
3935                 if (rcu_access_pointer(fn->leaf) == rt) {
3936                         struct fib6_info *last_sibling, *replace_rt;
3937
3938                         last_sibling = list_last_entry(&rt->fib6_siblings,
3939                                                        struct fib6_info,
3940                                                        fib6_siblings);
3941                         replace_rt = rcu_dereference_protected(
3942                                             last_sibling->fib6_next,
3943                                             lockdep_is_held(&table->tb6_lock));
3944                         if (replace_rt)
3945                                 call_fib6_entry_notifiers_replace(net,
3946                                                                   replace_rt);
3947                         else
3948                                 call_fib6_multipath_entry_notifiers(net,
3949                                                        FIB_EVENT_ENTRY_DEL,
3950                                                        rt, rt->fib6_nsiblings,
3951                                                        NULL);
3952                 }
3953                 list_for_each_entry_safe(sibling, next_sibling,
3954                                          &rt->fib6_siblings,
3955                                          fib6_siblings) {
3956                         err = fib6_del(sibling, info);
3957                         if (err)
3958                                 goto out_unlock;
3959                 }
3960         }
3961
3962         err = fib6_del(rt, info);
3963 out_unlock:
3964         spin_unlock_bh(&table->tb6_lock);
3965 out_put:
3966         fib6_info_release(rt);
3967
3968         if (skb) {
3969                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3970                             info->nlh, gfp_any());
3971         }
3972         return err;
3973 }
3974
3975 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3976 {
3977         int rc = -ESRCH;
3978
3979         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3980                 goto out;
3981
3982         if (cfg->fc_flags & RTF_GATEWAY &&
3983             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3984                 goto out;
3985
3986         rc = rt6_remove_exception_rt(rt);
3987 out:
3988         return rc;
3989 }
3990
3991 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3992                              struct fib6_nh *nh)
3993 {
3994         struct fib6_result res = {
3995                 .f6i = rt,
3996                 .nh = nh,
3997         };
3998         struct rt6_info *rt_cache;
3999
4000         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
4001         if (rt_cache)
4002                 return __ip6_del_cached_rt(rt_cache, cfg);
4003
4004         return 0;
4005 }
4006
4007 struct fib6_nh_del_cached_rt_arg {
4008         struct fib6_config *cfg;
4009         struct fib6_info *f6i;
4010 };
4011
4012 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4013 {
4014         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4015         int rc;
4016
4017         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4018         return rc != -ESRCH ? rc : 0;
4019 }
4020
4021 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4022 {
4023         struct fib6_nh_del_cached_rt_arg arg = {
4024                 .cfg = cfg,
4025                 .f6i = f6i
4026         };
4027
4028         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4029 }
4030
4031 static int ip6_route_del(struct fib6_config *cfg,
4032                          struct netlink_ext_ack *extack)
4033 {
4034         struct fib6_table *table;
4035         struct fib6_info *rt;
4036         struct fib6_node *fn;
4037         int err = -ESRCH;
4038
4039         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4040         if (!table) {
4041                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4042                 return err;
4043         }
4044
4045         rcu_read_lock();
4046
4047         fn = fib6_locate(&table->tb6_root,
4048                          &cfg->fc_dst, cfg->fc_dst_len,
4049                          &cfg->fc_src, cfg->fc_src_len,
4050                          !(cfg->fc_flags & RTF_CACHE));
4051
4052         if (fn) {
4053                 for_each_fib6_node_rt_rcu(fn) {
4054                         struct fib6_nh *nh;
4055
4056                         if (rt->nh && cfg->fc_nh_id &&
4057                             rt->nh->id != cfg->fc_nh_id)
4058                                 continue;
4059
4060                         if (cfg->fc_flags & RTF_CACHE) {
4061                                 int rc = 0;
4062
4063                                 if (rt->nh) {
4064                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4065                                 } else if (cfg->fc_nh_id) {
4066                                         continue;
4067                                 } else {
4068                                         nh = rt->fib6_nh;
4069                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4070                                 }
4071                                 if (rc != -ESRCH) {
4072                                         rcu_read_unlock();
4073                                         return rc;
4074                                 }
4075                                 continue;
4076                         }
4077
4078                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4079                                 continue;
4080                         if (cfg->fc_protocol &&
4081                             cfg->fc_protocol != rt->fib6_protocol)
4082                                 continue;
4083
4084                         if (rt->nh) {
4085                                 if (!fib6_info_hold_safe(rt))
4086                                         continue;
4087                                 rcu_read_unlock();
4088
4089                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4090                         }
4091                         if (cfg->fc_nh_id)
4092                                 continue;
4093
4094                         nh = rt->fib6_nh;
4095                         if (cfg->fc_ifindex &&
4096                             (!nh->fib_nh_dev ||
4097                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4098                                 continue;
4099                         if (cfg->fc_flags & RTF_GATEWAY &&
4100                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4101                                 continue;
4102                         if (!fib6_info_hold_safe(rt))
4103                                 continue;
4104                         rcu_read_unlock();
4105
4106                         /* if gateway was specified only delete the one hop */
4107                         if (cfg->fc_flags & RTF_GATEWAY)
4108                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4109
4110                         return __ip6_del_rt_siblings(rt, cfg);
4111                 }
4112         }
4113         rcu_read_unlock();
4114
4115         return err;
4116 }
4117
4118 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4119 {
4120         struct netevent_redirect netevent;
4121         struct rt6_info *rt, *nrt = NULL;
4122         struct fib6_result res = {};
4123         struct ndisc_options ndopts;
4124         struct inet6_dev *in6_dev;
4125         struct neighbour *neigh;
4126         struct rd_msg *msg;
4127         int optlen, on_link;
4128         u8 *lladdr;
4129
4130         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4131         optlen -= sizeof(*msg);
4132
4133         if (optlen < 0) {
4134                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4135                 return;
4136         }
4137
4138         msg = (struct rd_msg *)icmp6_hdr(skb);
4139
4140         if (ipv6_addr_is_multicast(&msg->dest)) {
4141                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4142                 return;
4143         }
4144
4145         on_link = 0;
4146         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4147                 on_link = 1;
4148         } else if (ipv6_addr_type(&msg->target) !=
4149                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4150                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4151                 return;
4152         }
4153
4154         in6_dev = __in6_dev_get(skb->dev);
4155         if (!in6_dev)
4156                 return;
4157         if (READ_ONCE(in6_dev->cnf.forwarding) ||
4158             !READ_ONCE(in6_dev->cnf.accept_redirects))
4159                 return;
4160
4161         /* RFC2461 8.1:
4162          *      The IP source address of the Redirect MUST be the same as the current
4163          *      first-hop router for the specified ICMP Destination Address.
4164          */
4165
4166         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4167                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4168                 return;
4169         }
4170
4171         lladdr = NULL;
4172         if (ndopts.nd_opts_tgt_lladdr) {
4173                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4174                                              skb->dev);
4175                 if (!lladdr) {
4176                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4177                         return;
4178                 }
4179         }
4180
4181         rt = dst_rt6_info(dst);
4182         if (rt->rt6i_flags & RTF_REJECT) {
4183                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4184                 return;
4185         }
4186
4187         /* Redirect received -> path was valid.
4188          * Look, redirects are sent only in response to data packets,
4189          * so that this nexthop apparently is reachable. --ANK
4190          */
4191         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4192
4193         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4194         if (!neigh)
4195                 return;
4196
4197         /*
4198          *      We have finally decided to accept it.
4199          */
4200
4201         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4202                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4203                      NEIGH_UPDATE_F_OVERRIDE|
4204                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4205                                      NEIGH_UPDATE_F_ISROUTER)),
4206                      NDISC_REDIRECT, &ndopts);
4207
4208         rcu_read_lock();
4209         res.f6i = rcu_dereference(rt->from);
4210         if (!res.f6i)
4211                 goto out;
4212
4213         if (res.f6i->nh) {
4214                 struct fib6_nh_match_arg arg = {
4215                         .dev = dst->dev,
4216                         .gw = &rt->rt6i_gateway,
4217                 };
4218
4219                 nexthop_for_each_fib6_nh(res.f6i->nh,
4220                                          fib6_nh_find_match, &arg);
4221
4222                 /* fib6_info uses a nexthop that does not have fib6_nh
4223                  * using the dst->dev. Should be impossible
4224                  */
4225                 if (!arg.match)
4226                         goto out;
4227                 res.nh = arg.match;
4228         } else {
4229                 res.nh = res.f6i->fib6_nh;
4230         }
4231
4232         res.fib6_flags = res.f6i->fib6_flags;
4233         res.fib6_type = res.f6i->fib6_type;
4234         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4235         if (!nrt)
4236                 goto out;
4237
4238         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4239         if (on_link)
4240                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4241
4242         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4243
4244         /* rt6_insert_exception() will take care of duplicated exceptions */
4245         if (rt6_insert_exception(nrt, &res)) {
4246                 dst_release_immediate(&nrt->dst);
4247                 goto out;
4248         }
4249
4250         netevent.old = &rt->dst;
4251         netevent.new = &nrt->dst;
4252         netevent.daddr = &msg->dest;
4253         netevent.neigh = neigh;
4254         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4255
4256 out:
4257         rcu_read_unlock();
4258         neigh_release(neigh);
4259 }
4260
4261 #ifdef CONFIG_IPV6_ROUTE_INFO
4262 static struct fib6_info *rt6_get_route_info(struct net *net,
4263                                            const struct in6_addr *prefix, int prefixlen,
4264                                            const struct in6_addr *gwaddr,
4265                                            struct net_device *dev)
4266 {
4267         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4268         int ifindex = dev->ifindex;
4269         struct fib6_node *fn;
4270         struct fib6_info *rt = NULL;
4271         struct fib6_table *table;
4272
4273         table = fib6_get_table(net, tb_id);
4274         if (!table)
4275                 return NULL;
4276
4277         rcu_read_lock();
4278         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4279         if (!fn)
4280                 goto out;
4281
4282         for_each_fib6_node_rt_rcu(fn) {
4283                 /* these routes do not use nexthops */
4284                 if (rt->nh)
4285                         continue;
4286                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4287                         continue;
4288                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4289                     !rt->fib6_nh->fib_nh_gw_family)
4290                         continue;
4291                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4292                         continue;
4293                 if (!fib6_info_hold_safe(rt))
4294                         continue;
4295                 break;
4296         }
4297 out:
4298         rcu_read_unlock();
4299         return rt;
4300 }
4301
4302 static struct fib6_info *rt6_add_route_info(struct net *net,
4303                                            const struct in6_addr *prefix, int prefixlen,
4304                                            const struct in6_addr *gwaddr,
4305                                            struct net_device *dev,
4306                                            unsigned int pref)
4307 {
4308         struct fib6_config cfg = {
4309                 .fc_metric      = IP6_RT_PRIO_USER,
4310                 .fc_ifindex     = dev->ifindex,
4311                 .fc_dst_len     = prefixlen,
4312                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4313                                   RTF_UP | RTF_PREF(pref),
4314                 .fc_protocol = RTPROT_RA,
4315                 .fc_type = RTN_UNICAST,
4316                 .fc_nlinfo.portid = 0,
4317                 .fc_nlinfo.nlh = NULL,
4318                 .fc_nlinfo.nl_net = net,
4319         };
4320
4321         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4322         cfg.fc_dst = *prefix;
4323         cfg.fc_gateway = *gwaddr;
4324
4325         /* We should treat it as a default route if prefix length is 0. */
4326         if (!prefixlen)
4327                 cfg.fc_flags |= RTF_DEFAULT;
4328
4329         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4330
4331         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4332 }
4333 #endif
4334
4335 struct fib6_info *rt6_get_dflt_router(struct net *net,
4336                                      const struct in6_addr *addr,
4337                                      struct net_device *dev)
4338 {
4339         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4340         struct fib6_info *rt;
4341         struct fib6_table *table;
4342
4343         table = fib6_get_table(net, tb_id);
4344         if (!table)
4345                 return NULL;
4346
4347         rcu_read_lock();
4348         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4349                 struct fib6_nh *nh;
4350
4351                 /* RA routes do not use nexthops */
4352                 if (rt->nh)
4353                         continue;
4354
4355                 nh = rt->fib6_nh;
4356                 if (dev == nh->fib_nh_dev &&
4357                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4358                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4359                         break;
4360         }
4361         if (rt && !fib6_info_hold_safe(rt))
4362                 rt = NULL;
4363         rcu_read_unlock();
4364         return rt;
4365 }
4366
4367 struct fib6_info *rt6_add_dflt_router(struct net *net,
4368                                      const struct in6_addr *gwaddr,
4369                                      struct net_device *dev,
4370                                      unsigned int pref,
4371                                      u32 defrtr_usr_metric,
4372                                      int lifetime)
4373 {
4374         struct fib6_config cfg = {
4375                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4376                 .fc_metric      = defrtr_usr_metric,
4377                 .fc_ifindex     = dev->ifindex,
4378                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4379                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4380                 .fc_protocol = RTPROT_RA,
4381                 .fc_type = RTN_UNICAST,
4382                 .fc_nlinfo.portid = 0,
4383                 .fc_nlinfo.nlh = NULL,
4384                 .fc_nlinfo.nl_net = net,
4385                 .fc_expires = jiffies_to_clock_t(lifetime * HZ),
4386         };
4387
4388         cfg.fc_gateway = *gwaddr;
4389
4390         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4391                 struct fib6_table *table;
4392
4393                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4394                 if (table)
4395                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4396         }
4397
4398         return rt6_get_dflt_router(net, gwaddr, dev);
4399 }
4400
4401 static void __rt6_purge_dflt_routers(struct net *net,
4402                                      struct fib6_table *table)
4403 {
4404         struct fib6_info *rt;
4405
4406 restart:
4407         rcu_read_lock();
4408         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4409                 struct net_device *dev = fib6_info_nh_dev(rt);
4410                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4411
4412                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4413                     (!idev || idev->cnf.accept_ra != 2) &&
4414                     fib6_info_hold_safe(rt)) {
4415                         rcu_read_unlock();
4416                         ip6_del_rt(net, rt, false);
4417                         goto restart;
4418                 }
4419         }
4420         rcu_read_unlock();
4421
4422         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4423 }
4424
4425 void rt6_purge_dflt_routers(struct net *net)
4426 {
4427         struct fib6_table *table;
4428         struct hlist_head *head;
4429         unsigned int h;
4430
4431         rcu_read_lock();
4432
4433         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4434                 head = &net->ipv6.fib_table_hash[h];
4435                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4436                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4437                                 __rt6_purge_dflt_routers(net, table);
4438                 }
4439         }
4440
4441         rcu_read_unlock();
4442 }
4443
4444 static void rtmsg_to_fib6_config(struct net *net,
4445                                  struct in6_rtmsg *rtmsg,
4446                                  struct fib6_config *cfg)
4447 {
4448         *cfg = (struct fib6_config){
4449                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4450                          : RT6_TABLE_MAIN,
4451                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4452                 .fc_metric = rtmsg->rtmsg_metric,
4453                 .fc_expires = rtmsg->rtmsg_info,
4454                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4455                 .fc_src_len = rtmsg->rtmsg_src_len,
4456                 .fc_flags = rtmsg->rtmsg_flags,
4457                 .fc_type = rtmsg->rtmsg_type,
4458
4459                 .fc_nlinfo.nl_net = net,
4460
4461                 .fc_dst = rtmsg->rtmsg_dst,
4462                 .fc_src = rtmsg->rtmsg_src,
4463                 .fc_gateway = rtmsg->rtmsg_gateway,
4464         };
4465 }
4466
4467 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4468 {
4469         struct fib6_config cfg;
4470         int err;
4471
4472         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4473                 return -EINVAL;
4474         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4475                 return -EPERM;
4476
4477         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4478
4479         rtnl_lock();
4480         switch (cmd) {
4481         case SIOCADDRT:
4482                 /* Only do the default setting of fc_metric in route adding */
4483                 if (cfg.fc_metric == 0)
4484                         cfg.fc_metric = IP6_RT_PRIO_USER;
4485                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4486                 break;
4487         case SIOCDELRT:
4488                 err = ip6_route_del(&cfg, NULL);
4489                 break;
4490         }
4491         rtnl_unlock();
4492         return err;
4493 }
4494
4495 /*
4496  *      Drop the packet on the floor
4497  */
4498
4499 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4500 {
4501         struct dst_entry *dst = skb_dst(skb);
4502         struct net *net = dev_net(dst->dev);
4503         struct inet6_dev *idev;
4504         SKB_DR(reason);
4505         int type;
4506
4507         if (netif_is_l3_master(skb->dev) ||
4508             dst->dev == net->loopback_dev)
4509                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4510         else
4511                 idev = ip6_dst_idev(dst);
4512
4513         switch (ipstats_mib_noroutes) {
4514         case IPSTATS_MIB_INNOROUTES:
4515                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4516                 if (type == IPV6_ADDR_ANY) {
4517                         SKB_DR_SET(reason, IP_INADDRERRORS);
4518                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4519                         break;
4520                 }
4521                 SKB_DR_SET(reason, IP_INNOROUTES);
4522                 fallthrough;
4523         case IPSTATS_MIB_OUTNOROUTES:
4524                 SKB_DR_OR(reason, IP_OUTNOROUTES);
4525                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4526                 break;
4527         }
4528
4529         /* Start over by dropping the dst for l3mdev case */
4530         if (netif_is_l3_master(skb->dev))
4531                 skb_dst_drop(skb);
4532
4533         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4534         kfree_skb_reason(skb, reason);
4535         return 0;
4536 }
4537
4538 static int ip6_pkt_discard(struct sk_buff *skb)
4539 {
4540         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4541 }
4542
4543 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4544 {
4545         skb->dev = skb_dst(skb)->dev;
4546         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4547 }
4548
4549 static int ip6_pkt_prohibit(struct sk_buff *skb)
4550 {
4551         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4552 }
4553
4554 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4555 {
4556         skb->dev = skb_dst(skb)->dev;
4557         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4558 }
4559
4560 /*
4561  *      Allocate a dst for local (unicast / anycast) address.
4562  */
4563
4564 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4565                                      struct inet6_dev *idev,
4566                                      const struct in6_addr *addr,
4567                                      bool anycast, gfp_t gfp_flags,
4568                                      struct netlink_ext_ack *extack)
4569 {
4570         struct fib6_config cfg = {
4571                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4572                 .fc_ifindex = idev->dev->ifindex,
4573                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4574                 .fc_dst = *addr,
4575                 .fc_dst_len = 128,
4576                 .fc_protocol = RTPROT_KERNEL,
4577                 .fc_nlinfo.nl_net = net,
4578                 .fc_ignore_dev_down = true,
4579         };
4580         struct fib6_info *f6i;
4581
4582         if (anycast) {
4583                 cfg.fc_type = RTN_ANYCAST;
4584                 cfg.fc_flags |= RTF_ANYCAST;
4585         } else {
4586                 cfg.fc_type = RTN_LOCAL;
4587                 cfg.fc_flags |= RTF_LOCAL;
4588         }
4589
4590         f6i = ip6_route_info_create(&cfg, gfp_flags, extack);
4591         if (!IS_ERR(f6i)) {
4592                 f6i->dst_nocount = true;
4593
4594                 if (!anycast &&
4595                     (READ_ONCE(net->ipv6.devconf_all->disable_policy) ||
4596                      READ_ONCE(idev->cnf.disable_policy)))
4597                         f6i->dst_nopolicy = true;
4598         }
4599
4600         return f6i;
4601 }
4602
4603 /* remove deleted ip from prefsrc entries */
4604 struct arg_dev_net_ip {
4605         struct net *net;
4606         struct in6_addr *addr;
4607 };
4608
4609 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4610 {
4611         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4612         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4613
4614         if (!rt->nh &&
4615             rt != net->ipv6.fib6_null_entry &&
4616             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr) &&
4617             !ipv6_chk_addr(net, addr, rt->fib6_nh->fib_nh_dev, 0)) {
4618                 spin_lock_bh(&rt6_exception_lock);
4619                 /* remove prefsrc entry */
4620                 rt->fib6_prefsrc.plen = 0;
4621                 spin_unlock_bh(&rt6_exception_lock);
4622         }
4623         return 0;
4624 }
4625
4626 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4627 {
4628         struct net *net = dev_net(ifp->idev->dev);
4629         struct arg_dev_net_ip adni = {
4630                 .net = net,
4631                 .addr = &ifp->addr,
4632         };
4633         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4634 }
4635
4636 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4637
4638 /* Remove routers and update dst entries when gateway turn into host. */
4639 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4640 {
4641         struct in6_addr *gateway = (struct in6_addr *)arg;
4642         struct fib6_nh *nh;
4643
4644         /* RA routes do not use nexthops */
4645         if (rt->nh)
4646                 return 0;
4647
4648         nh = rt->fib6_nh;
4649         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4650             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4651                 return -1;
4652
4653         /* Further clean up cached routes in exception table.
4654          * This is needed because cached route may have a different
4655          * gateway than its 'parent' in the case of an ip redirect.
4656          */
4657         fib6_nh_exceptions_clean_tohost(nh, gateway);
4658
4659         return 0;
4660 }
4661
4662 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4663 {
4664         fib6_clean_all(net, fib6_clean_tohost, gateway);
4665 }
4666
4667 struct arg_netdev_event {
4668         const struct net_device *dev;
4669         union {
4670                 unsigned char nh_flags;
4671                 unsigned long event;
4672         };
4673 };
4674
4675 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4676 {
4677         struct fib6_info *iter;
4678         struct fib6_node *fn;
4679
4680         fn = rcu_dereference_protected(rt->fib6_node,
4681                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4682         iter = rcu_dereference_protected(fn->leaf,
4683                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4684         while (iter) {
4685                 if (iter->fib6_metric == rt->fib6_metric &&
4686                     rt6_qualify_for_ecmp(iter))
4687                         return iter;
4688                 iter = rcu_dereference_protected(iter->fib6_next,
4689                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4690         }
4691
4692         return NULL;
4693 }
4694
4695 /* only called for fib entries with builtin fib6_nh */
4696 static bool rt6_is_dead(const struct fib6_info *rt)
4697 {
4698         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4699             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4700              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4701                 return true;
4702
4703         return false;
4704 }
4705
4706 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4707 {
4708         struct fib6_info *iter;
4709         int total = 0;
4710
4711         if (!rt6_is_dead(rt))
4712                 total += rt->fib6_nh->fib_nh_weight;
4713
4714         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4715                 if (!rt6_is_dead(iter))
4716                         total += iter->fib6_nh->fib_nh_weight;
4717         }
4718
4719         return total;
4720 }
4721
4722 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4723 {
4724         int upper_bound = -1;
4725
4726         if (!rt6_is_dead(rt)) {
4727                 *weight += rt->fib6_nh->fib_nh_weight;
4728                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4729                                                     total) - 1;
4730         }
4731         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4732 }
4733
4734 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4735 {
4736         struct fib6_info *iter;
4737         int weight = 0;
4738
4739         rt6_upper_bound_set(rt, &weight, total);
4740
4741         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4742                 rt6_upper_bound_set(iter, &weight, total);
4743 }
4744
4745 void rt6_multipath_rebalance(struct fib6_info *rt)
4746 {
4747         struct fib6_info *first;
4748         int total;
4749
4750         /* In case the entire multipath route was marked for flushing,
4751          * then there is no need to rebalance upon the removal of every
4752          * sibling route.
4753          */
4754         if (!rt->fib6_nsiblings || rt->should_flush)
4755                 return;
4756
4757         /* During lookup routes are evaluated in order, so we need to
4758          * make sure upper bounds are assigned from the first sibling
4759          * onwards.
4760          */
4761         first = rt6_multipath_first_sibling(rt);
4762         if (WARN_ON_ONCE(!first))
4763                 return;
4764
4765         total = rt6_multipath_total_weight(first);
4766         rt6_multipath_upper_bound_set(first, total);
4767 }
4768
4769 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4770 {
4771         const struct arg_netdev_event *arg = p_arg;
4772         struct net *net = dev_net(arg->dev);
4773
4774         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4775             rt->fib6_nh->fib_nh_dev == arg->dev) {
4776                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4777                 fib6_update_sernum_upto_root(net, rt);
4778                 rt6_multipath_rebalance(rt);
4779         }
4780
4781         return 0;
4782 }
4783
4784 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4785 {
4786         struct arg_netdev_event arg = {
4787                 .dev = dev,
4788                 {
4789                         .nh_flags = nh_flags,
4790                 },
4791         };
4792
4793         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4794                 arg.nh_flags |= RTNH_F_LINKDOWN;
4795
4796         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4797 }
4798
4799 /* only called for fib entries with inline fib6_nh */
4800 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4801                                    const struct net_device *dev)
4802 {
4803         struct fib6_info *iter;
4804
4805         if (rt->fib6_nh->fib_nh_dev == dev)
4806                 return true;
4807         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4808                 if (iter->fib6_nh->fib_nh_dev == dev)
4809                         return true;
4810
4811         return false;
4812 }
4813
4814 static void rt6_multipath_flush(struct fib6_info *rt)
4815 {
4816         struct fib6_info *iter;
4817
4818         rt->should_flush = 1;
4819         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4820                 iter->should_flush = 1;
4821 }
4822
4823 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4824                                              const struct net_device *down_dev)
4825 {
4826         struct fib6_info *iter;
4827         unsigned int dead = 0;
4828
4829         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4830             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4831                 dead++;
4832         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4833                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4834                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4835                         dead++;
4836
4837         return dead;
4838 }
4839
4840 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4841                                        const struct net_device *dev,
4842                                        unsigned char nh_flags)
4843 {
4844         struct fib6_info *iter;
4845
4846         if (rt->fib6_nh->fib_nh_dev == dev)
4847                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4848         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4849                 if (iter->fib6_nh->fib_nh_dev == dev)
4850                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4851 }
4852
4853 /* called with write lock held for table with rt */
4854 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4855 {
4856         const struct arg_netdev_event *arg = p_arg;
4857         const struct net_device *dev = arg->dev;
4858         struct net *net = dev_net(dev);
4859
4860         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4861                 return 0;
4862
4863         switch (arg->event) {
4864         case NETDEV_UNREGISTER:
4865                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4866         case NETDEV_DOWN:
4867                 if (rt->should_flush)
4868                         return -1;
4869                 if (!rt->fib6_nsiblings)
4870                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4871                 if (rt6_multipath_uses_dev(rt, dev)) {
4872                         unsigned int count;
4873
4874                         count = rt6_multipath_dead_count(rt, dev);
4875                         if (rt->fib6_nsiblings + 1 == count) {
4876                                 rt6_multipath_flush(rt);
4877                                 return -1;
4878                         }
4879                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4880                                                    RTNH_F_LINKDOWN);
4881                         fib6_update_sernum(net, rt);
4882                         rt6_multipath_rebalance(rt);
4883                 }
4884                 return -2;
4885         case NETDEV_CHANGE:
4886                 if (rt->fib6_nh->fib_nh_dev != dev ||
4887                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4888                         break;
4889                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4890                 rt6_multipath_rebalance(rt);
4891                 break;
4892         }
4893
4894         return 0;
4895 }
4896
4897 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4898 {
4899         struct arg_netdev_event arg = {
4900                 .dev = dev,
4901                 {
4902                         .event = event,
4903                 },
4904         };
4905         struct net *net = dev_net(dev);
4906
4907         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4908                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4909         else
4910                 fib6_clean_all(net, fib6_ifdown, &arg);
4911 }
4912
4913 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4914 {
4915         rt6_sync_down_dev(dev, event);
4916         rt6_uncached_list_flush_dev(dev);
4917         neigh_ifdown(&nd_tbl, dev);
4918 }
4919
4920 struct rt6_mtu_change_arg {
4921         struct net_device *dev;
4922         unsigned int mtu;
4923         struct fib6_info *f6i;
4924 };
4925
4926 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4927 {
4928         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4929         struct fib6_info *f6i = arg->f6i;
4930
4931         /* For administrative MTU increase, there is no way to discover
4932          * IPv6 PMTU increase, so PMTU increase should be updated here.
4933          * Since RFC 1981 doesn't include administrative MTU increase
4934          * update PMTU increase is a MUST. (i.e. jumbo frame)
4935          */
4936         if (nh->fib_nh_dev == arg->dev) {
4937                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4938                 u32 mtu = f6i->fib6_pmtu;
4939
4940                 if (mtu >= arg->mtu ||
4941                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4942                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4943
4944                 spin_lock_bh(&rt6_exception_lock);
4945                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4946                 spin_unlock_bh(&rt6_exception_lock);
4947         }
4948
4949         return 0;
4950 }
4951
4952 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4953 {
4954         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4955         struct inet6_dev *idev;
4956
4957         /* In IPv6 pmtu discovery is not optional,
4958            so that RTAX_MTU lock cannot disable it.
4959            We still use this lock to block changes
4960            caused by addrconf/ndisc.
4961         */
4962
4963         idev = __in6_dev_get(arg->dev);
4964         if (!idev)
4965                 return 0;
4966
4967         if (fib6_metric_locked(f6i, RTAX_MTU))
4968                 return 0;
4969
4970         arg->f6i = f6i;
4971         if (f6i->nh) {
4972                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4973                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4974                                                 arg);
4975         }
4976
4977         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4978 }
4979
4980 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4981 {
4982         struct rt6_mtu_change_arg arg = {
4983                 .dev = dev,
4984                 .mtu = mtu,
4985         };
4986
4987         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4988 }
4989
4990 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4991         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4992         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4993         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4994         [RTA_OIF]               = { .type = NLA_U32 },
4995         [RTA_IIF]               = { .type = NLA_U32 },
4996         [RTA_PRIORITY]          = { .type = NLA_U32 },
4997         [RTA_METRICS]           = { .type = NLA_NESTED },
4998         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4999         [RTA_PREF]              = { .type = NLA_U8 },
5000         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
5001         [RTA_ENCAP]             = { .type = NLA_NESTED },
5002         [RTA_EXPIRES]           = { .type = NLA_U32 },
5003         [RTA_UID]               = { .type = NLA_U32 },
5004         [RTA_MARK]              = { .type = NLA_U32 },
5005         [RTA_TABLE]             = { .type = NLA_U32 },
5006         [RTA_IP_PROTO]          = { .type = NLA_U8 },
5007         [RTA_SPORT]             = { .type = NLA_U16 },
5008         [RTA_DPORT]             = { .type = NLA_U16 },
5009         [RTA_NH_ID]             = { .type = NLA_U32 },
5010 };
5011
5012 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
5013                               struct fib6_config *cfg,
5014                               struct netlink_ext_ack *extack)
5015 {
5016         struct rtmsg *rtm;
5017         struct nlattr *tb[RTA_MAX+1];
5018         unsigned int pref;
5019         int err;
5020
5021         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5022                                      rtm_ipv6_policy, extack);
5023         if (err < 0)
5024                 goto errout;
5025
5026         err = -EINVAL;
5027         rtm = nlmsg_data(nlh);
5028
5029         if (rtm->rtm_tos) {
5030                 NL_SET_ERR_MSG(extack,
5031                                "Invalid dsfield (tos): option not available for IPv6");
5032                 goto errout;
5033         }
5034
5035         *cfg = (struct fib6_config){
5036                 .fc_table = rtm->rtm_table,
5037                 .fc_dst_len = rtm->rtm_dst_len,
5038                 .fc_src_len = rtm->rtm_src_len,
5039                 .fc_flags = RTF_UP,
5040                 .fc_protocol = rtm->rtm_protocol,
5041                 .fc_type = rtm->rtm_type,
5042
5043                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5044                 .fc_nlinfo.nlh = nlh,
5045                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5046         };
5047
5048         if (rtm->rtm_type == RTN_UNREACHABLE ||
5049             rtm->rtm_type == RTN_BLACKHOLE ||
5050             rtm->rtm_type == RTN_PROHIBIT ||
5051             rtm->rtm_type == RTN_THROW)
5052                 cfg->fc_flags |= RTF_REJECT;
5053
5054         if (rtm->rtm_type == RTN_LOCAL)
5055                 cfg->fc_flags |= RTF_LOCAL;
5056
5057         if (rtm->rtm_flags & RTM_F_CLONED)
5058                 cfg->fc_flags |= RTF_CACHE;
5059
5060         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5061
5062         if (tb[RTA_NH_ID]) {
5063                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5064                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5065                         NL_SET_ERR_MSG(extack,
5066                                        "Nexthop specification and nexthop id are mutually exclusive");
5067                         goto errout;
5068                 }
5069                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5070         }
5071
5072         if (tb[RTA_GATEWAY]) {
5073                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5074                 cfg->fc_flags |= RTF_GATEWAY;
5075         }
5076         if (tb[RTA_VIA]) {
5077                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5078                 goto errout;
5079         }
5080
5081         if (tb[RTA_DST]) {
5082                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5083
5084                 if (nla_len(tb[RTA_DST]) < plen)
5085                         goto errout;
5086
5087                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5088         }
5089
5090         if (tb[RTA_SRC]) {
5091                 int plen = (rtm->rtm_src_len + 7) >> 3;
5092
5093                 if (nla_len(tb[RTA_SRC]) < plen)
5094                         goto errout;
5095
5096                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5097         }
5098
5099         if (tb[RTA_PREFSRC])
5100                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5101
5102         if (tb[RTA_OIF])
5103                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5104
5105         if (tb[RTA_PRIORITY])
5106                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5107
5108         if (tb[RTA_METRICS]) {
5109                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5110                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5111         }
5112
5113         if (tb[RTA_TABLE])
5114                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5115
5116         if (tb[RTA_MULTIPATH]) {
5117                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5118                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5119
5120                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5121                                                      cfg->fc_mp_len, extack);
5122                 if (err < 0)
5123                         goto errout;
5124         }
5125
5126         if (tb[RTA_PREF]) {
5127                 pref = nla_get_u8(tb[RTA_PREF]);
5128                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5129                     pref != ICMPV6_ROUTER_PREF_HIGH)
5130                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5131                 cfg->fc_flags |= RTF_PREF(pref);
5132         }
5133
5134         if (tb[RTA_ENCAP])
5135                 cfg->fc_encap = tb[RTA_ENCAP];
5136
5137         if (tb[RTA_ENCAP_TYPE]) {
5138                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5139
5140                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5141                 if (err < 0)
5142                         goto errout;
5143         }
5144
5145         if (tb[RTA_EXPIRES]) {
5146                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5147
5148                 if (addrconf_finite_timeout(timeout)) {
5149                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5150                         cfg->fc_flags |= RTF_EXPIRES;
5151                 }
5152         }
5153
5154         err = 0;
5155 errout:
5156         return err;
5157 }
5158
5159 struct rt6_nh {
5160         struct fib6_info *fib6_info;
5161         struct fib6_config r_cfg;
5162         struct list_head next;
5163 };
5164
5165 static int ip6_route_info_append(struct net *net,
5166                                  struct list_head *rt6_nh_list,
5167                                  struct fib6_info *rt,
5168                                  struct fib6_config *r_cfg)
5169 {
5170         struct rt6_nh *nh;
5171         int err = -EEXIST;
5172
5173         list_for_each_entry(nh, rt6_nh_list, next) {
5174                 /* check if fib6_info already exists */
5175                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5176                         return err;
5177         }
5178
5179         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5180         if (!nh)
5181                 return -ENOMEM;
5182         nh->fib6_info = rt;
5183         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5184         list_add_tail(&nh->next, rt6_nh_list);
5185
5186         return 0;
5187 }
5188
5189 static void ip6_route_mpath_notify(struct fib6_info *rt,
5190                                    struct fib6_info *rt_last,
5191                                    struct nl_info *info,
5192                                    __u16 nlflags)
5193 {
5194         /* if this is an APPEND route, then rt points to the first route
5195          * inserted and rt_last points to last route inserted. Userspace
5196          * wants a consistent dump of the route which starts at the first
5197          * nexthop. Since sibling routes are always added at the end of
5198          * the list, find the first sibling of the last route appended
5199          */
5200         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5201                 rt = list_first_entry(&rt_last->fib6_siblings,
5202                                       struct fib6_info,
5203                                       fib6_siblings);
5204         }
5205
5206         if (rt)
5207                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5208 }
5209
5210 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5211 {
5212         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5213         bool should_notify = false;
5214         struct fib6_info *leaf;
5215         struct fib6_node *fn;
5216
5217         rcu_read_lock();
5218         fn = rcu_dereference(rt->fib6_node);
5219         if (!fn)
5220                 goto out;
5221
5222         leaf = rcu_dereference(fn->leaf);
5223         if (!leaf)
5224                 goto out;
5225
5226         if (rt == leaf ||
5227             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5228              rt6_qualify_for_ecmp(leaf)))
5229                 should_notify = true;
5230 out:
5231         rcu_read_unlock();
5232
5233         return should_notify;
5234 }
5235
5236 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5237                              struct netlink_ext_ack *extack)
5238 {
5239         if (nla_len(nla) < sizeof(*gw)) {
5240                 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5241                 return -EINVAL;
5242         }
5243
5244         *gw = nla_get_in6_addr(nla);
5245
5246         return 0;
5247 }
5248
5249 static int ip6_route_multipath_add(struct fib6_config *cfg,
5250                                    struct netlink_ext_ack *extack)
5251 {
5252         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5253         struct nl_info *info = &cfg->fc_nlinfo;
5254         struct fib6_config r_cfg;
5255         struct rtnexthop *rtnh;
5256         struct fib6_info *rt;
5257         struct rt6_nh *err_nh;
5258         struct rt6_nh *nh, *nh_safe;
5259         __u16 nlflags;
5260         int remaining;
5261         int attrlen;
5262         int err = 1;
5263         int nhn = 0;
5264         int replace = (cfg->fc_nlinfo.nlh &&
5265                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5266         LIST_HEAD(rt6_nh_list);
5267
5268         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5269         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5270                 nlflags |= NLM_F_APPEND;
5271
5272         remaining = cfg->fc_mp_len;
5273         rtnh = (struct rtnexthop *)cfg->fc_mp;
5274
5275         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5276          * fib6_info structs per nexthop
5277          */
5278         while (rtnh_ok(rtnh, remaining)) {
5279                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5280                 if (rtnh->rtnh_ifindex)
5281                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5282
5283                 attrlen = rtnh_attrlen(rtnh);
5284                 if (attrlen > 0) {
5285                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5286
5287                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5288                         if (nla) {
5289                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5290                                                         extack);
5291                                 if (err)
5292                                         goto cleanup;
5293
5294                                 r_cfg.fc_flags |= RTF_GATEWAY;
5295                         }
5296                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5297
5298                         /* RTA_ENCAP_TYPE length checked in
5299                          * lwtunnel_valid_encap_type_attr
5300                          */
5301                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5302                         if (nla)
5303                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5304                 }
5305
5306                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5307                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5308                 if (IS_ERR(rt)) {
5309                         err = PTR_ERR(rt);
5310                         rt = NULL;
5311                         goto cleanup;
5312                 }
5313                 if (!rt6_qualify_for_ecmp(rt)) {
5314                         err = -EINVAL;
5315                         NL_SET_ERR_MSG(extack,
5316                                        "Device only routes can not be added for IPv6 using the multipath API.");
5317                         fib6_info_release(rt);
5318                         goto cleanup;
5319                 }
5320
5321                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5322
5323                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5324                                             rt, &r_cfg);
5325                 if (err) {
5326                         fib6_info_release(rt);
5327                         goto cleanup;
5328                 }
5329
5330                 rtnh = rtnh_next(rtnh, &remaining);
5331         }
5332
5333         if (list_empty(&rt6_nh_list)) {
5334                 NL_SET_ERR_MSG(extack,
5335                                "Invalid nexthop configuration - no valid nexthops");
5336                 return -EINVAL;
5337         }
5338
5339         /* for add and replace send one notification with all nexthops.
5340          * Skip the notification in fib6_add_rt2node and send one with
5341          * the full route when done
5342          */
5343         info->skip_notify = 1;
5344
5345         /* For add and replace, send one notification with all nexthops. For
5346          * append, send one notification with all appended nexthops.
5347          */
5348         info->skip_notify_kernel = 1;
5349
5350         err_nh = NULL;
5351         list_for_each_entry(nh, &rt6_nh_list, next) {
5352                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5353
5354                 if (err) {
5355                         if (replace && nhn)
5356                                 NL_SET_ERR_MSG_MOD(extack,
5357                                                    "multipath route replace failed (check consistency of installed routes)");
5358                         err_nh = nh;
5359                         goto add_errout;
5360                 }
5361                 /* save reference to last route successfully inserted */
5362                 rt_last = nh->fib6_info;
5363
5364                 /* save reference to first route for notification */
5365                 if (!rt_notif)
5366                         rt_notif = nh->fib6_info;
5367
5368                 /* Because each route is added like a single route we remove
5369                  * these flags after the first nexthop: if there is a collision,
5370                  * we have already failed to add the first nexthop:
5371                  * fib6_add_rt2node() has rejected it; when replacing, old
5372                  * nexthops have been replaced by first new, the rest should
5373                  * be added to it.
5374                  */
5375                 if (cfg->fc_nlinfo.nlh) {
5376                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5377                                                              NLM_F_REPLACE);
5378                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5379                 }
5380                 nhn++;
5381         }
5382
5383         /* An in-kernel notification should only be sent in case the new
5384          * multipath route is added as the first route in the node, or if
5385          * it was appended to it. We pass 'rt_notif' since it is the first
5386          * sibling and might allow us to skip some checks in the replace case.
5387          */
5388         if (ip6_route_mpath_should_notify(rt_notif)) {
5389                 enum fib_event_type fib_event;
5390
5391                 if (rt_notif->fib6_nsiblings != nhn - 1)
5392                         fib_event = FIB_EVENT_ENTRY_APPEND;
5393                 else
5394                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5395
5396                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5397                                                           fib_event, rt_notif,
5398                                                           nhn - 1, extack);
5399                 if (err) {
5400                         /* Delete all the siblings that were just added */
5401                         err_nh = NULL;
5402                         goto add_errout;
5403                 }
5404         }
5405
5406         /* success ... tell user about new route */
5407         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5408         goto cleanup;
5409
5410 add_errout:
5411         /* send notification for routes that were added so that
5412          * the delete notifications sent by ip6_route_del are
5413          * coherent
5414          */
5415         if (rt_notif)
5416                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5417
5418         /* Delete routes that were already added */
5419         list_for_each_entry(nh, &rt6_nh_list, next) {
5420                 if (err_nh == nh)
5421                         break;
5422                 ip6_route_del(&nh->r_cfg, extack);
5423         }
5424
5425 cleanup:
5426         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5427                 fib6_info_release(nh->fib6_info);
5428                 list_del(&nh->next);
5429                 kfree(nh);
5430         }
5431
5432         return err;
5433 }
5434
5435 static int ip6_route_multipath_del(struct fib6_config *cfg,
5436                                    struct netlink_ext_ack *extack)
5437 {
5438         struct fib6_config r_cfg;
5439         struct rtnexthop *rtnh;
5440         int last_err = 0;
5441         int remaining;
5442         int attrlen;
5443         int err;
5444
5445         remaining = cfg->fc_mp_len;
5446         rtnh = (struct rtnexthop *)cfg->fc_mp;
5447
5448         /* Parse a Multipath Entry */
5449         while (rtnh_ok(rtnh, remaining)) {
5450                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5451                 if (rtnh->rtnh_ifindex)
5452                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5453
5454                 attrlen = rtnh_attrlen(rtnh);
5455                 if (attrlen > 0) {
5456                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5457
5458                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5459                         if (nla) {
5460                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5461                                                         extack);
5462                                 if (err) {
5463                                         last_err = err;
5464                                         goto next_rtnh;
5465                                 }
5466
5467                                 r_cfg.fc_flags |= RTF_GATEWAY;
5468                         }
5469                 }
5470                 err = ip6_route_del(&r_cfg, extack);
5471                 if (err)
5472                         last_err = err;
5473
5474 next_rtnh:
5475                 rtnh = rtnh_next(rtnh, &remaining);
5476         }
5477
5478         return last_err;
5479 }
5480
5481 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5482                               struct netlink_ext_ack *extack)
5483 {
5484         struct fib6_config cfg;
5485         int err;
5486
5487         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5488         if (err < 0)
5489                 return err;
5490
5491         if (cfg.fc_nh_id &&
5492             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5493                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5494                 return -EINVAL;
5495         }
5496
5497         if (cfg.fc_mp)
5498                 return ip6_route_multipath_del(&cfg, extack);
5499         else {
5500                 cfg.fc_delete_all_nh = 1;
5501                 return ip6_route_del(&cfg, extack);
5502         }
5503 }
5504
5505 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5506                               struct netlink_ext_ack *extack)
5507 {
5508         struct fib6_config cfg;
5509         int err;
5510
5511         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5512         if (err < 0)
5513                 return err;
5514
5515         if (cfg.fc_metric == 0)
5516                 cfg.fc_metric = IP6_RT_PRIO_USER;
5517
5518         if (cfg.fc_mp)
5519                 return ip6_route_multipath_add(&cfg, extack);
5520         else
5521                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5522 }
5523
5524 /* add the overhead of this fib6_nh to nexthop_len */
5525 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5526 {
5527         int *nexthop_len = arg;
5528
5529         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5530                      + NLA_ALIGN(sizeof(struct rtnexthop))
5531                      + nla_total_size(16); /* RTA_GATEWAY */
5532
5533         if (nh->fib_nh_lws) {
5534                 /* RTA_ENCAP_TYPE */
5535                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5536                 /* RTA_ENCAP */
5537                 *nexthop_len += nla_total_size(2);
5538         }
5539
5540         return 0;
5541 }
5542
5543 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5544 {
5545         int nexthop_len;
5546
5547         if (f6i->nh) {
5548                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5549                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5550                                          &nexthop_len);
5551         } else {
5552                 struct fib6_info *sibling, *next_sibling;
5553                 struct fib6_nh *nh = f6i->fib6_nh;
5554
5555                 nexthop_len = 0;
5556                 if (f6i->fib6_nsiblings) {
5557                         rt6_nh_nlmsg_size(nh, &nexthop_len);
5558
5559                         list_for_each_entry_safe(sibling, next_sibling,
5560                                                  &f6i->fib6_siblings, fib6_siblings) {
5561                                 rt6_nh_nlmsg_size(sibling->fib6_nh, &nexthop_len);
5562                         }
5563                 }
5564                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5565         }
5566
5567         return NLMSG_ALIGN(sizeof(struct rtmsg))
5568                + nla_total_size(16) /* RTA_SRC */
5569                + nla_total_size(16) /* RTA_DST */
5570                + nla_total_size(16) /* RTA_GATEWAY */
5571                + nla_total_size(16) /* RTA_PREFSRC */
5572                + nla_total_size(4) /* RTA_TABLE */
5573                + nla_total_size(4) /* RTA_IIF */
5574                + nla_total_size(4) /* RTA_OIF */
5575                + nla_total_size(4) /* RTA_PRIORITY */
5576                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5577                + nla_total_size(sizeof(struct rta_cacheinfo))
5578                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5579                + nla_total_size(1) /* RTA_PREF */
5580                + nexthop_len;
5581 }
5582
5583 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5584                                  unsigned char *flags)
5585 {
5586         if (nexthop_is_multipath(nh)) {
5587                 struct nlattr *mp;
5588
5589                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5590                 if (!mp)
5591                         goto nla_put_failure;
5592
5593                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5594                         goto nla_put_failure;
5595
5596                 nla_nest_end(skb, mp);
5597         } else {
5598                 struct fib6_nh *fib6_nh;
5599
5600                 fib6_nh = nexthop_fib6_nh(nh);
5601                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5602                                      flags, false) < 0)
5603                         goto nla_put_failure;
5604         }
5605
5606         return 0;
5607
5608 nla_put_failure:
5609         return -EMSGSIZE;
5610 }
5611
5612 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5613                          struct fib6_info *rt, struct dst_entry *dst,
5614                          struct in6_addr *dest, struct in6_addr *src,
5615                          int iif, int type, u32 portid, u32 seq,
5616                          unsigned int flags)
5617 {
5618         struct rt6_info *rt6 = dst_rt6_info(dst);
5619         struct rt6key *rt6_dst, *rt6_src;
5620         u32 *pmetrics, table, rt6_flags;
5621         unsigned char nh_flags = 0;
5622         struct nlmsghdr *nlh;
5623         struct rtmsg *rtm;
5624         long expires = 0;
5625
5626         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5627         if (!nlh)
5628                 return -EMSGSIZE;
5629
5630         if (rt6) {
5631                 rt6_dst = &rt6->rt6i_dst;
5632                 rt6_src = &rt6->rt6i_src;
5633                 rt6_flags = rt6->rt6i_flags;
5634         } else {
5635                 rt6_dst = &rt->fib6_dst;
5636                 rt6_src = &rt->fib6_src;
5637                 rt6_flags = rt->fib6_flags;
5638         }
5639
5640         rtm = nlmsg_data(nlh);
5641         rtm->rtm_family = AF_INET6;
5642         rtm->rtm_dst_len = rt6_dst->plen;
5643         rtm->rtm_src_len = rt6_src->plen;
5644         rtm->rtm_tos = 0;
5645         if (rt->fib6_table)
5646                 table = rt->fib6_table->tb6_id;
5647         else
5648                 table = RT6_TABLE_UNSPEC;
5649         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5650         if (nla_put_u32(skb, RTA_TABLE, table))
5651                 goto nla_put_failure;
5652
5653         rtm->rtm_type = rt->fib6_type;
5654         rtm->rtm_flags = 0;
5655         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5656         rtm->rtm_protocol = rt->fib6_protocol;
5657
5658         if (rt6_flags & RTF_CACHE)
5659                 rtm->rtm_flags |= RTM_F_CLONED;
5660
5661         if (dest) {
5662                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5663                         goto nla_put_failure;
5664                 rtm->rtm_dst_len = 128;
5665         } else if (rtm->rtm_dst_len)
5666                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5667                         goto nla_put_failure;
5668 #ifdef CONFIG_IPV6_SUBTREES
5669         if (src) {
5670                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5671                         goto nla_put_failure;
5672                 rtm->rtm_src_len = 128;
5673         } else if (rtm->rtm_src_len &&
5674                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5675                 goto nla_put_failure;
5676 #endif
5677         if (iif) {
5678 #ifdef CONFIG_IPV6_MROUTE
5679                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5680                         int err = ip6mr_get_route(net, skb, rtm, portid);
5681
5682                         if (err == 0)
5683                                 return 0;
5684                         if (err < 0)
5685                                 goto nla_put_failure;
5686                 } else
5687 #endif
5688                         if (nla_put_u32(skb, RTA_IIF, iif))
5689                                 goto nla_put_failure;
5690         } else if (dest) {
5691                 struct in6_addr saddr_buf;
5692                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5693                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5694                         goto nla_put_failure;
5695         }
5696
5697         if (rt->fib6_prefsrc.plen) {
5698                 struct in6_addr saddr_buf;
5699                 saddr_buf = rt->fib6_prefsrc.addr;
5700                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5701                         goto nla_put_failure;
5702         }
5703
5704         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5705         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5706                 goto nla_put_failure;
5707
5708         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5709                 goto nla_put_failure;
5710
5711         /* For multipath routes, walk the siblings list and add
5712          * each as a nexthop within RTA_MULTIPATH.
5713          */
5714         if (rt6) {
5715                 if (rt6_flags & RTF_GATEWAY &&
5716                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5717                         goto nla_put_failure;
5718
5719                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5720                         goto nla_put_failure;
5721
5722                 if (dst->lwtstate &&
5723                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5724                         goto nla_put_failure;
5725         } else if (rt->fib6_nsiblings) {
5726                 struct fib6_info *sibling, *next_sibling;
5727                 struct nlattr *mp;
5728
5729                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5730                 if (!mp)
5731                         goto nla_put_failure;
5732
5733                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5734                                     rt->fib6_nh->fib_nh_weight, AF_INET6,
5735                                     0) < 0)
5736                         goto nla_put_failure;
5737
5738                 list_for_each_entry_safe(sibling, next_sibling,
5739                                          &rt->fib6_siblings, fib6_siblings) {
5740                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5741                                             sibling->fib6_nh->fib_nh_weight,
5742                                             AF_INET6, 0) < 0)
5743                                 goto nla_put_failure;
5744                 }
5745
5746                 nla_nest_end(skb, mp);
5747         } else if (rt->nh) {
5748                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5749                         goto nla_put_failure;
5750
5751                 if (nexthop_is_blackhole(rt->nh))
5752                         rtm->rtm_type = RTN_BLACKHOLE;
5753
5754                 if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5755                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5756                         goto nla_put_failure;
5757
5758                 rtm->rtm_flags |= nh_flags;
5759         } else {
5760                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5761                                      &nh_flags, false) < 0)
5762                         goto nla_put_failure;
5763
5764                 rtm->rtm_flags |= nh_flags;
5765         }
5766
5767         if (rt6_flags & RTF_EXPIRES) {
5768                 expires = dst ? dst->expires : rt->expires;
5769                 expires -= jiffies;
5770         }
5771
5772         if (!dst) {
5773                 if (READ_ONCE(rt->offload))
5774                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5775                 if (READ_ONCE(rt->trap))
5776                         rtm->rtm_flags |= RTM_F_TRAP;
5777                 if (READ_ONCE(rt->offload_failed))
5778                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5779         }
5780
5781         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5782                 goto nla_put_failure;
5783
5784         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5785                 goto nla_put_failure;
5786
5787
5788         nlmsg_end(skb, nlh);
5789         return 0;
5790
5791 nla_put_failure:
5792         nlmsg_cancel(skb, nlh);
5793         return -EMSGSIZE;
5794 }
5795
5796 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5797 {
5798         const struct net_device *dev = arg;
5799
5800         if (nh->fib_nh_dev == dev)
5801                 return 1;
5802
5803         return 0;
5804 }
5805
5806 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5807                                const struct net_device *dev)
5808 {
5809         if (f6i->nh) {
5810                 struct net_device *_dev = (struct net_device *)dev;
5811
5812                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5813                                                   fib6_info_nh_uses_dev,
5814                                                   _dev);
5815         }
5816
5817         if (f6i->fib6_nh->fib_nh_dev == dev)
5818                 return true;
5819
5820         if (f6i->fib6_nsiblings) {
5821                 struct fib6_info *sibling, *next_sibling;
5822
5823                 list_for_each_entry_safe(sibling, next_sibling,
5824                                          &f6i->fib6_siblings, fib6_siblings) {
5825                         if (sibling->fib6_nh->fib_nh_dev == dev)
5826                                 return true;
5827                 }
5828         }
5829
5830         return false;
5831 }
5832
5833 struct fib6_nh_exception_dump_walker {
5834         struct rt6_rtnl_dump_arg *dump;
5835         struct fib6_info *rt;
5836         unsigned int flags;
5837         unsigned int skip;
5838         unsigned int count;
5839 };
5840
5841 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5842 {
5843         struct fib6_nh_exception_dump_walker *w = arg;
5844         struct rt6_rtnl_dump_arg *dump = w->dump;
5845         struct rt6_exception_bucket *bucket;
5846         struct rt6_exception *rt6_ex;
5847         int i, err;
5848
5849         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5850         if (!bucket)
5851                 return 0;
5852
5853         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5854                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5855                         if (w->skip) {
5856                                 w->skip--;
5857                                 continue;
5858                         }
5859
5860                         /* Expiration of entries doesn't bump sernum, insertion
5861                          * does. Removal is triggered by insertion, so we can
5862                          * rely on the fact that if entries change between two
5863                          * partial dumps, this node is scanned again completely,
5864                          * see rt6_insert_exception() and fib6_dump_table().
5865                          *
5866                          * Count expired entries we go through as handled
5867                          * entries that we'll skip next time, in case of partial
5868                          * node dump. Otherwise, if entries expire meanwhile,
5869                          * we'll skip the wrong amount.
5870                          */
5871                         if (rt6_check_expired(rt6_ex->rt6i)) {
5872                                 w->count++;
5873                                 continue;
5874                         }
5875
5876                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5877                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5878                                             RTM_NEWROUTE,
5879                                             NETLINK_CB(dump->cb->skb).portid,
5880                                             dump->cb->nlh->nlmsg_seq, w->flags);
5881                         if (err)
5882                                 return err;
5883
5884                         w->count++;
5885                 }
5886                 bucket++;
5887         }
5888
5889         return 0;
5890 }
5891
5892 /* Return -1 if done with node, number of handled routes on partial dump */
5893 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5894 {
5895         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5896         struct fib_dump_filter *filter = &arg->filter;
5897         unsigned int flags = NLM_F_MULTI;
5898         struct net *net = arg->net;
5899         int count = 0;
5900
5901         if (rt == net->ipv6.fib6_null_entry)
5902                 return -1;
5903
5904         if ((filter->flags & RTM_F_PREFIX) &&
5905             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5906                 /* success since this is not a prefix route */
5907                 return -1;
5908         }
5909         if (filter->filter_set &&
5910             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5911              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5912              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5913                 return -1;
5914         }
5915
5916         if (filter->filter_set ||
5917             !filter->dump_routes || !filter->dump_exceptions) {
5918                 flags |= NLM_F_DUMP_FILTERED;
5919         }
5920
5921         if (filter->dump_routes) {
5922                 if (skip) {
5923                         skip--;
5924                 } else {
5925                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5926                                           0, RTM_NEWROUTE,
5927                                           NETLINK_CB(arg->cb->skb).portid,
5928                                           arg->cb->nlh->nlmsg_seq, flags)) {
5929                                 return 0;
5930                         }
5931                         count++;
5932                 }
5933         }
5934
5935         if (filter->dump_exceptions) {
5936                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5937                                                            .rt = rt,
5938                                                            .flags = flags,
5939                                                            .skip = skip,
5940                                                            .count = 0 };
5941                 int err;
5942
5943                 rcu_read_lock();
5944                 if (rt->nh) {
5945                         err = nexthop_for_each_fib6_nh(rt->nh,
5946                                                        rt6_nh_dump_exceptions,
5947                                                        &w);
5948                 } else {
5949                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5950                 }
5951                 rcu_read_unlock();
5952
5953                 if (err)
5954                         return count + w.count;
5955         }
5956
5957         return -1;
5958 }
5959
5960 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5961                                         const struct nlmsghdr *nlh,
5962                                         struct nlattr **tb,
5963                                         struct netlink_ext_ack *extack)
5964 {
5965         struct rtmsg *rtm;
5966         int i, err;
5967
5968         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5969                 NL_SET_ERR_MSG_MOD(extack,
5970                                    "Invalid header for get route request");
5971                 return -EINVAL;
5972         }
5973
5974         if (!netlink_strict_get_check(skb))
5975                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5976                                               rtm_ipv6_policy, extack);
5977
5978         rtm = nlmsg_data(nlh);
5979         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5980             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5981             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5982             rtm->rtm_type) {
5983                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5984                 return -EINVAL;
5985         }
5986         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5987                 NL_SET_ERR_MSG_MOD(extack,
5988                                    "Invalid flags for get route request");
5989                 return -EINVAL;
5990         }
5991
5992         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5993                                             rtm_ipv6_policy, extack);
5994         if (err)
5995                 return err;
5996
5997         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5998             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5999                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
6000                 return -EINVAL;
6001         }
6002
6003         for (i = 0; i <= RTA_MAX; i++) {
6004                 if (!tb[i])
6005                         continue;
6006
6007                 switch (i) {
6008                 case RTA_SRC:
6009                 case RTA_DST:
6010                 case RTA_IIF:
6011                 case RTA_OIF:
6012                 case RTA_MARK:
6013                 case RTA_UID:
6014                 case RTA_SPORT:
6015                 case RTA_DPORT:
6016                 case RTA_IP_PROTO:
6017                         break;
6018                 default:
6019                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
6020                         return -EINVAL;
6021                 }
6022         }
6023
6024         return 0;
6025 }
6026
6027 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
6028                               struct netlink_ext_ack *extack)
6029 {
6030         struct net *net = sock_net(in_skb->sk);
6031         struct nlattr *tb[RTA_MAX+1];
6032         int err, iif = 0, oif = 0;
6033         struct fib6_info *from;
6034         struct dst_entry *dst;
6035         struct rt6_info *rt;
6036         struct sk_buff *skb;
6037         struct rtmsg *rtm;
6038         struct flowi6 fl6 = {};
6039         bool fibmatch;
6040
6041         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
6042         if (err < 0)
6043                 goto errout;
6044
6045         err = -EINVAL;
6046         rtm = nlmsg_data(nlh);
6047         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6048         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6049
6050         if (tb[RTA_SRC]) {
6051                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6052                         goto errout;
6053
6054                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6055         }
6056
6057         if (tb[RTA_DST]) {
6058                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6059                         goto errout;
6060
6061                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6062         }
6063
6064         if (tb[RTA_IIF])
6065                 iif = nla_get_u32(tb[RTA_IIF]);
6066
6067         if (tb[RTA_OIF])
6068                 oif = nla_get_u32(tb[RTA_OIF]);
6069
6070         if (tb[RTA_MARK])
6071                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6072
6073         if (tb[RTA_UID])
6074                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6075                                            nla_get_u32(tb[RTA_UID]));
6076         else
6077                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6078
6079         if (tb[RTA_SPORT])
6080                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6081
6082         if (tb[RTA_DPORT])
6083                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6084
6085         if (tb[RTA_IP_PROTO]) {
6086                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6087                                                   &fl6.flowi6_proto, AF_INET6,
6088                                                   extack);
6089                 if (err)
6090                         goto errout;
6091         }
6092
6093         if (iif) {
6094                 struct net_device *dev;
6095                 int flags = 0;
6096
6097                 rcu_read_lock();
6098
6099                 dev = dev_get_by_index_rcu(net, iif);
6100                 if (!dev) {
6101                         rcu_read_unlock();
6102                         err = -ENODEV;
6103                         goto errout;
6104                 }
6105
6106                 fl6.flowi6_iif = iif;
6107
6108                 if (!ipv6_addr_any(&fl6.saddr))
6109                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6110
6111                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6112
6113                 rcu_read_unlock();
6114         } else {
6115                 fl6.flowi6_oif = oif;
6116
6117                 dst = ip6_route_output(net, NULL, &fl6);
6118         }
6119
6120
6121         rt = dst_rt6_info(dst);
6122         if (rt->dst.error) {
6123                 err = rt->dst.error;
6124                 ip6_rt_put(rt);
6125                 goto errout;
6126         }
6127
6128         if (rt == net->ipv6.ip6_null_entry) {
6129                 err = rt->dst.error;
6130                 ip6_rt_put(rt);
6131                 goto errout;
6132         }
6133
6134         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6135         if (!skb) {
6136                 ip6_rt_put(rt);
6137                 err = -ENOBUFS;
6138                 goto errout;
6139         }
6140
6141         skb_dst_set(skb, &rt->dst);
6142
6143         rcu_read_lock();
6144         from = rcu_dereference(rt->from);
6145         if (from) {
6146                 if (fibmatch)
6147                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6148                                             iif, RTM_NEWROUTE,
6149                                             NETLINK_CB(in_skb).portid,
6150                                             nlh->nlmsg_seq, 0);
6151                 else
6152                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6153                                             &fl6.saddr, iif, RTM_NEWROUTE,
6154                                             NETLINK_CB(in_skb).portid,
6155                                             nlh->nlmsg_seq, 0);
6156         } else {
6157                 err = -ENETUNREACH;
6158         }
6159         rcu_read_unlock();
6160
6161         if (err < 0) {
6162                 kfree_skb(skb);
6163                 goto errout;
6164         }
6165
6166         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6167 errout:
6168         return err;
6169 }
6170
6171 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6172                      unsigned int nlm_flags)
6173 {
6174         struct sk_buff *skb;
6175         struct net *net = info->nl_net;
6176         u32 seq;
6177         int err;
6178
6179         err = -ENOBUFS;
6180         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6181
6182         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6183         if (!skb)
6184                 goto errout;
6185
6186         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6187                             event, info->portid, seq, nlm_flags);
6188         if (err < 0) {
6189                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6190                 WARN_ON(err == -EMSGSIZE);
6191                 kfree_skb(skb);
6192                 goto errout;
6193         }
6194         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6195                     info->nlh, gfp_any());
6196         return;
6197 errout:
6198         if (err < 0)
6199                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6200 }
6201
6202 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6203                     struct nl_info *info)
6204 {
6205         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6206         struct sk_buff *skb;
6207         int err = -ENOBUFS;
6208
6209         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6210         if (!skb)
6211                 goto errout;
6212
6213         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6214                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6215         if (err < 0) {
6216                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6217                 WARN_ON(err == -EMSGSIZE);
6218                 kfree_skb(skb);
6219                 goto errout;
6220         }
6221         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6222                     info->nlh, gfp_any());
6223         return;
6224 errout:
6225         if (err < 0)
6226                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6227 }
6228
6229 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6230                             bool offload, bool trap, bool offload_failed)
6231 {
6232         struct sk_buff *skb;
6233         int err;
6234
6235         if (READ_ONCE(f6i->offload) == offload &&
6236             READ_ONCE(f6i->trap) == trap &&
6237             READ_ONCE(f6i->offload_failed) == offload_failed)
6238                 return;
6239
6240         WRITE_ONCE(f6i->offload, offload);
6241         WRITE_ONCE(f6i->trap, trap);
6242
6243         /* 2 means send notifications only if offload_failed was changed. */
6244         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6245             READ_ONCE(f6i->offload_failed) == offload_failed)
6246                 return;
6247
6248         WRITE_ONCE(f6i->offload_failed, offload_failed);
6249
6250         if (!rcu_access_pointer(f6i->fib6_node))
6251                 /* The route was removed from the tree, do not send
6252                  * notification.
6253                  */
6254                 return;
6255
6256         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6257                 return;
6258
6259         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6260         if (!skb) {
6261                 err = -ENOBUFS;
6262                 goto errout;
6263         }
6264
6265         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6266                             0, 0);
6267         if (err < 0) {
6268                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6269                 WARN_ON(err == -EMSGSIZE);
6270                 kfree_skb(skb);
6271                 goto errout;
6272         }
6273
6274         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6275         return;
6276
6277 errout:
6278         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6279 }
6280 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6281
6282 static int ip6_route_dev_notify(struct notifier_block *this,
6283                                 unsigned long event, void *ptr)
6284 {
6285         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6286         struct net *net = dev_net(dev);
6287
6288         if (!(dev->flags & IFF_LOOPBACK))
6289                 return NOTIFY_OK;
6290
6291         if (event == NETDEV_REGISTER) {
6292                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6293                 net->ipv6.ip6_null_entry->dst.dev = dev;
6294                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6295 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6296                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6297                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6298                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6299                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6300 #endif
6301          } else if (event == NETDEV_UNREGISTER &&
6302                     dev->reg_state != NETREG_UNREGISTERED) {
6303                 /* NETDEV_UNREGISTER could be fired for multiple times by
6304                  * netdev_wait_allrefs(). Make sure we only call this once.
6305                  */
6306                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6307 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6308                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6309                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6310 #endif
6311         }
6312
6313         return NOTIFY_OK;
6314 }
6315
6316 /*
6317  *      /proc
6318  */
6319
6320 #ifdef CONFIG_PROC_FS
6321 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6322 {
6323         struct net *net = (struct net *)seq->private;
6324         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6325                    net->ipv6.rt6_stats->fib_nodes,
6326                    net->ipv6.rt6_stats->fib_route_nodes,
6327                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6328                    net->ipv6.rt6_stats->fib_rt_entries,
6329                    net->ipv6.rt6_stats->fib_rt_cache,
6330                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6331                    net->ipv6.rt6_stats->fib_discarded_routes);
6332
6333         return 0;
6334 }
6335 #endif  /* CONFIG_PROC_FS */
6336
6337 #ifdef CONFIG_SYSCTL
6338
6339 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6340                               void *buffer, size_t *lenp, loff_t *ppos)
6341 {
6342         struct net *net;
6343         int delay;
6344         int ret;
6345         if (!write)
6346                 return -EINVAL;
6347
6348         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6349         if (ret)
6350                 return ret;
6351
6352         net = (struct net *)ctl->extra1;
6353         delay = net->ipv6.sysctl.flush_delay;
6354         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6355         return 0;
6356 }
6357
6358 static struct ctl_table ipv6_route_table_template[] = {
6359         {
6360                 .procname       =       "max_size",
6361                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6362                 .maxlen         =       sizeof(int),
6363                 .mode           =       0644,
6364                 .proc_handler   =       proc_dointvec,
6365         },
6366         {
6367                 .procname       =       "gc_thresh",
6368                 .data           =       &ip6_dst_ops_template.gc_thresh,
6369                 .maxlen         =       sizeof(int),
6370                 .mode           =       0644,
6371                 .proc_handler   =       proc_dointvec,
6372         },
6373         {
6374                 .procname       =       "flush",
6375                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6376                 .maxlen         =       sizeof(int),
6377                 .mode           =       0200,
6378                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6379         },
6380         {
6381                 .procname       =       "gc_min_interval",
6382                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6383                 .maxlen         =       sizeof(int),
6384                 .mode           =       0644,
6385                 .proc_handler   =       proc_dointvec_jiffies,
6386         },
6387         {
6388                 .procname       =       "gc_timeout",
6389                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6390                 .maxlen         =       sizeof(int),
6391                 .mode           =       0644,
6392                 .proc_handler   =       proc_dointvec_jiffies,
6393         },
6394         {
6395                 .procname       =       "gc_interval",
6396                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6397                 .maxlen         =       sizeof(int),
6398                 .mode           =       0644,
6399                 .proc_handler   =       proc_dointvec_jiffies,
6400         },
6401         {
6402                 .procname       =       "gc_elasticity",
6403                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6404                 .maxlen         =       sizeof(int),
6405                 .mode           =       0644,
6406                 .proc_handler   =       proc_dointvec,
6407         },
6408         {
6409                 .procname       =       "mtu_expires",
6410                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6411                 .maxlen         =       sizeof(int),
6412                 .mode           =       0644,
6413                 .proc_handler   =       proc_dointvec_jiffies,
6414         },
6415         {
6416                 .procname       =       "min_adv_mss",
6417                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6418                 .maxlen         =       sizeof(int),
6419                 .mode           =       0644,
6420                 .proc_handler   =       proc_dointvec,
6421         },
6422         {
6423                 .procname       =       "gc_min_interval_ms",
6424                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6425                 .maxlen         =       sizeof(int),
6426                 .mode           =       0644,
6427                 .proc_handler   =       proc_dointvec_ms_jiffies,
6428         },
6429         {
6430                 .procname       =       "skip_notify_on_dev_down",
6431                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6432                 .maxlen         =       sizeof(u8),
6433                 .mode           =       0644,
6434                 .proc_handler   =       proc_dou8vec_minmax,
6435                 .extra1         =       SYSCTL_ZERO,
6436                 .extra2         =       SYSCTL_ONE,
6437         },
6438 };
6439
6440 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6441 {
6442         struct ctl_table *table;
6443
6444         table = kmemdup(ipv6_route_table_template,
6445                         sizeof(ipv6_route_table_template),
6446                         GFP_KERNEL);
6447
6448         if (table) {
6449                 table[0].data = &net->ipv6.sysctl.ip6_rt_max_size;
6450                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6451                 table[2].data = &net->ipv6.sysctl.flush_delay;
6452                 table[2].extra1 = net;
6453                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6454                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6455                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6456                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6457                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6458                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6459                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6460                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6461         }
6462
6463         return table;
6464 }
6465
6466 size_t ipv6_route_sysctl_table_size(struct net *net)
6467 {
6468         /* Don't export sysctls to unprivileged users */
6469         if (net->user_ns != &init_user_ns)
6470                 return 1;
6471
6472         return ARRAY_SIZE(ipv6_route_table_template);
6473 }
6474 #endif
6475
6476 static int __net_init ip6_route_net_init(struct net *net)
6477 {
6478         int ret = -ENOMEM;
6479
6480         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6481                sizeof(net->ipv6.ip6_dst_ops));
6482
6483         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6484                 goto out_ip6_dst_ops;
6485
6486         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6487         if (!net->ipv6.fib6_null_entry)
6488                 goto out_ip6_dst_entries;
6489         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6490                sizeof(*net->ipv6.fib6_null_entry));
6491
6492         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6493                                            sizeof(*net->ipv6.ip6_null_entry),
6494                                            GFP_KERNEL);
6495         if (!net->ipv6.ip6_null_entry)
6496                 goto out_fib6_null_entry;
6497         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6498         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6499                          ip6_template_metrics, true);
6500         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->dst.rt_uncached);
6501
6502 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6503         net->ipv6.fib6_has_custom_rules = false;
6504         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6505                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6506                                                GFP_KERNEL);
6507         if (!net->ipv6.ip6_prohibit_entry)
6508                 goto out_ip6_null_entry;
6509         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6510         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6511                          ip6_template_metrics, true);
6512         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->dst.rt_uncached);
6513
6514         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6515                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6516                                                GFP_KERNEL);
6517         if (!net->ipv6.ip6_blk_hole_entry)
6518                 goto out_ip6_prohibit_entry;
6519         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6520         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6521                          ip6_template_metrics, true);
6522         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->dst.rt_uncached);
6523 #ifdef CONFIG_IPV6_SUBTREES
6524         net->ipv6.fib6_routes_require_src = 0;
6525 #endif
6526 #endif
6527
6528         net->ipv6.sysctl.flush_delay = 0;
6529         net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
6530         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6531         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6532         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6533         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6534         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6535         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6536         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6537
6538         atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6539
6540         ret = 0;
6541 out:
6542         return ret;
6543
6544 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6545 out_ip6_prohibit_entry:
6546         kfree(net->ipv6.ip6_prohibit_entry);
6547 out_ip6_null_entry:
6548         kfree(net->ipv6.ip6_null_entry);
6549 #endif
6550 out_fib6_null_entry:
6551         kfree(net->ipv6.fib6_null_entry);
6552 out_ip6_dst_entries:
6553         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6554 out_ip6_dst_ops:
6555         goto out;
6556 }
6557
6558 static void __net_exit ip6_route_net_exit(struct net *net)
6559 {
6560         kfree(net->ipv6.fib6_null_entry);
6561         kfree(net->ipv6.ip6_null_entry);
6562 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6563         kfree(net->ipv6.ip6_prohibit_entry);
6564         kfree(net->ipv6.ip6_blk_hole_entry);
6565 #endif
6566         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6567 }
6568
6569 static int __net_init ip6_route_net_init_late(struct net *net)
6570 {
6571 #ifdef CONFIG_PROC_FS
6572         if (!proc_create_net("ipv6_route", 0, net->proc_net,
6573                              &ipv6_route_seq_ops,
6574                              sizeof(struct ipv6_route_iter)))
6575                 return -ENOMEM;
6576
6577         if (!proc_create_net_single("rt6_stats", 0444, net->proc_net,
6578                                     rt6_stats_seq_show, NULL)) {
6579                 remove_proc_entry("ipv6_route", net->proc_net);
6580                 return -ENOMEM;
6581         }
6582 #endif
6583         return 0;
6584 }
6585
6586 static void __net_exit ip6_route_net_exit_late(struct net *net)
6587 {
6588 #ifdef CONFIG_PROC_FS
6589         remove_proc_entry("ipv6_route", net->proc_net);
6590         remove_proc_entry("rt6_stats", net->proc_net);
6591 #endif
6592 }
6593
6594 static struct pernet_operations ip6_route_net_ops = {
6595         .init = ip6_route_net_init,
6596         .exit = ip6_route_net_exit,
6597 };
6598
6599 static int __net_init ipv6_inetpeer_init(struct net *net)
6600 {
6601         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6602
6603         if (!bp)
6604                 return -ENOMEM;
6605         inet_peer_base_init(bp);
6606         net->ipv6.peers = bp;
6607         return 0;
6608 }
6609
6610 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6611 {
6612         struct inet_peer_base *bp = net->ipv6.peers;
6613
6614         net->ipv6.peers = NULL;
6615         inetpeer_invalidate_tree(bp);
6616         kfree(bp);
6617 }
6618
6619 static struct pernet_operations ipv6_inetpeer_ops = {
6620         .init   =       ipv6_inetpeer_init,
6621         .exit   =       ipv6_inetpeer_exit,
6622 };
6623
6624 static struct pernet_operations ip6_route_net_late_ops = {
6625         .init = ip6_route_net_init_late,
6626         .exit = ip6_route_net_exit_late,
6627 };
6628
6629 static struct notifier_block ip6_route_dev_notifier = {
6630         .notifier_call = ip6_route_dev_notify,
6631         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6632 };
6633
6634 void __init ip6_route_init_special_entries(void)
6635 {
6636         /* Registering of the loopback is done before this portion of code,
6637          * the loopback reference in rt6_info will not be taken, do it
6638          * manually for init_net */
6639         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6640         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6641         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6642   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6643         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6644         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6645         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6646         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6647   #endif
6648 }
6649
6650 #if IS_BUILTIN(CONFIG_IPV6)
6651 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6652 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6653
6654 BTF_ID_LIST(btf_fib6_info_id)
6655 BTF_ID(struct, fib6_info)
6656
6657 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6658         .seq_ops                = &ipv6_route_seq_ops,
6659         .init_seq_private       = bpf_iter_init_seq_net,
6660         .fini_seq_private       = bpf_iter_fini_seq_net,
6661         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6662 };
6663
6664 static struct bpf_iter_reg ipv6_route_reg_info = {
6665         .target                 = "ipv6_route",
6666         .ctx_arg_info_size      = 1,
6667         .ctx_arg_info           = {
6668                 { offsetof(struct bpf_iter__ipv6_route, rt),
6669                   PTR_TO_BTF_ID_OR_NULL },
6670         },
6671         .seq_info               = &ipv6_route_seq_info,
6672 };
6673
6674 static int __init bpf_iter_register(void)
6675 {
6676         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6677         return bpf_iter_reg_target(&ipv6_route_reg_info);
6678 }
6679
6680 static void bpf_iter_unregister(void)
6681 {
6682         bpf_iter_unreg_target(&ipv6_route_reg_info);
6683 }
6684 #endif
6685 #endif
6686
6687 int __init ip6_route_init(void)
6688 {
6689         int ret;
6690         int cpu;
6691
6692         ret = -ENOMEM;
6693         ip6_dst_ops_template.kmem_cachep =
6694                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6695                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6696         if (!ip6_dst_ops_template.kmem_cachep)
6697                 goto out;
6698
6699         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6700         if (ret)
6701                 goto out_kmem_cache;
6702
6703         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6704         if (ret)
6705                 goto out_dst_entries;
6706
6707         ret = register_pernet_subsys(&ip6_route_net_ops);
6708         if (ret)
6709                 goto out_register_inetpeer;
6710
6711         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6712
6713         ret = fib6_init();
6714         if (ret)
6715                 goto out_register_subsys;
6716
6717         ret = xfrm6_init();
6718         if (ret)
6719                 goto out_fib6_init;
6720
6721         ret = fib6_rules_init();
6722         if (ret)
6723                 goto xfrm6_init;
6724
6725         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6726         if (ret)
6727                 goto fib6_rules_init;
6728
6729         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6730                                    inet6_rtm_newroute, NULL, 0);
6731         if (ret < 0)
6732                 goto out_register_late_subsys;
6733
6734         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6735                                    inet6_rtm_delroute, NULL, 0);
6736         if (ret < 0)
6737                 goto out_register_late_subsys;
6738
6739         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6740                                    inet6_rtm_getroute, NULL,
6741                                    RTNL_FLAG_DOIT_UNLOCKED);
6742         if (ret < 0)
6743                 goto out_register_late_subsys;
6744
6745         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6746         if (ret)
6747                 goto out_register_late_subsys;
6748
6749 #if IS_BUILTIN(CONFIG_IPV6)
6750 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6751         ret = bpf_iter_register();
6752         if (ret)
6753                 goto out_register_late_subsys;
6754 #endif
6755 #endif
6756
6757         for_each_possible_cpu(cpu) {
6758                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6759
6760                 INIT_LIST_HEAD(&ul->head);
6761                 INIT_LIST_HEAD(&ul->quarantine);
6762                 spin_lock_init(&ul->lock);
6763         }
6764
6765 out:
6766         return ret;
6767
6768 out_register_late_subsys:
6769         rtnl_unregister_all(PF_INET6);
6770         unregister_pernet_subsys(&ip6_route_net_late_ops);
6771 fib6_rules_init:
6772         fib6_rules_cleanup();
6773 xfrm6_init:
6774         xfrm6_fini();
6775 out_fib6_init:
6776         fib6_gc_cleanup();
6777 out_register_subsys:
6778         unregister_pernet_subsys(&ip6_route_net_ops);
6779 out_register_inetpeer:
6780         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6781 out_dst_entries:
6782         dst_entries_destroy(&ip6_dst_blackhole_ops);
6783 out_kmem_cache:
6784         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6785         goto out;
6786 }
6787
6788 void ip6_route_cleanup(void)
6789 {
6790 #if IS_BUILTIN(CONFIG_IPV6)
6791 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6792         bpf_iter_unregister();
6793 #endif
6794 #endif
6795         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6796         unregister_pernet_subsys(&ip6_route_net_late_ops);
6797         fib6_rules_cleanup();
6798         xfrm6_fini();
6799         fib6_gc_cleanup();
6800         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6801         unregister_pernet_subsys(&ip6_route_net_ops);
6802         dst_entries_destroy(&ip6_dst_blackhole_ops);
6803         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6804 }