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 <net/net_namespace.h>
  45 #include <net/snmp.h>
  46 #include <net/ipv6.h>
  47 #include <net/ip6_fib.h>
  48 #include <net/ip6_route.h>
  49 #include <net/ndisc.h>
  50 #include <net/addrconf.h>
  51 #include <net/tcp.h>
  52 #include <linux/rtnetlink.h>
  53 #include <net/dst.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/xfrm.h>
  56 #include <net/netevent.h>
  57 #include <net/netlink.h>
  58 #include <net/rtnh.h>
  59 #include <net/lwtunnel.h>
  60 #include <net/ip_tunnels.h>
  61 #include <net/l3mdev.h>
  62 #include <net/ip.h>
  63 #include <linux/uaccess.h>
  64 #include <linux/btf_ids.h>
  65
  66 #ifdef CONFIG_SYSCTL
  67 #include <linux/sysctl.h>
  68 #endif
  69
  70 static int ip6_rt_type_to_error(u8 fib6_type);
  71
  72 #define CREATE_TRACE_POINTS
  73 #include <trace/events/fib6.h>
  74 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  75 #undef CREATE_TRACE_POINTS
  76
  77 enum rt6_nud_state {
  78         RT6_NUD_FAIL_HARD = -3,
  79         RT6_NUD_FAIL_PROBE = -2,
  80         RT6_NUD_FAIL_DO_RR = -1,
  81         RT6_NUD_SUCCEED = 1
  82 };
  83
  84 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  85 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  86 static unsigned int      ip6_mtu(const struct dst_entry *dst);
  87 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  88 static void             ip6_dst_destroy(struct dst_entry *);
  89 static void             ip6_dst_ifdown(struct dst_entry *,
  90                                        struct net_device *dev, int how);
  91 static int               ip6_dst_gc(struct dst_ops *ops);
  92
  93 static int              ip6_pkt_discard(struct sk_buff *skb);
  94 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  95 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  96 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  97 static void             ip6_link_failure(struct sk_buff *skb);
  98 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
  99                                            struct sk_buff *skb, u32 mtu,
 100                                            bool confirm_neigh);
 101 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 102                                         struct sk_buff *skb);
 103 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 104                            int strict);
 105 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 106 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 107                          struct fib6_info *rt, struct dst_entry *dst,
 108                          struct in6_addr *dest, struct in6_addr *src,
 109                          int iif, int type, u32 portid, u32 seq,
 110                          unsigned int flags);
 111 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 112                                            const struct in6_addr *daddr,
 113                                            const struct in6_addr *saddr);
 114
 115 #ifdef CONFIG_IPV6_ROUTE_INFO
 116 static struct fib6_info *rt6_add_route_info(struct net *net,
 117                                            const struct in6_addr *prefix, int prefixlen,
 118                                            const struct in6_addr *gwaddr,
 119                                            struct net_device *dev,
 120                                            unsigned int pref);
 121 static struct fib6_info *rt6_get_route_info(struct net *net,
 122                                            const struct in6_addr *prefix, int prefixlen,
 123                                            const struct in6_addr *gwaddr,
 124                                            struct net_device *dev);
 125 #endif
 126
 127 struct uncached_list {
 128         spinlock_t              lock;
 129         struct list_head        head;
 130 };
 131
 132 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 133
 134 void rt6_uncached_list_add(struct rt6_info *rt)
 135 {
 136         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 137
 138         rt->rt6i_uncached_list = ul;
 139
 140         spin_lock_bh(&ul->lock);
 141         list_add_tail(&rt->rt6i_uncached, &ul->head);
 142         spin_unlock_bh(&ul->lock);
 143 }
 144
 145 void rt6_uncached_list_del(struct rt6_info *rt)
 146 {
 147         if (!list_empty(&rt->rt6i_uncached)) {
 148                 struct uncached_list *ul = rt->rt6i_uncached_list;
 149                 struct net *net = dev_net(rt->dst.dev);
 150
 151                 spin_lock_bh(&ul->lock);
 152                 list_del(&rt->rt6i_uncached);
 153                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 154                 spin_unlock_bh(&ul->lock);
 155         }
 156 }
 157
 158 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 159 {
 160         struct net_device *loopback_dev = net->loopback_dev;
 161         int cpu;
 162
 163         if (dev == loopback_dev)
 164                 return;
 165
 166         for_each_possible_cpu(cpu) {
 167                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 168                 struct rt6_info *rt;
 169
 170                 spin_lock_bh(&ul->lock);
 171                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 172                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 173                         struct net_device *rt_dev = rt->dst.dev;
 174
 175                         if (rt_idev->dev == dev) {
 176                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 177                                 in6_dev_put(rt_idev);
 178                         }
 179
 180                         if (rt_dev == dev) {
 181                                 rt->dst.dev = blackhole_netdev;
 182                                 dev_hold(rt->dst.dev);
 183                                 dev_put(rt_dev);
 184                         }
 185                 }
 186                 spin_unlock_bh(&ul->lock);
 187         }
 188 }
 189
 190 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 191                                              struct sk_buff *skb,
 192                                              const void *daddr)
 193 {
 194         if (!ipv6_addr_any(p))
 195                 return (const void *) p;
 196         else if (skb)
 197                 return &ipv6_hdr(skb)->daddr;
 198         return daddr;
 199 }
 200
 201 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 202                                    struct net_device *dev,
 203                                    struct sk_buff *skb,
 204                                    const void *daddr)
 205 {
 206         struct neighbour *n;
 207
 208         daddr = choose_neigh_daddr(gw, skb, daddr);
 209         n = __ipv6_neigh_lookup(dev, daddr);
 210         if (n)
 211                 return n;
 212
 213         n = neigh_create(&nd_tbl, daddr, dev);
 214         return IS_ERR(n) ? NULL : n;
 215 }
 216
 217 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 218                                               struct sk_buff *skb,
 219                                               const void *daddr)
 220 {
 221         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 222
 223         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 224                                 dst->dev, skb, daddr);
 225 }
 226
 227 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 228 {
 229         struct net_device *dev = dst->dev;
 230         struct rt6_info *rt = (struct rt6_info *)dst;
 231
 232         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 233         if (!daddr)
 234                 return;
 235         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 236                 return;
 237         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 238                 return;
 239         __ipv6_confirm_neigh(dev, daddr);
 240 }
 241
 242 static struct dst_ops ip6_dst_ops_template = {
 243         .family                 =       AF_INET6,
 244         .gc                     =       ip6_dst_gc,
 245         .gc_thresh              =       1024,
 246         .check                  =       ip6_dst_check,
 247         .default_advmss         =       ip6_default_advmss,
 248         .mtu                    =       ip6_mtu,
 249         .cow_metrics            =       dst_cow_metrics_generic,
 250         .destroy                =       ip6_dst_destroy,
 251         .ifdown                 =       ip6_dst_ifdown,
 252         .negative_advice        =       ip6_negative_advice,
 253         .link_failure           =       ip6_link_failure,
 254         .update_pmtu            =       ip6_rt_update_pmtu,
 255         .redirect               =       rt6_do_redirect,
 256         .local_out              =       __ip6_local_out,
 257         .neigh_lookup           =       ip6_dst_neigh_lookup,
 258         .confirm_neigh          =       ip6_confirm_neigh,
 259 };
 260
 261 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 262 {
 263         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 264
 265         return mtu ? : dst->dev->mtu;
 266 }
 267
 268 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 269                                          struct sk_buff *skb, u32 mtu,
 270                                          bool confirm_neigh)
 271 {
 272 }
 273
 274 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 275                                       struct sk_buff *skb)
 276 {
 277 }
 278
 279 static struct dst_ops ip6_dst_blackhole_ops = {
 280         .family                 =       AF_INET6,
 281         .destroy                =       ip6_dst_destroy,
 282         .check                  =       ip6_dst_check,
 283         .mtu                    =       ip6_blackhole_mtu,
 284         .default_advmss         =       ip6_default_advmss,
 285         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 286         .redirect               =       ip6_rt_blackhole_redirect,
 287         .cow_metrics            =       dst_cow_metrics_generic,
 288         .neigh_lookup           =       ip6_dst_neigh_lookup,
 289 };
 290
 291 static const u32 ip6_template_metrics[RTAX_MAX] = {
 292         [RTAX_HOPLIMIT - 1] = 0,
 293 };
 294
 295 static const struct fib6_info fib6_null_entry_template = {
 296         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 297         .fib6_protocol  = RTPROT_KERNEL,
 298         .fib6_metric    = ~(u32)0,
 299         .fib6_ref       = REFCOUNT_INIT(1),
 300         .fib6_type      = RTN_UNREACHABLE,
 301         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 302 };
 303
 304 static const struct rt6_info ip6_null_entry_template = {
 305         .dst = {
 306                 .__refcnt       = ATOMIC_INIT(1),
 307                 .__use          = 1,
 308                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 309                 .error          = -ENETUNREACH,
 310                 .input          = ip6_pkt_discard,
 311                 .output         = ip6_pkt_discard_out,
 312         },
 313         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 314 };
 315
 316 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 317
 318 static const struct rt6_info ip6_prohibit_entry_template = {
 319         .dst = {
 320                 .__refcnt       = ATOMIC_INIT(1),
 321                 .__use          = 1,
 322                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 323                 .error          = -EACCES,
 324                 .input          = ip6_pkt_prohibit,
 325                 .output         = ip6_pkt_prohibit_out,
 326         },
 327         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 328 };
 329
 330 static const struct rt6_info ip6_blk_hole_entry_template = {
 331         .dst = {
 332                 .__refcnt       = ATOMIC_INIT(1),
 333                 .__use          = 1,
 334                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 335                 .error          = -EINVAL,
 336                 .input          = dst_discard,
 337                 .output         = dst_discard_out,
 338         },
 339         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 340 };
 341
 342 #endif
 343
 344 static void rt6_info_init(struct rt6_info *rt)
 345 {
 346         struct dst_entry *dst = &rt->dst;
 347
 348         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 349         INIT_LIST_HEAD(&rt->rt6i_uncached);
 350 }
 351
 352 /* allocate dst with ip6_dst_ops */
 353 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 354                                int flags)
 355 {
 356         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 357                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 358
 359         if (rt) {
 360                 rt6_info_init(rt);
 361                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 362         }
 363
 364         return rt;
 365 }
 366 EXPORT_SYMBOL(ip6_dst_alloc);
 367
 368 static void ip6_dst_destroy(struct dst_entry *dst)
 369 {
 370         struct rt6_info *rt = (struct rt6_info *)dst;
 371         struct fib6_info *from;
 372         struct inet6_dev *idev;
 373
 374         ip_dst_metrics_put(dst);
 375         rt6_uncached_list_del(rt);
 376
 377         idev = rt->rt6i_idev;
 378         if (idev) {
 379                 rt->rt6i_idev = NULL;
 380                 in6_dev_put(idev);
 381         }
 382
 383         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 384         fib6_info_release(from);
 385 }
 386
 387 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 388                            int how)
 389 {
 390         struct rt6_info *rt = (struct rt6_info *)dst;
 391         struct inet6_dev *idev = rt->rt6i_idev;
 392         struct net_device *loopback_dev =
 393                 dev_net(dev)->loopback_dev;
 394
 395         if (idev && idev->dev != loopback_dev) {
 396                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 397                 if (loopback_idev) {
 398                         rt->rt6i_idev = loopback_idev;
 399                         in6_dev_put(idev);
 400                 }
 401         }
 402 }
 403
 404 static bool __rt6_check_expired(const struct rt6_info *rt)
 405 {
 406         if (rt->rt6i_flags & RTF_EXPIRES)
 407                 return time_after(jiffies, rt->dst.expires);
 408         else
 409                 return false;
 410 }
 411
 412 static bool rt6_check_expired(const struct rt6_info *rt)
 413 {
 414         struct fib6_info *from;
 415
 416         from = rcu_dereference(rt->from);
 417
 418         if (rt->rt6i_flags & RTF_EXPIRES) {
 419                 if (time_after(jiffies, rt->dst.expires))
 420                         return true;
 421         } else if (from) {
 422                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 423                         fib6_check_expired(from);
 424         }
 425         return false;
 426 }
 427
 428 void fib6_select_path(const struct net *net, struct fib6_result *res,
 429                       struct flowi6 *fl6, int oif, bool have_oif_match,
 430                       const struct sk_buff *skb, int strict)
 431 {
 432         struct fib6_info *sibling, *next_sibling;
 433         struct fib6_info *match = res->f6i;
 434
 435         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 436                 goto out;
 437
 438         if (match->nh && have_oif_match && res->nh)
 439                 return;
 440
 441         /* We might have already computed the hash for ICMPv6 errors. In such
 442          * case it will always be non-zero. Otherwise now is the time to do it.
 443          */
 444         if (!fl6->mp_hash &&
 445             (!match->nh || nexthop_is_multipath(match->nh)))
 446                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 447
 448         if (unlikely(match->nh)) {
 449                 nexthop_path_fib6_result(res, fl6->mp_hash);
 450                 return;
 451         }
 452
 453         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 454                 goto out;
 455
 456         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 457                                  fib6_siblings) {
 458                 const struct fib6_nh *nh = sibling->fib6_nh;
 459                 int nh_upper_bound;
 460
 461                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 462                 if (fl6->mp_hash > nh_upper_bound)
 463                         continue;
 464                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 465                         break;
 466                 match = sibling;
 467                 break;
 468         }
 469
 470 out:
 471         res->f6i = match;
 472         res->nh = match->fib6_nh;
 473 }
 474
 475 /*
 476  *      Route lookup. rcu_read_lock() should be held.
 477  */
 478
 479 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 480                                const struct in6_addr *saddr, int oif, int flags)
 481 {
 482         const struct net_device *dev;
 483
 484         if (nh->fib_nh_flags & RTNH_F_DEAD)
 485                 return false;
 486
 487         dev = nh->fib_nh_dev;
 488         if (oif) {
 489                 if (dev->ifindex == oif)
 490                         return true;
 491         } else {
 492                 if (ipv6_chk_addr(net, saddr, dev,
 493                                   flags & RT6_LOOKUP_F_IFACE))
 494                         return true;
 495         }
 496
 497         return false;
 498 }
 499
 500 struct fib6_nh_dm_arg {
 501         struct net              *net;
 502         const struct in6_addr   *saddr;
 503         int                     oif;
 504         int                     flags;
 505         struct fib6_nh          *nh;
 506 };
 507
 508 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 509 {
 510         struct fib6_nh_dm_arg *arg = _arg;
 511
 512         arg->nh = nh;
 513         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 514                                   arg->flags);
 515 }
 516
 517 /* returns fib6_nh from nexthop or NULL */
 518 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 519                                         struct fib6_result *res,
 520                                         const struct in6_addr *saddr,
 521                                         int oif, int flags)
 522 {
 523         struct fib6_nh_dm_arg arg = {
 524                 .net   = net,
 525                 .saddr = saddr,
 526                 .oif   = oif,
 527                 .flags = flags,
 528         };
 529
 530         if (nexthop_is_blackhole(nh))
 531                 return NULL;
 532
 533         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 534                 return arg.nh;
 535
 536         return NULL;
 537 }
 538
 539 static void rt6_device_match(struct net *net, struct fib6_result *res,
 540                              const struct in6_addr *saddr, int oif, int flags)
 541 {
 542         struct fib6_info *f6i = res->f6i;
 543         struct fib6_info *spf6i;
 544         struct fib6_nh *nh;
 545
 546         if (!oif && ipv6_addr_any(saddr)) {
 547                 if (unlikely(f6i->nh)) {
 548                         nh = nexthop_fib6_nh(f6i->nh);
 549                         if (nexthop_is_blackhole(f6i->nh))
 550                                 goto out_blackhole;
 551                 } else {
 552                         nh = f6i->fib6_nh;
 553                 }
 554                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 555                         goto out;
 556         }
 557
 558         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 559                 bool matched = false;
 560
 561                 if (unlikely(spf6i->nh)) {
 562                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 563                                               oif, flags);
 564                         if (nh)
 565                                 matched = true;
 566                 } else {
 567                         nh = spf6i->fib6_nh;
 568                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 569                                 matched = true;
 570                 }
 571                 if (matched) {
 572                         res->f6i = spf6i;
 573                         goto out;
 574                 }
 575         }
 576
 577         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 578                 res->f6i = net->ipv6.fib6_null_entry;
 579                 nh = res->f6i->fib6_nh;
 580                 goto out;
 581         }
 582
 583         if (unlikely(f6i->nh)) {
 584                 nh = nexthop_fib6_nh(f6i->nh);
 585                 if (nexthop_is_blackhole(f6i->nh))
 586                         goto out_blackhole;
 587         } else {
 588                 nh = f6i->fib6_nh;
 589         }
 590
 591         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 592                 res->f6i = net->ipv6.fib6_null_entry;
 593                 nh = res->f6i->fib6_nh;
 594         }
 595 out:
 596         res->nh = nh;
 597         res->fib6_type = res->f6i->fib6_type;
 598         res->fib6_flags = res->f6i->fib6_flags;
 599         return;
 600
 601 out_blackhole:
 602         res->fib6_flags |= RTF_REJECT;
 603         res->fib6_type = RTN_BLACKHOLE;
 604         res->nh = nh;
 605 }
 606
 607 #ifdef CONFIG_IPV6_ROUTER_PREF
 608 struct __rt6_probe_work {
 609         struct work_struct work;
 610         struct in6_addr target;
 611         struct net_device *dev;
 612 };
 613
 614 static void rt6_probe_deferred(struct work_struct *w)
 615 {
 616         struct in6_addr mcaddr;
 617         struct __rt6_probe_work *work =
 618                 container_of(w, struct __rt6_probe_work, work);
 619
 620         addrconf_addr_solict_mult(&work->target, &mcaddr);
 621         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 622         dev_put(work->dev);
 623         kfree(work);
 624 }
 625
 626 static void rt6_probe(struct fib6_nh *fib6_nh)
 627 {
 628         struct __rt6_probe_work *work = NULL;
 629         const struct in6_addr *nh_gw;
 630         unsigned long last_probe;
 631         struct neighbour *neigh;
 632         struct net_device *dev;
 633         struct inet6_dev *idev;
 634
 635         /*
 636          * Okay, this does not seem to be appropriate
 637          * for now, however, we need to check if it
 638          * is really so; aka Router Reachability Probing.
 639          *
 640          * Router Reachability Probe MUST be rate-limited
 641          * to no more than one per minute.
 642          */
 643         if (!fib6_nh->fib_nh_gw_family)
 644                 return;
 645
 646         nh_gw = &fib6_nh->fib_nh_gw6;
 647         dev = fib6_nh->fib_nh_dev;
 648         rcu_read_lock_bh();
 649         last_probe = READ_ONCE(fib6_nh->last_probe);
 650         idev = __in6_dev_get(dev);
 651         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 652         if (neigh) {
 653                 if (neigh->nud_state & NUD_VALID)
 654                         goto out;
 655
 656                 write_lock(&neigh->lock);
 657                 if (!(neigh->nud_state & NUD_VALID) &&
 658                     time_after(jiffies,
 659                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 660                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 661                         if (work)
 662                                 __neigh_set_probe_once(neigh);
 663                 }
 664                 write_unlock(&neigh->lock);
 665         } else if (time_after(jiffies, last_probe +
 666                                        idev->cnf.rtr_probe_interval)) {
 667                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 668         }
 669
 670         if (!work || cmpxchg(&fib6_nh->last_probe,
 671                              last_probe, jiffies) != last_probe) {
 672                 kfree(work);
 673         } else {
 674                 INIT_WORK(&work->work, rt6_probe_deferred);
 675                 work->target = *nh_gw;
 676                 dev_hold(dev);
 677                 work->dev = dev;
 678                 schedule_work(&work->work);
 679         }
 680
 681 out:
 682         rcu_read_unlock_bh();
 683 }
 684 #else
 685 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 686 {
 687 }
 688 #endif
 689
 690 /*
 691  * Default Router Selection (RFC 2461 6.3.6)
 692  */
 693 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 694 {
 695         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 696         struct neighbour *neigh;
 697
 698         rcu_read_lock_bh();
 699         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 700                                           &fib6_nh->fib_nh_gw6);
 701         if (neigh) {
 702                 read_lock(&neigh->lock);
 703                 if (neigh->nud_state & NUD_VALID)
 704                         ret = RT6_NUD_SUCCEED;
 705 #ifdef CONFIG_IPV6_ROUTER_PREF
 706                 else if (!(neigh->nud_state & NUD_FAILED))
 707                         ret = RT6_NUD_SUCCEED;
 708                 else
 709                         ret = RT6_NUD_FAIL_PROBE;
 710 #endif
 711                 read_unlock(&neigh->lock);
 712         } else {
 713                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 714                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 715         }
 716         rcu_read_unlock_bh();
 717
 718         return ret;
 719 }
 720
 721 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 722                            int strict)
 723 {
 724         int m = 0;
 725
 726         if (!oif || nh->fib_nh_dev->ifindex == oif)
 727                 m = 2;
 728
 729         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 730                 return RT6_NUD_FAIL_HARD;
 731 #ifdef CONFIG_IPV6_ROUTER_PREF
 732         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 733 #endif
 734         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 735             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 736                 int n = rt6_check_neigh(nh);
 737                 if (n < 0)
 738                         return n;
 739         }
 740         return m;
 741 }
 742
 743 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 744                        int oif, int strict, int *mpri, bool *do_rr)
 745 {
 746         bool match_do_rr = false;
 747         bool rc = false;
 748         int m;
 749
 750         if (nh->fib_nh_flags & RTNH_F_DEAD)
 751                 goto out;
 752
 753         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 754             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 755             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 756                 goto out;
 757
 758         m = rt6_score_route(nh, fib6_flags, oif, strict);
 759         if (m == RT6_NUD_FAIL_DO_RR) {
 760                 match_do_rr = true;
 761                 m = 0; /* lowest valid score */
 762         } else if (m == RT6_NUD_FAIL_HARD) {
 763                 goto out;
 764         }
 765
 766         if (strict & RT6_LOOKUP_F_REACHABLE)
 767                 rt6_probe(nh);
 768
 769         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 770         if (m > *mpri) {
 771                 *do_rr = match_do_rr;
 772                 *mpri = m;
 773                 rc = true;
 774         }
 775 out:
 776         return rc;
 777 }
 778
 779 struct fib6_nh_frl_arg {
 780         u32             flags;
 781         int             oif;
 782         int             strict;
 783         int             *mpri;
 784         bool            *do_rr;
 785         struct fib6_nh  *nh;
 786 };
 787
 788 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 789 {
 790         struct fib6_nh_frl_arg *arg = _arg;
 791
 792         arg->nh = nh;
 793         return find_match(nh, arg->flags, arg->oif, arg->strict,
 794                           arg->mpri, arg->do_rr);
 795 }
 796
 797 static void __find_rr_leaf(struct fib6_info *f6i_start,
 798                            struct fib6_info *nomatch, u32 metric,
 799                            struct fib6_result *res, struct fib6_info **cont,
 800                            int oif, int strict, bool *do_rr, int *mpri)
 801 {
 802         struct fib6_info *f6i;
 803
 804         for (f6i = f6i_start;
 805              f6i && f6i != nomatch;
 806              f6i = rcu_dereference(f6i->fib6_next)) {
 807                 bool matched = false;
 808                 struct fib6_nh *nh;
 809
 810                 if (cont && f6i->fib6_metric != metric) {
 811                         *cont = f6i;
 812                         return;
 813                 }
 814
 815                 if (fib6_check_expired(f6i))
 816                         continue;
 817
 818                 if (unlikely(f6i->nh)) {
 819                         struct fib6_nh_frl_arg arg = {
 820                                 .flags  = f6i->fib6_flags,
 821                                 .oif    = oif,
 822                                 .strict = strict,
 823                                 .mpri   = mpri,
 824                                 .do_rr  = do_rr
 825                         };
 826
 827                         if (nexthop_is_blackhole(f6i->nh)) {
 828                                 res->fib6_flags = RTF_REJECT;
 829                                 res->fib6_type = RTN_BLACKHOLE;
 830                                 res->f6i = f6i;
 831                                 res->nh = nexthop_fib6_nh(f6i->nh);
 832                                 return;
 833                         }
 834                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 835                                                      &arg)) {
 836                                 matched = true;
 837                                 nh = arg.nh;
 838                         }
 839                 } else {
 840                         nh = f6i->fib6_nh;
 841                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 842                                        mpri, do_rr))
 843                                 matched = true;
 844                 }
 845                 if (matched) {
 846                         res->f6i = f6i;
 847                         res->nh = nh;
 848                         res->fib6_flags = f6i->fib6_flags;
 849                         res->fib6_type = f6i->fib6_type;
 850                 }
 851         }
 852 }
 853
 854 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 855                          struct fib6_info *rr_head, int oif, int strict,
 856                          bool *do_rr, struct fib6_result *res)
 857 {
 858         u32 metric = rr_head->fib6_metric;
 859         struct fib6_info *cont = NULL;
 860         int mpri = -1;
 861
 862         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 863                        oif, strict, do_rr, &mpri);
 864
 865         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 866                        oif, strict, do_rr, &mpri);
 867
 868         if (res->f6i || !cont)
 869                 return;
 870
 871         __find_rr_leaf(cont, NULL, metric, res, NULL,
 872                        oif, strict, do_rr, &mpri);
 873 }
 874
 875 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 876                        struct fib6_result *res, int strict)
 877 {
 878         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 879         struct fib6_info *rt0;
 880         bool do_rr = false;
 881         int key_plen;
 882
 883         /* make sure this function or its helpers sets f6i */
 884         res->f6i = NULL;
 885
 886         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 887                 goto out;
 888
 889         rt0 = rcu_dereference(fn->rr_ptr);
 890         if (!rt0)
 891                 rt0 = leaf;
 892
 893         /* Double check to make sure fn is not an intermediate node
 894          * and fn->leaf does not points to its child's leaf
 895          * (This might happen if all routes under fn are deleted from
 896          * the tree and fib6_repair_tree() is called on the node.)
 897          */
 898         key_plen = rt0->fib6_dst.plen;
 899 #ifdef CONFIG_IPV6_SUBTREES
 900         if (rt0->fib6_src.plen)
 901                 key_plen = rt0->fib6_src.plen;
 902 #endif
 903         if (fn->fn_bit != key_plen)
 904                 goto out;
 905
 906         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 907         if (do_rr) {
 908                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 909
 910                 /* no entries matched; do round-robin */
 911                 if (!next || next->fib6_metric != rt0->fib6_metric)
 912                         next = leaf;
 913
 914                 if (next != rt0) {
 915                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 916                         /* make sure next is not being deleted from the tree */
 917                         if (next->fib6_node)
 918                                 rcu_assign_pointer(fn->rr_ptr, next);
 919                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 920                 }
 921         }
 922
 923 out:
 924         if (!res->f6i) {
 925                 res->f6i = net->ipv6.fib6_null_entry;
 926                 res->nh = res->f6i->fib6_nh;
 927                 res->fib6_flags = res->f6i->fib6_flags;
 928                 res->fib6_type = res->f6i->fib6_type;
 929         }
 930 }
 931
 932 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 933 {
 934         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 935                res->nh->fib_nh_gw_family;
 936 }
 937
 938 #ifdef CONFIG_IPV6_ROUTE_INFO
 939 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 940                   const struct in6_addr *gwaddr)
 941 {
 942         struct net *net = dev_net(dev);
 943         struct route_info *rinfo = (struct route_info *) opt;
 944         struct in6_addr prefix_buf, *prefix;
 945         unsigned int pref;
 946         unsigned long lifetime;
 947         struct fib6_info *rt;
 948
 949         if (len < sizeof(struct route_info)) {
 950                 return -EINVAL;
 951         }
 952
 953         /* Sanity check for prefix_len and length */
 954         if (rinfo->length > 3) {
 955                 return -EINVAL;
 956         } else if (rinfo->prefix_len > 128) {
 957                 return -EINVAL;
 958         } else if (rinfo->prefix_len > 64) {
 959                 if (rinfo->length < 2) {
 960                         return -EINVAL;
 961                 }
 962         } else if (rinfo->prefix_len > 0) {
 963                 if (rinfo->length < 1) {
 964                         return -EINVAL;
 965                 }
 966         }
 967
 968         pref = rinfo->route_pref;
 969         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 970                 return -EINVAL;
 971
 972         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 973
 974         if (rinfo->length == 3)
 975                 prefix = (struct in6_addr *)rinfo->prefix;
 976         else {
 977                 /* this function is safe */
 978                 ipv6_addr_prefix(&prefix_buf,
 979                                  (struct in6_addr *)rinfo->prefix,
 980                                  rinfo->prefix_len);
 981                 prefix = &prefix_buf;
 982         }
 983
 984         if (rinfo->prefix_len == 0)
 985                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 986         else
 987                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 988                                         gwaddr, dev);
 989
 990         if (rt && !lifetime) {
 991                 ip6_del_rt(net, rt, false);
 992                 rt = NULL;
 993         }
 994
 995         if (!rt && lifetime)
 996                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 997                                         dev, pref);
 998         else if (rt)
 999                 rt->fib6_flags = RTF_ROUTEINFO |
1000                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
1001
1002         if (rt) {
1003                 if (!addrconf_finite_timeout(lifetime))
1004                         fib6_clean_expires(rt);
1005                 else
1006                         fib6_set_expires(rt, jiffies + HZ * lifetime);
1007
1008                 fib6_info_release(rt);
1009         }
1010         return 0;
1011 }
1012 #endif
1013
1014 /*
1015  *      Misc support functions
1016  */
1017
1018 /* called with rcu_lock held */
1019 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1020 {
1021         struct net_device *dev = res->nh->fib_nh_dev;
1022
1023         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1024                 /* for copies of local routes, dst->dev needs to be the
1025                  * device if it is a master device, the master device if
1026                  * device is enslaved, and the loopback as the default
1027                  */
1028                 if (netif_is_l3_slave(dev) &&
1029                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1030                         dev = l3mdev_master_dev_rcu(dev);
1031                 else if (!netif_is_l3_master(dev))
1032                         dev = dev_net(dev)->loopback_dev;
1033                 /* last case is netif_is_l3_master(dev) is true in which
1034                  * case we want dev returned to be dev
1035                  */
1036         }
1037
1038         return dev;
1039 }
1040
1041 static const int fib6_prop[RTN_MAX + 1] = {
1042         [RTN_UNSPEC]    = 0,
1043         [RTN_UNICAST]   = 0,
1044         [RTN_LOCAL]     = 0,
1045         [RTN_BROADCAST] = 0,
1046         [RTN_ANYCAST]   = 0,
1047         [RTN_MULTICAST] = 0,
1048         [RTN_BLACKHOLE] = -EINVAL,
1049         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1050         [RTN_PROHIBIT]  = -EACCES,
1051         [RTN_THROW]     = -EAGAIN,
1052         [RTN_NAT]       = -EINVAL,
1053         [RTN_XRESOLVE]  = -EINVAL,
1054 };
1055
1056 static int ip6_rt_type_to_error(u8 fib6_type)
1057 {
1058         return fib6_prop[fib6_type];
1059 }
1060
1061 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1062 {
1063         unsigned short flags = 0;
1064
1065         if (rt->dst_nocount)
1066                 flags |= DST_NOCOUNT;
1067         if (rt->dst_nopolicy)
1068                 flags |= DST_NOPOLICY;
1069
1070         return flags;
1071 }
1072
1073 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1074 {
1075         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1076
1077         switch (fib6_type) {
1078         case RTN_BLACKHOLE:
1079                 rt->dst.output = dst_discard_out;
1080                 rt->dst.input = dst_discard;
1081                 break;
1082         case RTN_PROHIBIT:
1083                 rt->dst.output = ip6_pkt_prohibit_out;
1084                 rt->dst.input = ip6_pkt_prohibit;
1085                 break;
1086         case RTN_THROW:
1087         case RTN_UNREACHABLE:
1088         default:
1089                 rt->dst.output = ip6_pkt_discard_out;
1090                 rt->dst.input = ip6_pkt_discard;
1091                 break;
1092         }
1093 }
1094
1095 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1096 {
1097         struct fib6_info *f6i = res->f6i;
1098
1099         if (res->fib6_flags & RTF_REJECT) {
1100                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1101                 return;
1102         }
1103
1104         rt->dst.error = 0;
1105         rt->dst.output = ip6_output;
1106
1107         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1108                 rt->dst.input = ip6_input;
1109         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1110                 rt->dst.input = ip6_mc_input;
1111         } else {
1112                 rt->dst.input = ip6_forward;
1113         }
1114
1115         if (res->nh->fib_nh_lws) {
1116                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1117                 lwtunnel_set_redirect(&rt->dst);
1118         }
1119
1120         rt->dst.lastuse = jiffies;
1121 }
1122
1123 /* Caller must already hold reference to @from */
1124 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1125 {
1126         rt->rt6i_flags &= ~RTF_EXPIRES;
1127         rcu_assign_pointer(rt->from, from);
1128         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1129 }
1130
1131 /* Caller must already hold reference to f6i in result */
1132 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1133 {
1134         const struct fib6_nh *nh = res->nh;
1135         const struct net_device *dev = nh->fib_nh_dev;
1136         struct fib6_info *f6i = res->f6i;
1137
1138         ip6_rt_init_dst(rt, res);
1139
1140         rt->rt6i_dst = f6i->fib6_dst;
1141         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1142         rt->rt6i_flags = res->fib6_flags;
1143         if (nh->fib_nh_gw_family) {
1144                 rt->rt6i_gateway = nh->fib_nh_gw6;
1145                 rt->rt6i_flags |= RTF_GATEWAY;
1146         }
1147         rt6_set_from(rt, f6i);
1148 #ifdef CONFIG_IPV6_SUBTREES
1149         rt->rt6i_src = f6i->fib6_src;
1150 #endif
1151 }
1152
1153 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1154                                         struct in6_addr *saddr)
1155 {
1156         struct fib6_node *pn, *sn;
1157         while (1) {
1158                 if (fn->fn_flags & RTN_TL_ROOT)
1159                         return NULL;
1160                 pn = rcu_dereference(fn->parent);
1161                 sn = FIB6_SUBTREE(pn);
1162                 if (sn && sn != fn)
1163                         fn = fib6_node_lookup(sn, NULL, saddr);
1164                 else
1165                         fn = pn;
1166                 if (fn->fn_flags & RTN_RTINFO)
1167                         return fn;
1168         }
1169 }
1170
1171 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1172 {
1173         struct rt6_info *rt = *prt;
1174
1175         if (dst_hold_safe(&rt->dst))
1176                 return true;
1177         if (net) {
1178                 rt = net->ipv6.ip6_null_entry;
1179                 dst_hold(&rt->dst);
1180         } else {
1181                 rt = NULL;
1182         }
1183         *prt = rt;
1184         return false;
1185 }
1186
1187 /* called with rcu_lock held */
1188 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1189 {
1190         struct net_device *dev = res->nh->fib_nh_dev;
1191         struct fib6_info *f6i = res->f6i;
1192         unsigned short flags;
1193         struct rt6_info *nrt;
1194
1195         if (!fib6_info_hold_safe(f6i))
1196                 goto fallback;
1197
1198         flags = fib6_info_dst_flags(f6i);
1199         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1200         if (!nrt) {
1201                 fib6_info_release(f6i);
1202                 goto fallback;
1203         }
1204
1205         ip6_rt_copy_init(nrt, res);
1206         return nrt;
1207
1208 fallback:
1209         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1210         dst_hold(&nrt->dst);
1211         return nrt;
1212 }
1213
1214 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1215                                              struct fib6_table *table,
1216                                              struct flowi6 *fl6,
1217                                              const struct sk_buff *skb,
1218                                              int flags)
1219 {
1220         struct fib6_result res = {};
1221         struct fib6_node *fn;
1222         struct rt6_info *rt;
1223
1224         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1225                 flags &= ~RT6_LOOKUP_F_IFACE;
1226
1227         rcu_read_lock();
1228         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1229 restart:
1230         res.f6i = rcu_dereference(fn->leaf);
1231         if (!res.f6i)
1232                 res.f6i = net->ipv6.fib6_null_entry;
1233         else
1234                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1235                                  flags);
1236
1237         if (res.f6i == net->ipv6.fib6_null_entry) {
1238                 fn = fib6_backtrack(fn, &fl6->saddr);
1239                 if (fn)
1240                         goto restart;
1241
1242                 rt = net->ipv6.ip6_null_entry;
1243                 dst_hold(&rt->dst);
1244                 goto out;
1245         } else if (res.fib6_flags & RTF_REJECT) {
1246                 goto do_create;
1247         }
1248
1249         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1250                          fl6->flowi6_oif != 0, skb, flags);
1251
1252         /* Search through exception table */
1253         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1254         if (rt) {
1255                 if (ip6_hold_safe(net, &rt))
1256                         dst_use_noref(&rt->dst, jiffies);
1257         } else {
1258 do_create:
1259                 rt = ip6_create_rt_rcu(&res);
1260         }
1261
1262 out:
1263         trace_fib6_table_lookup(net, &res, table, fl6);
1264
1265         rcu_read_unlock();
1266
1267         return rt;
1268 }
1269
1270 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1271                                    const struct sk_buff *skb, int flags)
1272 {
1273         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1274 }
1275 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1276
1277 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1278                             const struct in6_addr *saddr, int oif,
1279                             const struct sk_buff *skb, int strict)
1280 {
1281         struct flowi6 fl6 = {
1282                 .flowi6_oif = oif,
1283                 .daddr = *daddr,
1284         };
1285         struct dst_entry *dst;
1286         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1287
1288         if (saddr) {
1289                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1290                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1291         }
1292
1293         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1294         if (dst->error == 0)
1295                 return (struct rt6_info *) dst;
1296
1297         dst_release(dst);
1298
1299         return NULL;
1300 }
1301 EXPORT_SYMBOL(rt6_lookup);
1302
1303 /* ip6_ins_rt is called with FREE table->tb6_lock.
1304  * It takes new route entry, the addition fails by any reason the
1305  * route is released.
1306  * Caller must hold dst before calling it.
1307  */
1308
1309 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1310                         struct netlink_ext_ack *extack)
1311 {
1312         int err;
1313         struct fib6_table *table;
1314
1315         table = rt->fib6_table;
1316         spin_lock_bh(&table->tb6_lock);
1317         err = fib6_add(&table->tb6_root, rt, info, extack);
1318         spin_unlock_bh(&table->tb6_lock);
1319
1320         return err;
1321 }
1322
1323 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1324 {
1325         struct nl_info info = { .nl_net = net, };
1326
1327         return __ip6_ins_rt(rt, &info, NULL);
1328 }
1329
1330 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1331                                            const struct in6_addr *daddr,
1332                                            const struct in6_addr *saddr)
1333 {
1334         struct fib6_info *f6i = res->f6i;
1335         struct net_device *dev;
1336         struct rt6_info *rt;
1337
1338         /*
1339          *      Clone the route.
1340          */
1341
1342         if (!fib6_info_hold_safe(f6i))
1343                 return NULL;
1344
1345         dev = ip6_rt_get_dev_rcu(res);
1346         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1347         if (!rt) {
1348                 fib6_info_release(f6i);
1349                 return NULL;
1350         }
1351
1352         ip6_rt_copy_init(rt, res);
1353         rt->rt6i_flags |= RTF_CACHE;
1354         rt->rt6i_dst.addr = *daddr;
1355         rt->rt6i_dst.plen = 128;
1356
1357         if (!rt6_is_gw_or_nonexthop(res)) {
1358                 if (f6i->fib6_dst.plen != 128 &&
1359                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1360                         rt->rt6i_flags |= RTF_ANYCAST;
1361 #ifdef CONFIG_IPV6_SUBTREES
1362                 if (rt->rt6i_src.plen && saddr) {
1363                         rt->rt6i_src.addr = *saddr;
1364                         rt->rt6i_src.plen = 128;
1365                 }
1366 #endif
1367         }
1368
1369         return rt;
1370 }
1371
1372 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1373 {
1374         struct fib6_info *f6i = res->f6i;
1375         unsigned short flags = fib6_info_dst_flags(f6i);
1376         struct net_device *dev;
1377         struct rt6_info *pcpu_rt;
1378
1379         if (!fib6_info_hold_safe(f6i))
1380                 return NULL;
1381
1382         rcu_read_lock();
1383         dev = ip6_rt_get_dev_rcu(res);
1384         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1385         rcu_read_unlock();
1386         if (!pcpu_rt) {
1387                 fib6_info_release(f6i);
1388                 return NULL;
1389         }
1390         ip6_rt_copy_init(pcpu_rt, res);
1391         pcpu_rt->rt6i_flags |= RTF_PCPU;
1392
1393         if (f6i->nh)
1394                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1395
1396         return pcpu_rt;
1397 }
1398
1399 static bool rt6_is_valid(const struct rt6_info *rt6)
1400 {
1401         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1402 }
1403
1404 /* It should be called with rcu_read_lock() acquired */
1405 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1406 {
1407         struct rt6_info *pcpu_rt;
1408
1409         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1410
1411         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1412                 struct rt6_info *prev, **p;
1413
1414                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
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 u32 seed __read_mostly;
1504         u32 val;
1505
1506         net_get_random_once(&seed, sizeof(seed));
1507         val = jhash2((const u32 *)dst, sizeof(*dst)/sizeof(u32), seed);
1508
1509 #ifdef CONFIG_IPV6_SUBTREES
1510         if (src)
1511                 val = jhash2((const u32 *)src, sizeof(*src)/sizeof(u32), val);
1512 #endif
1513         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1514 }
1515
1516 /* Helper function to find the cached rt in the hash table
1517  * and update bucket pointer to point to the bucket for this
1518  * (daddr, saddr) pair
1519  * Caller must hold rt6_exception_lock
1520  */
1521 static struct rt6_exception *
1522 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1523                               const struct in6_addr *daddr,
1524                               const struct in6_addr *saddr)
1525 {
1526         struct rt6_exception *rt6_ex;
1527         u32 hval;
1528
1529         if (!(*bucket) || !daddr)
1530                 return NULL;
1531
1532         hval = rt6_exception_hash(daddr, saddr);
1533         *bucket += hval;
1534
1535         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1536                 struct rt6_info *rt6 = rt6_ex->rt6i;
1537                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1538
1539 #ifdef CONFIG_IPV6_SUBTREES
1540                 if (matched && saddr)
1541                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1542 #endif
1543                 if (matched)
1544                         return rt6_ex;
1545         }
1546         return NULL;
1547 }
1548
1549 /* Helper function to find the cached rt in the hash table
1550  * and update bucket pointer to point to the bucket for this
1551  * (daddr, saddr) pair
1552  * Caller must hold rcu_read_lock()
1553  */
1554 static struct rt6_exception *
1555 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1556                          const struct in6_addr *daddr,
1557                          const struct in6_addr *saddr)
1558 {
1559         struct rt6_exception *rt6_ex;
1560         u32 hval;
1561
1562         WARN_ON_ONCE(!rcu_read_lock_held());
1563
1564         if (!(*bucket) || !daddr)
1565                 return NULL;
1566
1567         hval = rt6_exception_hash(daddr, saddr);
1568         *bucket += hval;
1569
1570         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1571                 struct rt6_info *rt6 = rt6_ex->rt6i;
1572                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1573
1574 #ifdef CONFIG_IPV6_SUBTREES
1575                 if (matched && saddr)
1576                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1577 #endif
1578                 if (matched)
1579                         return rt6_ex;
1580         }
1581         return NULL;
1582 }
1583
1584 static unsigned int fib6_mtu(const struct fib6_result *res)
1585 {
1586         const struct fib6_nh *nh = res->nh;
1587         unsigned int mtu;
1588
1589         if (res->f6i->fib6_pmtu) {
1590                 mtu = res->f6i->fib6_pmtu;
1591         } else {
1592                 struct net_device *dev = nh->fib_nh_dev;
1593                 struct inet6_dev *idev;
1594
1595                 rcu_read_lock();
1596                 idev = __in6_dev_get(dev);
1597                 mtu = idev->cnf.mtu6;
1598                 rcu_read_unlock();
1599         }
1600
1601         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1602
1603         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1604 }
1605
1606 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1607
1608 /* used when the flushed bit is not relevant, only access to the bucket
1609  * (ie., all bucket users except rt6_insert_exception);
1610  *
1611  * called under rcu lock; sometimes called with rt6_exception_lock held
1612  */
1613 static
1614 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1615                                                        spinlock_t *lock)
1616 {
1617         struct rt6_exception_bucket *bucket;
1618
1619         if (lock)
1620                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1621                                                    lockdep_is_held(lock));
1622         else
1623                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1624
1625         /* remove bucket flushed bit if set */
1626         if (bucket) {
1627                 unsigned long p = (unsigned long)bucket;
1628
1629                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1630                 bucket = (struct rt6_exception_bucket *)p;
1631         }
1632
1633         return bucket;
1634 }
1635
1636 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1637 {
1638         unsigned long p = (unsigned long)bucket;
1639
1640         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1641 }
1642
1643 /* called with rt6_exception_lock held */
1644 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1645                                               spinlock_t *lock)
1646 {
1647         struct rt6_exception_bucket *bucket;
1648         unsigned long p;
1649
1650         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1651                                            lockdep_is_held(lock));
1652
1653         p = (unsigned long)bucket;
1654         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1655         bucket = (struct rt6_exception_bucket *)p;
1656         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1657 }
1658
1659 static int rt6_insert_exception(struct rt6_info *nrt,
1660                                 const struct fib6_result *res)
1661 {
1662         struct net *net = dev_net(nrt->dst.dev);
1663         struct rt6_exception_bucket *bucket;
1664         struct fib6_info *f6i = res->f6i;
1665         struct in6_addr *src_key = NULL;
1666         struct rt6_exception *rt6_ex;
1667         struct fib6_nh *nh = res->nh;
1668         int err = 0;
1669
1670         spin_lock_bh(&rt6_exception_lock);
1671
1672         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1673                                           lockdep_is_held(&rt6_exception_lock));
1674         if (!bucket) {
1675                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1676                                  GFP_ATOMIC);
1677                 if (!bucket) {
1678                         err = -ENOMEM;
1679                         goto out;
1680                 }
1681                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1682         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1683                 err = -EINVAL;
1684                 goto out;
1685         }
1686
1687 #ifdef CONFIG_IPV6_SUBTREES
1688         /* fib6_src.plen != 0 indicates f6i is in subtree
1689          * and exception table is indexed by a hash of
1690          * both fib6_dst and fib6_src.
1691          * Otherwise, the exception table is indexed by
1692          * a hash of only fib6_dst.
1693          */
1694         if (f6i->fib6_src.plen)
1695                 src_key = &nrt->rt6i_src.addr;
1696 #endif
1697         /* rt6_mtu_change() might lower mtu on f6i.
1698          * Only insert this exception route if its mtu
1699          * is less than f6i's mtu value.
1700          */
1701         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1702                 err = -EINVAL;
1703                 goto out;
1704         }
1705
1706         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1707                                                src_key);
1708         if (rt6_ex)
1709                 rt6_remove_exception(bucket, rt6_ex);
1710
1711         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1712         if (!rt6_ex) {
1713                 err = -ENOMEM;
1714                 goto out;
1715         }
1716         rt6_ex->rt6i = nrt;
1717         rt6_ex->stamp = jiffies;
1718         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1719         bucket->depth++;
1720         net->ipv6.rt6_stats->fib_rt_cache++;
1721
1722         if (bucket->depth > FIB6_MAX_DEPTH)
1723                 rt6_exception_remove_oldest(bucket);
1724
1725 out:
1726         spin_unlock_bh(&rt6_exception_lock);
1727
1728         /* Update fn->fn_sernum to invalidate all cached dst */
1729         if (!err) {
1730                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1731                 fib6_update_sernum(net, f6i);
1732                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1733                 fib6_force_start_gc(net);
1734         }
1735
1736         return err;
1737 }
1738
1739 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1740 {
1741         struct rt6_exception_bucket *bucket;
1742         struct rt6_exception *rt6_ex;
1743         struct hlist_node *tmp;
1744         int i;
1745
1746         spin_lock_bh(&rt6_exception_lock);
1747
1748         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1749         if (!bucket)
1750                 goto out;
1751
1752         /* Prevent rt6_insert_exception() to recreate the bucket list */
1753         if (!from)
1754                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1755
1756         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1757                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1758                         if (!from ||
1759                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1760                                 rt6_remove_exception(bucket, rt6_ex);
1761                 }
1762                 WARN_ON_ONCE(!from && bucket->depth);
1763                 bucket++;
1764         }
1765 out:
1766         spin_unlock_bh(&rt6_exception_lock);
1767 }
1768
1769 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1770 {
1771         struct fib6_info *f6i = arg;
1772
1773         fib6_nh_flush_exceptions(nh, f6i);
1774
1775         return 0;
1776 }
1777
1778 void rt6_flush_exceptions(struct fib6_info *f6i)
1779 {
1780         if (f6i->nh)
1781                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1782                                          f6i);
1783         else
1784                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1785 }
1786
1787 /* Find cached rt in the hash table inside passed in rt
1788  * Caller has to hold rcu_read_lock()
1789  */
1790 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1791                                            const struct in6_addr *daddr,
1792                                            const struct in6_addr *saddr)
1793 {
1794         const struct in6_addr *src_key = NULL;
1795         struct rt6_exception_bucket *bucket;
1796         struct rt6_exception *rt6_ex;
1797         struct rt6_info *ret = NULL;
1798
1799 #ifdef CONFIG_IPV6_SUBTREES
1800         /* fib6i_src.plen != 0 indicates f6i is in subtree
1801          * and exception table is indexed by a hash of
1802          * both fib6_dst and fib6_src.
1803          * However, the src addr used to create the hash
1804          * might not be exactly the passed in saddr which
1805          * is a /128 addr from the flow.
1806          * So we need to use f6i->fib6_src to redo lookup
1807          * if the passed in saddr does not find anything.
1808          * (See the logic in ip6_rt_cache_alloc() on how
1809          * rt->rt6i_src is updated.)
1810          */
1811         if (res->f6i->fib6_src.plen)
1812                 src_key = saddr;
1813 find_ex:
1814 #endif
1815         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1816         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1817
1818         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1819                 ret = rt6_ex->rt6i;
1820
1821 #ifdef CONFIG_IPV6_SUBTREES
1822         /* Use fib6_src as src_key and redo lookup */
1823         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1824                 src_key = &res->f6i->fib6_src.addr;
1825                 goto find_ex;
1826         }
1827 #endif
1828
1829         return ret;
1830 }
1831
1832 /* Remove the passed in cached rt from the hash table that contains it */
1833 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1834                                     const struct rt6_info *rt)
1835 {
1836         const struct in6_addr *src_key = NULL;
1837         struct rt6_exception_bucket *bucket;
1838         struct rt6_exception *rt6_ex;
1839         int err;
1840
1841         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1842                 return -ENOENT;
1843
1844         spin_lock_bh(&rt6_exception_lock);
1845         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1846
1847 #ifdef CONFIG_IPV6_SUBTREES
1848         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1849          * and exception table is indexed by a hash of
1850          * both rt6i_dst and rt6i_src.
1851          * Otherwise, the exception table is indexed by
1852          * a hash of only rt6i_dst.
1853          */
1854         if (plen)
1855                 src_key = &rt->rt6i_src.addr;
1856 #endif
1857         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1858                                                &rt->rt6i_dst.addr,
1859                                                src_key);
1860         if (rt6_ex) {
1861                 rt6_remove_exception(bucket, rt6_ex);
1862                 err = 0;
1863         } else {
1864                 err = -ENOENT;
1865         }
1866
1867         spin_unlock_bh(&rt6_exception_lock);
1868         return err;
1869 }
1870
1871 struct fib6_nh_excptn_arg {
1872         struct rt6_info *rt;
1873         int             plen;
1874 };
1875
1876 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1877 {
1878         struct fib6_nh_excptn_arg *arg = _arg;
1879         int err;
1880
1881         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1882         if (err == 0)
1883                 return 1;
1884
1885         return 0;
1886 }
1887
1888 static int rt6_remove_exception_rt(struct rt6_info *rt)
1889 {
1890         struct fib6_info *from;
1891
1892         from = rcu_dereference(rt->from);
1893         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1894                 return -EINVAL;
1895
1896         if (from->nh) {
1897                 struct fib6_nh_excptn_arg arg = {
1898                         .rt = rt,
1899                         .plen = from->fib6_src.plen
1900                 };
1901                 int rc;
1902
1903                 /* rc = 1 means an entry was found */
1904                 rc = nexthop_for_each_fib6_nh(from->nh,
1905                                               rt6_nh_remove_exception_rt,
1906                                               &arg);
1907                 return rc ? 0 : -ENOENT;
1908         }
1909
1910         return fib6_nh_remove_exception(from->fib6_nh,
1911                                         from->fib6_src.plen, rt);
1912 }
1913
1914 /* Find rt6_ex which contains the passed in rt cache and
1915  * refresh its stamp
1916  */
1917 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1918                                      const struct rt6_info *rt)
1919 {
1920         const struct in6_addr *src_key = NULL;
1921         struct rt6_exception_bucket *bucket;
1922         struct rt6_exception *rt6_ex;
1923
1924         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1925 #ifdef CONFIG_IPV6_SUBTREES
1926         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1927          * and exception table is indexed by a hash of
1928          * both rt6i_dst and rt6i_src.
1929          * Otherwise, the exception table is indexed by
1930          * a hash of only rt6i_dst.
1931          */
1932         if (plen)
1933                 src_key = &rt->rt6i_src.addr;
1934 #endif
1935         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1936         if (rt6_ex)
1937                 rt6_ex->stamp = jiffies;
1938 }
1939
1940 struct fib6_nh_match_arg {
1941         const struct net_device *dev;
1942         const struct in6_addr   *gw;
1943         struct fib6_nh          *match;
1944 };
1945
1946 /* determine if fib6_nh has given device and gateway */
1947 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1948 {
1949         struct fib6_nh_match_arg *arg = _arg;
1950
1951         if (arg->dev != nh->fib_nh_dev ||
1952             (arg->gw && !nh->fib_nh_gw_family) ||
1953             (!arg->gw && nh->fib_nh_gw_family) ||
1954             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1955                 return 0;
1956
1957         arg->match = nh;
1958
1959         /* found a match, break the loop */
1960         return 1;
1961 }
1962
1963 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1964 {
1965         struct fib6_info *from;
1966         struct fib6_nh *fib6_nh;
1967
1968         rcu_read_lock();
1969
1970         from = rcu_dereference(rt->from);
1971         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1972                 goto unlock;
1973
1974         if (from->nh) {
1975                 struct fib6_nh_match_arg arg = {
1976                         .dev = rt->dst.dev,
1977                         .gw = &rt->rt6i_gateway,
1978                 };
1979
1980                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1981
1982                 if (!arg.match)
1983                         goto unlock;
1984                 fib6_nh = arg.match;
1985         } else {
1986                 fib6_nh = from->fib6_nh;
1987         }
1988         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1989 unlock:
1990         rcu_read_unlock();
1991 }
1992
1993 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1994                                          struct rt6_info *rt, int mtu)
1995 {
1996         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1997          * lowest MTU in the path: always allow updating the route PMTU to
1998          * reflect PMTU decreases.
1999          *
2000          * If the new MTU is higher, and the route PMTU is equal to the local
2001          * MTU, this means the old MTU is the lowest in the path, so allow
2002          * updating it: if other nodes now have lower MTUs, PMTU discovery will
2003          * handle this.
2004          */
2005
2006         if (dst_mtu(&rt->dst) >= mtu)
2007                 return true;
2008
2009         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2010                 return true;
2011
2012         return false;
2013 }
2014
2015 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2016                                        const struct fib6_nh *nh, int mtu)
2017 {
2018         struct rt6_exception_bucket *bucket;
2019         struct rt6_exception *rt6_ex;
2020         int i;
2021
2022         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2023         if (!bucket)
2024                 return;
2025
2026         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2027                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2028                         struct rt6_info *entry = rt6_ex->rt6i;
2029
2030                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2031                          * route), the metrics of its rt->from have already
2032                          * been updated.
2033                          */
2034                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2035                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2036                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2037                 }
2038                 bucket++;
2039         }
2040 }
2041
2042 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2043
2044 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2045                                             const struct in6_addr *gateway)
2046 {
2047         struct rt6_exception_bucket *bucket;
2048         struct rt6_exception *rt6_ex;
2049         struct hlist_node *tmp;
2050         int i;
2051
2052         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2053                 return;
2054
2055         spin_lock_bh(&rt6_exception_lock);
2056         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2057         if (bucket) {
2058                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2059                         hlist_for_each_entry_safe(rt6_ex, tmp,
2060                                                   &bucket->chain, hlist) {
2061                                 struct rt6_info *entry = rt6_ex->rt6i;
2062
2063                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2064                                     RTF_CACHE_GATEWAY &&
2065                                     ipv6_addr_equal(gateway,
2066                                                     &entry->rt6i_gateway)) {
2067                                         rt6_remove_exception(bucket, rt6_ex);
2068                                 }
2069                         }
2070                         bucket++;
2071                 }
2072         }
2073
2074         spin_unlock_bh(&rt6_exception_lock);
2075 }
2076
2077 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2078                                       struct rt6_exception *rt6_ex,
2079                                       struct fib6_gc_args *gc_args,
2080                                       unsigned long now)
2081 {
2082         struct rt6_info *rt = rt6_ex->rt6i;
2083
2084         /* we are pruning and obsoleting aged-out and non gateway exceptions
2085          * even if others have still references to them, so that on next
2086          * dst_check() such references can be dropped.
2087          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2088          * expired, independently from their aging, as per RFC 8201 section 4
2089          */
2090         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2091                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2092                         RT6_TRACE("aging clone %p\n", rt);
2093                         rt6_remove_exception(bucket, rt6_ex);
2094                         return;
2095                 }
2096         } else if (time_after(jiffies, rt->dst.expires)) {
2097                 RT6_TRACE("purging expired route %p\n", rt);
2098                 rt6_remove_exception(bucket, rt6_ex);
2099                 return;
2100         }
2101
2102         if (rt->rt6i_flags & RTF_GATEWAY) {
2103                 struct neighbour *neigh;
2104                 __u8 neigh_flags = 0;
2105
2106                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2107                 if (neigh)
2108                         neigh_flags = neigh->flags;
2109
2110                 if (!(neigh_flags & NTF_ROUTER)) {
2111                         RT6_TRACE("purging route %p via non-router but gateway\n",
2112                                   rt);
2113                         rt6_remove_exception(bucket, rt6_ex);
2114                         return;
2115                 }
2116         }
2117
2118         gc_args->more++;
2119 }
2120
2121 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2122                                    struct fib6_gc_args *gc_args,
2123                                    unsigned long now)
2124 {
2125         struct rt6_exception_bucket *bucket;
2126         struct rt6_exception *rt6_ex;
2127         struct hlist_node *tmp;
2128         int i;
2129
2130         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2131                 return;
2132
2133         rcu_read_lock_bh();
2134         spin_lock(&rt6_exception_lock);
2135         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2136         if (bucket) {
2137                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2138                         hlist_for_each_entry_safe(rt6_ex, tmp,
2139                                                   &bucket->chain, hlist) {
2140                                 rt6_age_examine_exception(bucket, rt6_ex,
2141                                                           gc_args, now);
2142                         }
2143                         bucket++;
2144                 }
2145         }
2146         spin_unlock(&rt6_exception_lock);
2147         rcu_read_unlock_bh();
2148 }
2149
2150 struct fib6_nh_age_excptn_arg {
2151         struct fib6_gc_args     *gc_args;
2152         unsigned long           now;
2153 };
2154
2155 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2156 {
2157         struct fib6_nh_age_excptn_arg *arg = _arg;
2158
2159         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2160         return 0;
2161 }
2162
2163 void rt6_age_exceptions(struct fib6_info *f6i,
2164                         struct fib6_gc_args *gc_args,
2165                         unsigned long now)
2166 {
2167         if (f6i->nh) {
2168                 struct fib6_nh_age_excptn_arg arg = {
2169                         .gc_args = gc_args,
2170                         .now = now
2171                 };
2172
2173                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2174                                          &arg);
2175         } else {
2176                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2177         }
2178 }
2179
2180 /* must be called with rcu lock held */
2181 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2182                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2183 {
2184         struct fib6_node *fn, *saved_fn;
2185
2186         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2187         saved_fn = fn;
2188
2189         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2190                 oif = 0;
2191
2192 redo_rt6_select:
2193         rt6_select(net, fn, oif, res, strict);
2194         if (res->f6i == net->ipv6.fib6_null_entry) {
2195                 fn = fib6_backtrack(fn, &fl6->saddr);
2196                 if (fn)
2197                         goto redo_rt6_select;
2198                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2199                         /* also consider unreachable route */
2200                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2201                         fn = saved_fn;
2202                         goto redo_rt6_select;
2203                 }
2204         }
2205
2206         trace_fib6_table_lookup(net, res, table, fl6);
2207
2208         return 0;
2209 }
2210
2211 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2212                                int oif, struct flowi6 *fl6,
2213                                const struct sk_buff *skb, int flags)
2214 {
2215         struct fib6_result res = {};
2216         struct rt6_info *rt = NULL;
2217         int strict = 0;
2218
2219         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2220                      !rcu_read_lock_held());
2221
2222         strict |= flags & RT6_LOOKUP_F_IFACE;
2223         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2224         if (net->ipv6.devconf_all->forwarding == 0)
2225                 strict |= RT6_LOOKUP_F_REACHABLE;
2226
2227         rcu_read_lock();
2228
2229         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2230         if (res.f6i == net->ipv6.fib6_null_entry)
2231                 goto out;
2232
2233         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2234
2235         /*Search through exception table */
2236         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2237         if (rt) {
2238                 goto out;
2239         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2240                             !res.nh->fib_nh_gw_family)) {
2241                 /* Create a RTF_CACHE clone which will not be
2242                  * owned by the fib6 tree.  It is for the special case where
2243                  * the daddr in the skb during the neighbor look-up is different
2244                  * from the fl6->daddr used to look-up route here.
2245                  */
2246                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2247
2248                 if (rt) {
2249                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2250                          * As rt6_uncached_list_add() does not consume refcnt,
2251                          * this refcnt is always returned to the caller even
2252                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2253                          */
2254                         rt6_uncached_list_add(rt);
2255                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2256                         rcu_read_unlock();
2257
2258                         return rt;
2259                 }
2260         } else {
2261                 /* Get a percpu copy */
2262                 local_bh_disable();
2263                 rt = rt6_get_pcpu_route(&res);
2264
2265                 if (!rt)
2266                         rt = rt6_make_pcpu_route(net, &res);
2267
2268                 local_bh_enable();
2269         }
2270 out:
2271         if (!rt)
2272                 rt = net->ipv6.ip6_null_entry;
2273         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2274                 ip6_hold_safe(net, &rt);
2275         rcu_read_unlock();
2276
2277         return rt;
2278 }
2279 EXPORT_SYMBOL_GPL(ip6_pol_route);
2280
2281 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2282                                             struct fib6_table *table,
2283                                             struct flowi6 *fl6,
2284                                             const struct sk_buff *skb,
2285                                             int flags)
2286 {
2287         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2288 }
2289
2290 struct dst_entry *ip6_route_input_lookup(struct net *net,
2291                                          struct net_device *dev,
2292                                          struct flowi6 *fl6,
2293                                          const struct sk_buff *skb,
2294                                          int flags)
2295 {
2296         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2297                 flags |= RT6_LOOKUP_F_IFACE;
2298
2299         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2300 }
2301 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2302
2303 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2304                                   struct flow_keys *keys,
2305                                   struct flow_keys *flkeys)
2306 {
2307         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2308         const struct ipv6hdr *key_iph = outer_iph;
2309         struct flow_keys *_flkeys = flkeys;
2310         const struct ipv6hdr *inner_iph;
2311         const struct icmp6hdr *icmph;
2312         struct ipv6hdr _inner_iph;
2313         struct icmp6hdr _icmph;
2314
2315         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2316                 goto out;
2317
2318         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2319                                    sizeof(_icmph), &_icmph);
2320         if (!icmph)
2321                 goto out;
2322
2323         if (!icmpv6_is_err(icmph->icmp6_type))
2324                 goto out;
2325
2326         inner_iph = skb_header_pointer(skb,
2327                                        skb_transport_offset(skb) + sizeof(*icmph),
2328                                        sizeof(_inner_iph), &_inner_iph);
2329         if (!inner_iph)
2330                 goto out;
2331
2332         key_iph = inner_iph;
2333         _flkeys = NULL;
2334 out:
2335         if (_flkeys) {
2336                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2337                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2338                 keys->tags.flow_label = _flkeys->tags.flow_label;
2339                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2340         } else {
2341                 keys->addrs.v6addrs.src = key_iph->saddr;
2342                 keys->addrs.v6addrs.dst = key_iph->daddr;
2343                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2344                 keys->basic.ip_proto = key_iph->nexthdr;
2345         }
2346 }
2347
2348 /* if skb is set it will be used and fl6 can be NULL */
2349 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2350                        const struct sk_buff *skb, struct flow_keys *flkeys)
2351 {
2352         struct flow_keys hash_keys;
2353         u32 mhash;
2354
2355         switch (ip6_multipath_hash_policy(net)) {
2356         case 0:
2357                 memset(&hash_keys, 0, sizeof(hash_keys));
2358                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2359                 if (skb) {
2360                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2361                 } else {
2362                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2363                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2364                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2365                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2366                 }
2367                 break;
2368         case 1:
2369                 if (skb) {
2370                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2371                         struct flow_keys keys;
2372
2373                         /* short-circuit if we already have L4 hash present */
2374                         if (skb->l4_hash)
2375                                 return skb_get_hash_raw(skb) >> 1;
2376
2377                         memset(&hash_keys, 0, sizeof(hash_keys));
2378
2379                         if (!flkeys) {
2380                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2381                                 flkeys = &keys;
2382                         }
2383                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2384                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2385                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2386                         hash_keys.ports.src = flkeys->ports.src;
2387                         hash_keys.ports.dst = flkeys->ports.dst;
2388                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2389                 } else {
2390                         memset(&hash_keys, 0, sizeof(hash_keys));
2391                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2392                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2393                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2394                         hash_keys.ports.src = fl6->fl6_sport;
2395                         hash_keys.ports.dst = fl6->fl6_dport;
2396                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2397                 }
2398                 break;
2399         case 2:
2400                 memset(&hash_keys, 0, sizeof(hash_keys));
2401                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2402                 if (skb) {
2403                         struct flow_keys keys;
2404
2405                         if (!flkeys) {
2406                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2407                                 flkeys = &keys;
2408                         }
2409
2410                         /* Inner can be v4 or v6 */
2411                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2412                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2413                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2414                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2415                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2416                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2417                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2418                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2419                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2420                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2421                         } else {
2422                                 /* Same as case 0 */
2423                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2424                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2425                         }
2426                 } else {
2427                         /* Same as case 0 */
2428                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2429                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2430                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2431                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2432                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2433                 }
2434                 break;
2435         }
2436         mhash = flow_hash_from_keys(&hash_keys);
2437
2438         return mhash >> 1;
2439 }
2440
2441 /* Called with rcu held */
2442 void ip6_route_input(struct sk_buff *skb)
2443 {
2444         const struct ipv6hdr *iph = ipv6_hdr(skb);
2445         struct net *net = dev_net(skb->dev);
2446         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2447         struct ip_tunnel_info *tun_info;
2448         struct flowi6 fl6 = {
2449                 .flowi6_iif = skb->dev->ifindex,
2450                 .daddr = iph->daddr,
2451                 .saddr = iph->saddr,
2452                 .flowlabel = ip6_flowinfo(iph),
2453                 .flowi6_mark = skb->mark,
2454                 .flowi6_proto = iph->nexthdr,
2455         };
2456         struct flow_keys *flkeys = NULL, _flkeys;
2457
2458         tun_info = skb_tunnel_info(skb);
2459         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2460                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2461
2462         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2463                 flkeys = &_flkeys;
2464
2465         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2466                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2467         skb_dst_drop(skb);
2468         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2469                                                       &fl6, skb, flags));
2470 }
2471
2472 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2473                                              struct fib6_table *table,
2474                                              struct flowi6 *fl6,
2475                                              const struct sk_buff *skb,
2476                                              int flags)
2477 {
2478         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2479 }
2480
2481 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2482                                                const struct sock *sk,
2483                                                struct flowi6 *fl6, int flags)
2484 {
2485         bool any_src;
2486
2487         if (ipv6_addr_type(&fl6->daddr) &
2488             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2489                 struct dst_entry *dst;
2490
2491                 /* This function does not take refcnt on the dst */
2492                 dst = l3mdev_link_scope_lookup(net, fl6);
2493                 if (dst)
2494                         return dst;
2495         }
2496
2497         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2498
2499         flags |= RT6_LOOKUP_F_DST_NOREF;
2500         any_src = ipv6_addr_any(&fl6->saddr);
2501         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2502             (fl6->flowi6_oif && any_src))
2503                 flags |= RT6_LOOKUP_F_IFACE;
2504
2505         if (!any_src)
2506                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2507         else if (sk)
2508                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2509
2510         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2511 }
2512 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2513
2514 struct dst_entry *ip6_route_output_flags(struct net *net,
2515                                          const struct sock *sk,
2516                                          struct flowi6 *fl6,
2517                                          int flags)
2518 {
2519         struct dst_entry *dst;
2520         struct rt6_info *rt6;
2521
2522         rcu_read_lock();
2523         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2524         rt6 = (struct rt6_info *)dst;
2525         /* For dst cached in uncached_list, refcnt is already taken. */
2526         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2527                 dst = &net->ipv6.ip6_null_entry->dst;
2528                 dst_hold(dst);
2529         }
2530         rcu_read_unlock();
2531
2532         return dst;
2533 }
2534 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2535
2536 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2537 {
2538         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2539         struct net_device *loopback_dev = net->loopback_dev;
2540         struct dst_entry *new = NULL;
2541
2542         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2543                        DST_OBSOLETE_DEAD, 0);
2544         if (rt) {
2545                 rt6_info_init(rt);
2546                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2547
2548                 new = &rt->dst;
2549                 new->__use = 1;
2550                 new->input = dst_discard;
2551                 new->output = dst_discard_out;
2552
2553                 dst_copy_metrics(new, &ort->dst);
2554
2555                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2556                 rt->rt6i_gateway = ort->rt6i_gateway;
2557                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2558
2559                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2560 #ifdef CONFIG_IPV6_SUBTREES
2561                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2562 #endif
2563         }
2564
2565         dst_release(dst_orig);
2566         return new ? new : ERR_PTR(-ENOMEM);
2567 }
2568
2569 /*
2570  *      Destination cache support functions
2571  */
2572
2573 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2574 {
2575         u32 rt_cookie = 0;
2576
2577         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2578                 return false;
2579
2580         if (fib6_check_expired(f6i))
2581                 return false;
2582
2583         return true;
2584 }
2585
2586 static struct dst_entry *rt6_check(struct rt6_info *rt,
2587                                    struct fib6_info *from,
2588                                    u32 cookie)
2589 {
2590         u32 rt_cookie = 0;
2591
2592         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2593             rt_cookie != cookie)
2594                 return NULL;
2595
2596         if (rt6_check_expired(rt))
2597                 return NULL;
2598
2599         return &rt->dst;
2600 }
2601
2602 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2603                                             struct fib6_info *from,
2604                                             u32 cookie)
2605 {
2606         if (!__rt6_check_expired(rt) &&
2607             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2608             fib6_check(from, cookie))
2609                 return &rt->dst;
2610         else
2611                 return NULL;
2612 }
2613
2614 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2615 {
2616         struct dst_entry *dst_ret;
2617         struct fib6_info *from;
2618         struct rt6_info *rt;
2619
2620         rt = container_of(dst, struct rt6_info, dst);
2621
2622         if (rt->sernum)
2623                 return rt6_is_valid(rt) ? dst : NULL;
2624
2625         rcu_read_lock();
2626
2627         /* All IPV6 dsts are created with ->obsolete set to the value
2628          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2629          * into this function always.
2630          */
2631
2632         from = rcu_dereference(rt->from);
2633
2634         if (from && (rt->rt6i_flags & RTF_PCPU ||
2635             unlikely(!list_empty(&rt->rt6i_uncached))))
2636                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2637         else
2638                 dst_ret = rt6_check(rt, from, cookie);
2639
2640         rcu_read_unlock();
2641
2642         return dst_ret;
2643 }
2644
2645 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2646 {
2647         struct rt6_info *rt = (struct rt6_info *) dst;
2648
2649         if (rt) {
2650                 if (rt->rt6i_flags & RTF_CACHE) {
2651                         rcu_read_lock();
2652                         if (rt6_check_expired(rt)) {
2653                                 rt6_remove_exception_rt(rt);
2654                                 dst = NULL;
2655                         }
2656                         rcu_read_unlock();
2657                 } else {
2658                         dst_release(dst);
2659                         dst = NULL;
2660                 }
2661         }
2662         return dst;
2663 }
2664
2665 static void ip6_link_failure(struct sk_buff *skb)
2666 {
2667         struct rt6_info *rt;
2668
2669         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2670
2671         rt = (struct rt6_info *) skb_dst(skb);
2672         if (rt) {
2673                 rcu_read_lock();
2674                 if (rt->rt6i_flags & RTF_CACHE) {
2675                         rt6_remove_exception_rt(rt);
2676                 } else {
2677                         struct fib6_info *from;
2678                         struct fib6_node *fn;
2679
2680                         from = rcu_dereference(rt->from);
2681                         if (from) {
2682                                 fn = rcu_dereference(from->fib6_node);
2683                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2684                                         fn->fn_sernum = -1;
2685                         }
2686                 }
2687                 rcu_read_unlock();
2688         }
2689 }
2690
2691 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2692 {
2693         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2694                 struct fib6_info *from;
2695
2696                 rcu_read_lock();
2697                 from = rcu_dereference(rt0->from);
2698                 if (from)
2699                         rt0->dst.expires = from->expires;
2700                 rcu_read_unlock();
2701         }
2702
2703         dst_set_expires(&rt0->dst, timeout);
2704         rt0->rt6i_flags |= RTF_EXPIRES;
2705 }
2706
2707 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2708 {
2709         struct net *net = dev_net(rt->dst.dev);
2710
2711         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2712         rt->rt6i_flags |= RTF_MODIFIED;
2713         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2714 }
2715
2716 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2717 {
2718         return !(rt->rt6i_flags & RTF_CACHE) &&
2719                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2720 }
2721
2722 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2723                                  const struct ipv6hdr *iph, u32 mtu,
2724                                  bool confirm_neigh)
2725 {
2726         const struct in6_addr *daddr, *saddr;
2727         struct rt6_info *rt6 = (struct rt6_info *)dst;
2728
2729         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2730          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2731          * [see also comment in rt6_mtu_change_route()]
2732          */
2733
2734         if (iph) {
2735                 daddr = &iph->daddr;
2736                 saddr = &iph->saddr;
2737         } else if (sk) {
2738                 daddr = &sk->sk_v6_daddr;
2739                 saddr = &inet6_sk(sk)->saddr;
2740         } else {
2741                 daddr = NULL;
2742                 saddr = NULL;
2743         }
2744
2745         if (confirm_neigh)
2746                 dst_confirm_neigh(dst, daddr);
2747
2748         if (mtu < IPV6_MIN_MTU)
2749                 return;
2750         if (mtu >= dst_mtu(dst))
2751                 return;
2752
2753         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2754                 rt6_do_update_pmtu(rt6, mtu);
2755                 /* update rt6_ex->stamp for cache */
2756                 if (rt6->rt6i_flags & RTF_CACHE)
2757                         rt6_update_exception_stamp_rt(rt6);
2758         } else if (daddr) {
2759                 struct fib6_result res = {};
2760                 struct rt6_info *nrt6;
2761
2762                 rcu_read_lock();
2763                 res.f6i = rcu_dereference(rt6->from);
2764                 if (!res.f6i)
2765                         goto out_unlock;
2766
2767                 res.fib6_flags = res.f6i->fib6_flags;
2768                 res.fib6_type = res.f6i->fib6_type;
2769
2770                 if (res.f6i->nh) {
2771                         struct fib6_nh_match_arg arg = {
2772                                 .dev = dst->dev,
2773                                 .gw = &rt6->rt6i_gateway,
2774                         };
2775
2776                         nexthop_for_each_fib6_nh(res.f6i->nh,
2777                                                  fib6_nh_find_match, &arg);
2778
2779                         /* fib6_info uses a nexthop that does not have fib6_nh
2780                          * using the dst->dev + gw. Should be impossible.
2781                          */
2782                         if (!arg.match)
2783                                 goto out_unlock;
2784
2785                         res.nh = arg.match;
2786                 } else {
2787                         res.nh = res.f6i->fib6_nh;
2788                 }
2789
2790                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2791                 if (nrt6) {
2792                         rt6_do_update_pmtu(nrt6, mtu);
2793                         if (rt6_insert_exception(nrt6, &res))
2794                                 dst_release_immediate(&nrt6->dst);
2795                 }
2796 out_unlock:
2797                 rcu_read_unlock();
2798         }
2799 }
2800
2801 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2802                                struct sk_buff *skb, u32 mtu,
2803                                bool confirm_neigh)
2804 {
2805         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2806                              confirm_neigh);
2807 }
2808
2809 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2810                      int oif, u32 mark, kuid_t uid)
2811 {
2812         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2813         struct dst_entry *dst;
2814         struct flowi6 fl6 = {
2815                 .flowi6_oif = oif,
2816                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2817                 .daddr = iph->daddr,
2818                 .saddr = iph->saddr,
2819                 .flowlabel = ip6_flowinfo(iph),
2820                 .flowi6_uid = uid,
2821         };
2822
2823         dst = ip6_route_output(net, NULL, &fl6);
2824         if (!dst->error)
2825                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2826         dst_release(dst);
2827 }
2828 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2829
2830 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2831 {
2832         int oif = sk->sk_bound_dev_if;
2833         struct dst_entry *dst;
2834
2835         if (!oif && skb->dev)
2836                 oif = l3mdev_master_ifindex(skb->dev);
2837
2838         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2839
2840         dst = __sk_dst_get(sk);
2841         if (!dst || !dst->obsolete ||
2842             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2843                 return;
2844
2845         bh_lock_sock(sk);
2846         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2847                 ip6_datagram_dst_update(sk, false);
2848         bh_unlock_sock(sk);
2849 }
2850 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2851
2852 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2853                            const struct flowi6 *fl6)
2854 {
2855 #ifdef CONFIG_IPV6_SUBTREES
2856         struct ipv6_pinfo *np = inet6_sk(sk);
2857 #endif
2858
2859         ip6_dst_store(sk, dst,
2860                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2861                       &sk->sk_v6_daddr : NULL,
2862 #ifdef CONFIG_IPV6_SUBTREES
2863                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2864                       &np->saddr :
2865 #endif
2866                       NULL);
2867 }
2868
2869 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2870                                   struct flowi6 *fl6,
2871                                   const struct in6_addr *gw,
2872                                   struct rt6_info **ret)
2873 {
2874         const struct fib6_nh *nh = res->nh;
2875
2876         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2877             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2878                 return false;
2879
2880         /* rt_cache's gateway might be different from its 'parent'
2881          * in the case of an ip redirect.
2882          * So we keep searching in the exception table if the gateway
2883          * is different.
2884          */
2885         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2886                 struct rt6_info *rt_cache;
2887
2888                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2889                 if (rt_cache &&
2890                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2891                         *ret = rt_cache;
2892                         return true;
2893                 }
2894                 return false;
2895         }
2896         return true;
2897 }
2898
2899 struct fib6_nh_rd_arg {
2900         struct fib6_result      *res;
2901         struct flowi6           *fl6;
2902         const struct in6_addr   *gw;
2903         struct rt6_info         **ret;
2904 };
2905
2906 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2907 {
2908         struct fib6_nh_rd_arg *arg = _arg;
2909
2910         arg->res->nh = nh;
2911         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2912 }
2913
2914 /* Handle redirects */
2915 struct ip6rd_flowi {
2916         struct flowi6 fl6;
2917         struct in6_addr gateway;
2918 };
2919
2920 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
2921                                              struct fib6_table *table,
2922                                              struct flowi6 *fl6,
2923                                              const struct sk_buff *skb,
2924                                              int flags)
2925 {
2926         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2927         struct rt6_info *ret = NULL;
2928         struct fib6_result res = {};
2929         struct fib6_nh_rd_arg arg = {
2930                 .res = &res,
2931                 .fl6 = fl6,
2932                 .gw  = &rdfl->gateway,
2933                 .ret = &ret
2934         };
2935         struct fib6_info *rt;
2936         struct fib6_node *fn;
2937
2938         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2939          * this case we must match on the real ingress device, so reset it
2940          */
2941         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2942                 fl6->flowi6_oif = skb->dev->ifindex;
2943
2944         /* Get the "current" route for this destination and
2945          * check if the redirect has come from appropriate router.
2946          *
2947          * RFC 4861 specifies that redirects should only be
2948          * accepted if they come from the nexthop to the target.
2949          * Due to the way the routes are chosen, this notion
2950          * is a bit fuzzy and one might need to check all possible
2951          * routes.
2952          */
2953
2954         rcu_read_lock();
2955         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2956 restart:
2957         for_each_fib6_node_rt_rcu(fn) {
2958                 res.f6i = rt;
2959                 if (fib6_check_expired(rt))
2960                         continue;
2961                 if (rt->fib6_flags & RTF_REJECT)
2962                         break;
2963                 if (unlikely(rt->nh)) {
2964                         if (nexthop_is_blackhole(rt->nh))
2965                                 continue;
2966                         /* on match, res->nh is filled in and potentially ret */
2967                         if (nexthop_for_each_fib6_nh(rt->nh,
2968                                                      fib6_nh_redirect_match,
2969                                                      &arg))
2970                                 goto out;
2971                 } else {
2972                         res.nh = rt->fib6_nh;
2973                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2974                                                   &ret))
2975                                 goto out;
2976                 }
2977         }
2978
2979         if (!rt)
2980                 rt = net->ipv6.fib6_null_entry;
2981         else if (rt->fib6_flags & RTF_REJECT) {
2982                 ret = net->ipv6.ip6_null_entry;
2983                 goto out;
2984         }
2985
2986         if (rt == net->ipv6.fib6_null_entry) {
2987                 fn = fib6_backtrack(fn, &fl6->saddr);
2988                 if (fn)
2989                         goto restart;
2990         }
2991
2992         res.f6i = rt;
2993         res.nh = rt->fib6_nh;
2994 out:
2995         if (ret) {
2996                 ip6_hold_safe(net, &ret);
2997         } else {
2998                 res.fib6_flags = res.f6i->fib6_flags;
2999                 res.fib6_type = res.f6i->fib6_type;
3000                 ret = ip6_create_rt_rcu(&res);
3001         }
3002
3003         rcu_read_unlock();
3004
3005         trace_fib6_table_lookup(net, &res, table, fl6);
3006         return ret;
3007 };
3008
3009 static struct dst_entry *ip6_route_redirect(struct net *net,
3010                                             const struct flowi6 *fl6,
3011                                             const struct sk_buff *skb,
3012                                             const struct in6_addr *gateway)
3013 {
3014         int flags = RT6_LOOKUP_F_HAS_SADDR;
3015         struct ip6rd_flowi rdfl;
3016
3017         rdfl.fl6 = *fl6;
3018         rdfl.gateway = *gateway;
3019
3020         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3021                                 flags, __ip6_route_redirect);
3022 }
3023
3024 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3025                   kuid_t uid)
3026 {
3027         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3028         struct dst_entry *dst;
3029         struct flowi6 fl6 = {
3030                 .flowi6_iif = LOOPBACK_IFINDEX,
3031                 .flowi6_oif = oif,
3032                 .flowi6_mark = mark,
3033                 .daddr = iph->daddr,
3034                 .saddr = iph->saddr,
3035                 .flowlabel = ip6_flowinfo(iph),
3036                 .flowi6_uid = uid,
3037         };
3038
3039         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3040         rt6_do_redirect(dst, NULL, skb);
3041         dst_release(dst);
3042 }
3043 EXPORT_SYMBOL_GPL(ip6_redirect);
3044
3045 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3046 {
3047         const struct ipv6hdr *iph = ipv6_hdr(skb);
3048         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3049         struct dst_entry *dst;
3050         struct flowi6 fl6 = {
3051                 .flowi6_iif = LOOPBACK_IFINDEX,
3052                 .flowi6_oif = oif,
3053                 .daddr = msg->dest,
3054                 .saddr = iph->daddr,
3055                 .flowi6_uid = sock_net_uid(net, NULL),
3056         };
3057
3058         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3059         rt6_do_redirect(dst, NULL, skb);
3060         dst_release(dst);
3061 }
3062
3063 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3064 {
3065         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3066                      sk->sk_uid);
3067 }
3068 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3069
3070 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3071 {
3072         struct net_device *dev = dst->dev;
3073         unsigned int mtu = dst_mtu(dst);
3074         struct net *net = dev_net(dev);
3075
3076         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3077
3078         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3079                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3080
3081         /*
3082          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3083          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3084          * IPV6_MAXPLEN is also valid and means: "any MSS,
3085          * rely only on pmtu discovery"
3086          */
3087         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3088                 mtu = IPV6_MAXPLEN;
3089         return mtu;
3090 }
3091
3092 static unsigned int ip6_mtu(const struct dst_entry *dst)
3093 {
3094         struct inet6_dev *idev;
3095         unsigned int mtu;
3096
3097         mtu = dst_metric_raw(dst, RTAX_MTU);
3098         if (mtu)
3099                 goto out;
3100
3101         mtu = IPV6_MIN_MTU;
3102
3103         rcu_read_lock();
3104         idev = __in6_dev_get(dst->dev);
3105         if (idev)
3106                 mtu = idev->cnf.mtu6;
3107         rcu_read_unlock();
3108
3109 out:
3110         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3111
3112         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3113 }
3114
3115 /* MTU selection:
3116  * 1. mtu on route is locked - use it
3117  * 2. mtu from nexthop exception
3118  * 3. mtu from egress device
3119  *
3120  * based on ip6_dst_mtu_forward and exception logic of
3121  * rt6_find_cached_rt; called with rcu_read_lock
3122  */
3123 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3124                       const struct in6_addr *daddr,
3125                       const struct in6_addr *saddr)
3126 {
3127         const struct fib6_nh *nh = res->nh;
3128         struct fib6_info *f6i = res->f6i;
3129         struct inet6_dev *idev;
3130         struct rt6_info *rt;
3131         u32 mtu = 0;
3132
3133         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3134                 mtu = f6i->fib6_pmtu;
3135                 if (mtu)
3136                         goto out;
3137         }
3138
3139         rt = rt6_find_cached_rt(res, daddr, saddr);
3140         if (unlikely(rt)) {
3141                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3142         } else {
3143                 struct net_device *dev = nh->fib_nh_dev;
3144
3145                 mtu = IPV6_MIN_MTU;
3146                 idev = __in6_dev_get(dev);
3147                 if (idev && idev->cnf.mtu6 > mtu)
3148                         mtu = idev->cnf.mtu6;
3149         }
3150
3151         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3152 out:
3153         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3154 }
3155
3156 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3157                                   struct flowi6 *fl6)
3158 {
3159         struct dst_entry *dst;
3160         struct rt6_info *rt;
3161         struct inet6_dev *idev = in6_dev_get(dev);
3162         struct net *net = dev_net(dev);
3163
3164         if (unlikely(!idev))
3165                 return ERR_PTR(-ENODEV);
3166
3167         rt = ip6_dst_alloc(net, dev, 0);
3168         if (unlikely(!rt)) {
3169                 in6_dev_put(idev);
3170                 dst = ERR_PTR(-ENOMEM);
3171                 goto out;
3172         }
3173
3174         rt->dst.input = ip6_input;
3175         rt->dst.output  = ip6_output;
3176         rt->rt6i_gateway  = fl6->daddr;
3177         rt->rt6i_dst.addr = fl6->daddr;
3178         rt->rt6i_dst.plen = 128;
3179         rt->rt6i_idev     = idev;
3180         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3181
3182         /* Add this dst into uncached_list so that rt6_disable_ip() can
3183          * do proper release of the net_device
3184          */
3185         rt6_uncached_list_add(rt);
3186         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3187
3188         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3189
3190 out:
3191         return dst;
3192 }
3193
3194 static int ip6_dst_gc(struct dst_ops *ops)
3195 {
3196         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3197         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3198         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3199         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3200         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3201         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3202         int entries;
3203
3204         entries = dst_entries_get_fast(ops);
3205         if (entries > rt_max_size)
3206                 entries = dst_entries_get_slow(ops);
3207
3208         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3209             entries <= rt_max_size)
3210                 goto out;
3211
3212         net->ipv6.ip6_rt_gc_expire++;
3213         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3214         entries = dst_entries_get_slow(ops);
3215         if (entries < ops->gc_thresh)
3216                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3217 out:
3218         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3219         return entries > rt_max_size;
3220 }
3221
3222 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3223                                const struct in6_addr *gw_addr, u32 tbid,
3224                                int flags, struct fib6_result *res)
3225 {
3226         struct flowi6 fl6 = {
3227                 .flowi6_oif = cfg->fc_ifindex,
3228                 .daddr = *gw_addr,
3229                 .saddr = cfg->fc_prefsrc,
3230         };
3231         struct fib6_table *table;
3232         int err;
3233
3234         table = fib6_get_table(net, tbid);
3235         if (!table)
3236                 return -EINVAL;
3237
3238         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3239                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3240
3241         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3242
3243         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3244         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3245                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3246                                  cfg->fc_ifindex != 0, NULL, flags);
3247
3248         return err;
3249 }
3250
3251 static int ip6_route_check_nh_onlink(struct net *net,
3252                                      struct fib6_config *cfg,
3253                                      const struct net_device *dev,
3254                                      struct netlink_ext_ack *extack)
3255 {
3256         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3257         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3258         struct fib6_result res = {};
3259         int err;
3260
3261         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3262         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3263             /* ignore match if it is the default route */
3264             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3265             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3266                 NL_SET_ERR_MSG(extack,
3267                                "Nexthop has invalid gateway or device mismatch");
3268                 err = -EINVAL;
3269         }
3270
3271         return err;
3272 }
3273
3274 static int ip6_route_check_nh(struct net *net,
3275                               struct fib6_config *cfg,
3276                               struct net_device **_dev,
3277                               struct inet6_dev **idev)
3278 {
3279         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3280         struct net_device *dev = _dev ? *_dev : NULL;
3281         int flags = RT6_LOOKUP_F_IFACE;
3282         struct fib6_result res = {};
3283         int err = -EHOSTUNREACH;
3284
3285         if (cfg->fc_table) {
3286                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3287                                           cfg->fc_table, flags, &res);
3288                 /* gw_addr can not require a gateway or resolve to a reject
3289                  * route. If a device is given, it must match the result.
3290                  */
3291                 if (err || res.fib6_flags & RTF_REJECT ||
3292                     res.nh->fib_nh_gw_family ||
3293                     (dev && dev != res.nh->fib_nh_dev))
3294                         err = -EHOSTUNREACH;
3295         }
3296
3297         if (err < 0) {
3298                 struct flowi6 fl6 = {
3299                         .flowi6_oif = cfg->fc_ifindex,
3300                         .daddr = *gw_addr,
3301                 };
3302
3303                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3304                 if (err || res.fib6_flags & RTF_REJECT ||
3305                     res.nh->fib_nh_gw_family)
3306                         err = -EHOSTUNREACH;
3307
3308                 if (err)
3309                         return err;
3310
3311                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3312                                  cfg->fc_ifindex != 0, NULL, flags);
3313         }
3314
3315         err = 0;
3316         if (dev) {
3317                 if (dev != res.nh->fib_nh_dev)
3318                         err = -EHOSTUNREACH;
3319         } else {
3320                 *_dev = dev = res.nh->fib_nh_dev;
3321                 dev_hold(dev);
3322                 *idev = in6_dev_get(dev);
3323         }
3324
3325         return err;
3326 }
3327
3328 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3329                            struct net_device **_dev, struct inet6_dev **idev,
3330                            struct netlink_ext_ack *extack)
3331 {
3332         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3333         int gwa_type = ipv6_addr_type(gw_addr);
3334         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3335         const struct net_device *dev = *_dev;
3336         bool need_addr_check = !dev;
3337         int err = -EINVAL;
3338
3339         /* if gw_addr is local we will fail to detect this in case
3340          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3341          * will return already-added prefix route via interface that
3342          * prefix route was assigned to, which might be non-loopback.
3343          */
3344         if (dev &&
3345             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3346                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3347                 goto out;
3348         }
3349
3350         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3351                 /* IPv6 strictly inhibits using not link-local
3352                  * addresses as nexthop address.
3353                  * Otherwise, router will not able to send redirects.
3354                  * It is very good, but in some (rare!) circumstances
3355                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3356                  * some exceptions. --ANK
3357                  * We allow IPv4-mapped nexthops to support RFC4798-type
3358                  * addressing
3359                  */
3360                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3361                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3362                         goto out;
3363                 }
3364
3365                 rcu_read_lock();
3366
3367                 if (cfg->fc_flags & RTNH_F_ONLINK)
3368                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3369                 else
3370                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3371
3372                 rcu_read_unlock();
3373
3374                 if (err)
3375                         goto out;
3376         }
3377
3378         /* reload in case device was changed */
3379         dev = *_dev;
3380
3381         err = -EINVAL;
3382         if (!dev) {
3383                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3384                 goto out;
3385         } else if (dev->flags & IFF_LOOPBACK) {
3386                 NL_SET_ERR_MSG(extack,
3387                                "Egress device can not be loopback device for this route");
3388                 goto out;
3389         }
3390
3391         /* if we did not check gw_addr above, do so now that the
3392          * egress device has been resolved.
3393          */
3394         if (need_addr_check &&
3395             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3396                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3397                 goto out;
3398         }
3399
3400         err = 0;
3401 out:
3402         return err;
3403 }
3404
3405 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3406 {
3407         if ((flags & RTF_REJECT) ||
3408             (dev && (dev->flags & IFF_LOOPBACK) &&
3409              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3410              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3411                 return true;
3412
3413         return false;
3414 }
3415
3416 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3417                  struct fib6_config *cfg, gfp_t gfp_flags,
3418                  struct netlink_ext_ack *extack)
3419 {
3420         struct net_device *dev = NULL;
3421         struct inet6_dev *idev = NULL;
3422         int addr_type;
3423         int err;
3424
3425         fib6_nh->fib_nh_family = AF_INET6;
3426 #ifdef CONFIG_IPV6_ROUTER_PREF
3427         fib6_nh->last_probe = jiffies;
3428 #endif
3429         if (cfg->fc_is_fdb) {
3430                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3431                 fib6_nh->fib_nh_gw_family = AF_INET6;
3432                 return 0;
3433         }
3434
3435         err = -ENODEV;
3436         if (cfg->fc_ifindex) {
3437                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3438                 if (!dev)
3439                         goto out;
3440                 idev = in6_dev_get(dev);
3441                 if (!idev)
3442                         goto out;
3443         }
3444
3445         if (cfg->fc_flags & RTNH_F_ONLINK) {
3446                 if (!dev) {
3447                         NL_SET_ERR_MSG(extack,
3448                                        "Nexthop device required for onlink");
3449                         goto out;
3450                 }
3451
3452                 if (!(dev->flags & IFF_UP)) {
3453                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3454                         err = -ENETDOWN;
3455                         goto out;
3456                 }
3457
3458                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3459         }
3460
3461         fib6_nh->fib_nh_weight = 1;
3462
3463         /* We cannot add true routes via loopback here,
3464          * they would result in kernel looping; promote them to reject routes
3465          */
3466         addr_type = ipv6_addr_type(&cfg->fc_dst);
3467         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3468                 /* hold loopback dev/idev if we haven't done so. */
3469                 if (dev != net->loopback_dev) {
3470                         if (dev) {
3471                                 dev_put(dev);
3472                                 in6_dev_put(idev);
3473                         }
3474                         dev = net->loopback_dev;
3475                         dev_hold(dev);
3476                         idev = in6_dev_get(dev);
3477                         if (!idev) {
3478                                 err = -ENODEV;
3479                                 goto out;
3480                         }
3481                 }
3482                 goto pcpu_alloc;
3483         }
3484
3485         if (cfg->fc_flags & RTF_GATEWAY) {
3486                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3487                 if (err)
3488                         goto out;
3489
3490                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3491                 fib6_nh->fib_nh_gw_family = AF_INET6;
3492         }
3493
3494         err = -ENODEV;
3495         if (!dev)
3496                 goto out;
3497
3498         if (idev->cnf.disable_ipv6) {
3499                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3500                 err = -EACCES;
3501                 goto out;
3502         }
3503
3504         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3505                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3506                 err = -ENETDOWN;
3507                 goto out;
3508         }
3509
3510         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3511             !netif_carrier_ok(dev))
3512                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3513
3514         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3515                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3516         if (err)
3517                 goto out;
3518
3519 pcpu_alloc:
3520         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3521         if (!fib6_nh->rt6i_pcpu) {
3522                 err = -ENOMEM;
3523                 goto out;
3524         }
3525
3526         fib6_nh->fib_nh_dev = dev;
3527         fib6_nh->fib_nh_oif = dev->ifindex;
3528         err = 0;
3529 out:
3530         if (idev)
3531                 in6_dev_put(idev);
3532
3533         if (err) {
3534                 lwtstate_put(fib6_nh->fib_nh_lws);
3535                 fib6_nh->fib_nh_lws = NULL;
3536                 if (dev)
3537                         dev_put(dev);
3538         }
3539
3540         return err;
3541 }
3542
3543 void fib6_nh_release(struct fib6_nh *fib6_nh)
3544 {
3545         struct rt6_exception_bucket *bucket;
3546
3547         rcu_read_lock();
3548
3549         fib6_nh_flush_exceptions(fib6_nh, NULL);
3550         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3551         if (bucket) {
3552                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3553                 kfree(bucket);
3554         }
3555
3556         rcu_read_unlock();
3557
3558         if (fib6_nh->rt6i_pcpu) {
3559                 int cpu;
3560
3561                 for_each_possible_cpu(cpu) {
3562                         struct rt6_info **ppcpu_rt;
3563                         struct rt6_info *pcpu_rt;
3564
3565                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3566                         pcpu_rt = *ppcpu_rt;
3567                         if (pcpu_rt) {
3568                                 dst_dev_put(&pcpu_rt->dst);
3569                                 dst_release(&pcpu_rt->dst);
3570                                 *ppcpu_rt = NULL;
3571                         }
3572                 }
3573
3574                 free_percpu(fib6_nh->rt6i_pcpu);
3575         }
3576
3577         fib_nh_common_release(&fib6_nh->nh_common);
3578 }
3579
3580 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3581                                               gfp_t gfp_flags,
3582                                               struct netlink_ext_ack *extack)
3583 {
3584         struct net *net = cfg->fc_nlinfo.nl_net;
3585         struct fib6_info *rt = NULL;
3586         struct nexthop *nh = NULL;
3587         struct fib6_table *table;
3588         struct fib6_nh *fib6_nh;
3589         int err = -EINVAL;
3590         int addr_type;
3591
3592         /* RTF_PCPU is an internal flag; can not be set by userspace */
3593         if (cfg->fc_flags & RTF_PCPU) {
3594                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3595                 goto out;
3596         }
3597
3598         /* RTF_CACHE is an internal flag; can not be set by userspace */
3599         if (cfg->fc_flags & RTF_CACHE) {
3600                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3601                 goto out;
3602         }
3603
3604         if (cfg->fc_type > RTN_MAX) {
3605                 NL_SET_ERR_MSG(extack, "Invalid route type");
3606                 goto out;
3607         }
3608
3609         if (cfg->fc_dst_len > 128) {
3610                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3611                 goto out;
3612         }
3613         if (cfg->fc_src_len > 128) {
3614                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3615                 goto out;
3616         }
3617 #ifndef CONFIG_IPV6_SUBTREES
3618         if (cfg->fc_src_len) {
3619                 NL_SET_ERR_MSG(extack,
3620                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3621                 goto out;
3622         }
3623 #endif
3624         if (cfg->fc_nh_id) {
3625                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3626                 if (!nh) {
3627                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3628                         goto out;
3629                 }
3630                 err = fib6_check_nexthop(nh, cfg, extack);
3631                 if (err)
3632                         goto out;
3633         }
3634
3635         err = -ENOBUFS;
3636         if (cfg->fc_nlinfo.nlh &&
3637             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3638                 table = fib6_get_table(net, cfg->fc_table);
3639                 if (!table) {
3640                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3641                         table = fib6_new_table(net, cfg->fc_table);
3642                 }
3643         } else {
3644                 table = fib6_new_table(net, cfg->fc_table);
3645         }
3646
3647         if (!table)
3648                 goto out;
3649
3650         err = -ENOMEM;
3651         rt = fib6_info_alloc(gfp_flags, !nh);
3652         if (!rt)
3653                 goto out;
3654
3655         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3656                                                extack);
3657         if (IS_ERR(rt->fib6_metrics)) {
3658                 err = PTR_ERR(rt->fib6_metrics);
3659                 /* Do not leave garbage there. */
3660                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3661                 goto out;
3662         }
3663
3664         if (cfg->fc_flags & RTF_ADDRCONF)
3665                 rt->dst_nocount = true;
3666
3667         if (cfg->fc_flags & RTF_EXPIRES)
3668                 fib6_set_expires(rt, jiffies +
3669                                 clock_t_to_jiffies(cfg->fc_expires));
3670         else
3671                 fib6_clean_expires(rt);
3672
3673         if (cfg->fc_protocol == RTPROT_UNSPEC)
3674                 cfg->fc_protocol = RTPROT_BOOT;
3675         rt->fib6_protocol = cfg->fc_protocol;
3676
3677         rt->fib6_table = table;
3678         rt->fib6_metric = cfg->fc_metric;
3679         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3680         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3681
3682         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3683         rt->fib6_dst.plen = cfg->fc_dst_len;
3684
3685 #ifdef CONFIG_IPV6_SUBTREES
3686         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3687         rt->fib6_src.plen = cfg->fc_src_len;
3688 #endif
3689         if (nh) {
3690                 if (rt->fib6_src.plen) {
3691                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3692                         goto out;
3693                 }
3694                 if (!nexthop_get(nh)) {
3695                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3696                         goto out;
3697                 }
3698                 rt->nh = nh;
3699                 fib6_nh = nexthop_fib6_nh(rt->nh);
3700         } else {
3701                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3702                 if (err)
3703                         goto out;
3704
3705                 fib6_nh = rt->fib6_nh;
3706
3707                 /* We cannot add true routes via loopback here, they would
3708                  * result in kernel looping; promote them to reject routes
3709                  */
3710                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3711                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3712                                    addr_type))
3713                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3714         }
3715
3716         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3717                 struct net_device *dev = fib6_nh->fib_nh_dev;
3718
3719                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3720                         NL_SET_ERR_MSG(extack, "Invalid source address");
3721                         err = -EINVAL;
3722                         goto out;
3723                 }
3724                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3725                 rt->fib6_prefsrc.plen = 128;
3726         } else
3727                 rt->fib6_prefsrc.plen = 0;
3728
3729         return rt;
3730 out:
3731         fib6_info_release(rt);
3732         return ERR_PTR(err);
3733 }
3734
3735 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3736                   struct netlink_ext_ack *extack)
3737 {
3738         struct fib6_info *rt;
3739         int err;
3740
3741         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3742         if (IS_ERR(rt))
3743                 return PTR_ERR(rt);
3744
3745         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3746         fib6_info_release(rt);
3747
3748         return err;
3749 }
3750
3751 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3752 {
3753         struct net *net = info->nl_net;
3754         struct fib6_table *table;
3755         int err;
3756
3757         if (rt == net->ipv6.fib6_null_entry) {
3758                 err = -ENOENT;
3759                 goto out;
3760         }
3761
3762         table = rt->fib6_table;
3763         spin_lock_bh(&table->tb6_lock);
3764         err = fib6_del(rt, info);
3765         spin_unlock_bh(&table->tb6_lock);
3766
3767 out:
3768         fib6_info_release(rt);
3769         return err;
3770 }
3771
3772 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3773 {
3774         struct nl_info info = {
3775                 .nl_net = net,
3776                 .skip_notify = skip_notify
3777         };
3778
3779         return __ip6_del_rt(rt, &info);
3780 }
3781
3782 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3783 {
3784         struct nl_info *info = &cfg->fc_nlinfo;
3785         struct net *net = info->nl_net;
3786         struct sk_buff *skb = NULL;
3787         struct fib6_table *table;
3788         int err = -ENOENT;
3789
3790         if (rt == net->ipv6.fib6_null_entry)
3791                 goto out_put;
3792         table = rt->fib6_table;
3793         spin_lock_bh(&table->tb6_lock);
3794
3795         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3796                 struct fib6_info *sibling, *next_sibling;
3797                 struct fib6_node *fn;
3798
3799                 /* prefer to send a single notification with all hops */
3800                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3801                 if (skb) {
3802                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3803
3804                         if (rt6_fill_node(net, skb, rt, NULL,
3805                                           NULL, NULL, 0, RTM_DELROUTE,
3806                                           info->portid, seq, 0) < 0) {
3807                                 kfree_skb(skb);
3808                                 skb = NULL;
3809                         } else
3810                                 info->skip_notify = 1;
3811                 }
3812
3813                 /* 'rt' points to the first sibling route. If it is not the
3814                  * leaf, then we do not need to send a notification. Otherwise,
3815                  * we need to check if the last sibling has a next route or not
3816                  * and emit a replace or delete notification, respectively.
3817                  */
3818                 info->skip_notify_kernel = 1;
3819                 fn = rcu_dereference_protected(rt->fib6_node,
3820                                             lockdep_is_held(&table->tb6_lock));
3821                 if (rcu_access_pointer(fn->leaf) == rt) {
3822                         struct fib6_info *last_sibling, *replace_rt;
3823
3824                         last_sibling = list_last_entry(&rt->fib6_siblings,
3825                                                        struct fib6_info,
3826                                                        fib6_siblings);
3827                         replace_rt = rcu_dereference_protected(
3828                                             last_sibling->fib6_next,
3829                                             lockdep_is_held(&table->tb6_lock));
3830                         if (replace_rt)
3831                                 call_fib6_entry_notifiers_replace(net,
3832                                                                   replace_rt);
3833                         else
3834                                 call_fib6_multipath_entry_notifiers(net,
3835                                                        FIB_EVENT_ENTRY_DEL,
3836                                                        rt, rt->fib6_nsiblings,
3837                                                        NULL);
3838                 }
3839                 list_for_each_entry_safe(sibling, next_sibling,
3840                                          &rt->fib6_siblings,
3841                                          fib6_siblings) {
3842                         err = fib6_del(sibling, info);
3843                         if (err)
3844                                 goto out_unlock;
3845                 }
3846         }
3847
3848         err = fib6_del(rt, info);
3849 out_unlock:
3850         spin_unlock_bh(&table->tb6_lock);
3851 out_put:
3852         fib6_info_release(rt);
3853
3854         if (skb) {
3855                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3856                             info->nlh, gfp_any());
3857         }
3858         return err;
3859 }
3860
3861 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3862 {
3863         int rc = -ESRCH;
3864
3865         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3866                 goto out;
3867
3868         if (cfg->fc_flags & RTF_GATEWAY &&
3869             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3870                 goto out;
3871
3872         rc = rt6_remove_exception_rt(rt);
3873 out:
3874         return rc;
3875 }
3876
3877 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3878                              struct fib6_nh *nh)
3879 {
3880         struct fib6_result res = {
3881                 .f6i = rt,
3882                 .nh = nh,
3883         };
3884         struct rt6_info *rt_cache;
3885
3886         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3887         if (rt_cache)
3888                 return __ip6_del_cached_rt(rt_cache, cfg);
3889
3890         return 0;
3891 }
3892
3893 struct fib6_nh_del_cached_rt_arg {
3894         struct fib6_config *cfg;
3895         struct fib6_info *f6i;
3896 };
3897
3898 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3899 {
3900         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3901         int rc;
3902
3903         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3904         return rc != -ESRCH ? rc : 0;
3905 }
3906
3907 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3908 {
3909         struct fib6_nh_del_cached_rt_arg arg = {
3910                 .cfg = cfg,
3911                 .f6i = f6i
3912         };
3913
3914         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3915 }
3916
3917 static int ip6_route_del(struct fib6_config *cfg,
3918                          struct netlink_ext_ack *extack)
3919 {
3920         struct fib6_table *table;
3921         struct fib6_info *rt;
3922         struct fib6_node *fn;
3923         int err = -ESRCH;
3924
3925         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3926         if (!table) {
3927                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3928                 return err;
3929         }
3930
3931         rcu_read_lock();
3932
3933         fn = fib6_locate(&table->tb6_root,
3934                          &cfg->fc_dst, cfg->fc_dst_len,
3935                          &cfg->fc_src, cfg->fc_src_len,
3936                          !(cfg->fc_flags & RTF_CACHE));
3937
3938         if (fn) {
3939                 for_each_fib6_node_rt_rcu(fn) {
3940                         struct fib6_nh *nh;
3941
3942                         if (rt->nh && cfg->fc_nh_id &&
3943                             rt->nh->id != cfg->fc_nh_id)
3944                                 continue;
3945
3946                         if (cfg->fc_flags & RTF_CACHE) {
3947                                 int rc = 0;
3948
3949                                 if (rt->nh) {
3950                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3951                                 } else if (cfg->fc_nh_id) {
3952                                         continue;
3953                                 } else {
3954                                         nh = rt->fib6_nh;
3955                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3956                                 }
3957                                 if (rc != -ESRCH) {
3958                                         rcu_read_unlock();
3959                                         return rc;
3960                                 }
3961                                 continue;
3962                         }
3963
3964                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3965                                 continue;
3966                         if (cfg->fc_protocol &&
3967                             cfg->fc_protocol != rt->fib6_protocol)
3968                                 continue;
3969
3970                         if (rt->nh) {
3971                                 if (!fib6_info_hold_safe(rt))
3972                                         continue;
3973                                 rcu_read_unlock();
3974
3975                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3976                         }
3977                         if (cfg->fc_nh_id)
3978                                 continue;
3979
3980                         nh = rt->fib6_nh;
3981                         if (cfg->fc_ifindex &&
3982                             (!nh->fib_nh_dev ||
3983                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3984                                 continue;
3985                         if (cfg->fc_flags & RTF_GATEWAY &&
3986                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3987                                 continue;
3988                         if (!fib6_info_hold_safe(rt))
3989                                 continue;
3990                         rcu_read_unlock();
3991
3992                         /* if gateway was specified only delete the one hop */
3993                         if (cfg->fc_flags & RTF_GATEWAY)
3994                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3995
3996                         return __ip6_del_rt_siblings(rt, cfg);
3997                 }
3998         }
3999         rcu_read_unlock();
4000
4001         return err;
4002 }
4003
4004 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4005 {
4006         struct netevent_redirect netevent;
4007         struct rt6_info *rt, *nrt = NULL;
4008         struct fib6_result res = {};
4009         struct ndisc_options ndopts;
4010         struct inet6_dev *in6_dev;
4011         struct neighbour *neigh;
4012         struct rd_msg *msg;
4013         int optlen, on_link;
4014         u8 *lladdr;
4015
4016         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4017         optlen -= sizeof(*msg);
4018
4019         if (optlen < 0) {
4020                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4021                 return;
4022         }
4023
4024         msg = (struct rd_msg *)icmp6_hdr(skb);
4025
4026         if (ipv6_addr_is_multicast(&msg->dest)) {
4027                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4028                 return;
4029         }
4030
4031         on_link = 0;
4032         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4033                 on_link = 1;
4034         } else if (ipv6_addr_type(&msg->target) !=
4035                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4036                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4037                 return;
4038         }
4039
4040         in6_dev = __in6_dev_get(skb->dev);
4041         if (!in6_dev)
4042                 return;
4043         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4044                 return;
4045
4046         /* RFC2461 8.1:
4047          *      The IP source address of the Redirect MUST be the same as the current
4048          *      first-hop router for the specified ICMP Destination Address.
4049          */
4050
4051         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4052                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4053                 return;
4054         }
4055
4056         lladdr = NULL;
4057         if (ndopts.nd_opts_tgt_lladdr) {
4058                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4059                                              skb->dev);
4060                 if (!lladdr) {
4061                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4062                         return;
4063                 }
4064         }
4065
4066         rt = (struct rt6_info *) dst;
4067         if (rt->rt6i_flags & RTF_REJECT) {
4068                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4069                 return;
4070         }
4071
4072         /* Redirect received -> path was valid.
4073          * Look, redirects are sent only in response to data packets,
4074          * so that this nexthop apparently is reachable. --ANK
4075          */
4076         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4077
4078         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4079         if (!neigh)
4080                 return;
4081
4082         /*
4083          *      We have finally decided to accept it.
4084          */
4085
4086         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4087                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4088                      NEIGH_UPDATE_F_OVERRIDE|
4089                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4090                                      NEIGH_UPDATE_F_ISROUTER)),
4091                      NDISC_REDIRECT, &ndopts);
4092
4093         rcu_read_lock();
4094         res.f6i = rcu_dereference(rt->from);
4095         if (!res.f6i)
4096                 goto out;
4097
4098         if (res.f6i->nh) {
4099                 struct fib6_nh_match_arg arg = {
4100                         .dev = dst->dev,
4101                         .gw = &rt->rt6i_gateway,
4102                 };
4103
4104                 nexthop_for_each_fib6_nh(res.f6i->nh,
4105                                          fib6_nh_find_match, &arg);
4106
4107                 /* fib6_info uses a nexthop that does not have fib6_nh
4108                  * using the dst->dev. Should be impossible
4109                  */
4110                 if (!arg.match)
4111                         goto out;
4112                 res.nh = arg.match;
4113         } else {
4114                 res.nh = res.f6i->fib6_nh;
4115         }
4116
4117         res.fib6_flags = res.f6i->fib6_flags;
4118         res.fib6_type = res.f6i->fib6_type;
4119         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4120         if (!nrt)
4121                 goto out;
4122
4123         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4124         if (on_link)
4125                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4126
4127         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4128
4129         /* rt6_insert_exception() will take care of duplicated exceptions */
4130         if (rt6_insert_exception(nrt, &res)) {
4131                 dst_release_immediate(&nrt->dst);
4132                 goto out;
4133         }
4134
4135         netevent.old = &rt->dst;
4136         netevent.new = &nrt->dst;
4137         netevent.daddr = &msg->dest;
4138         netevent.neigh = neigh;
4139         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4140
4141 out:
4142         rcu_read_unlock();
4143         neigh_release(neigh);
4144 }
4145
4146 #ifdef CONFIG_IPV6_ROUTE_INFO
4147 static struct fib6_info *rt6_get_route_info(struct net *net,
4148                                            const struct in6_addr *prefix, int prefixlen,
4149                                            const struct in6_addr *gwaddr,
4150                                            struct net_device *dev)
4151 {
4152         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4153         int ifindex = dev->ifindex;
4154         struct fib6_node *fn;
4155         struct fib6_info *rt = NULL;
4156         struct fib6_table *table;
4157
4158         table = fib6_get_table(net, tb_id);
4159         if (!table)
4160                 return NULL;
4161
4162         rcu_read_lock();
4163         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4164         if (!fn)
4165                 goto out;
4166
4167         for_each_fib6_node_rt_rcu(fn) {
4168                 /* these routes do not use nexthops */
4169                 if (rt->nh)
4170                         continue;
4171                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4172                         continue;
4173                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4174                     !rt->fib6_nh->fib_nh_gw_family)
4175                         continue;
4176                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4177                         continue;
4178                 if (!fib6_info_hold_safe(rt))
4179                         continue;
4180                 break;
4181         }
4182 out:
4183         rcu_read_unlock();
4184         return rt;
4185 }
4186
4187 static struct fib6_info *rt6_add_route_info(struct net *net,
4188                                            const struct in6_addr *prefix, int prefixlen,
4189                                            const struct in6_addr *gwaddr,
4190                                            struct net_device *dev,
4191                                            unsigned int pref)
4192 {
4193         struct fib6_config cfg = {
4194                 .fc_metric      = IP6_RT_PRIO_USER,
4195                 .fc_ifindex     = dev->ifindex,
4196                 .fc_dst_len     = prefixlen,
4197                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4198                                   RTF_UP | RTF_PREF(pref),
4199                 .fc_protocol = RTPROT_RA,
4200                 .fc_type = RTN_UNICAST,
4201                 .fc_nlinfo.portid = 0,
4202                 .fc_nlinfo.nlh = NULL,
4203                 .fc_nlinfo.nl_net = net,
4204         };
4205
4206         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4207         cfg.fc_dst = *prefix;
4208         cfg.fc_gateway = *gwaddr;
4209
4210         /* We should treat it as a default route if prefix length is 0. */
4211         if (!prefixlen)
4212                 cfg.fc_flags |= RTF_DEFAULT;
4213
4214         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4215
4216         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4217 }
4218 #endif
4219
4220 struct fib6_info *rt6_get_dflt_router(struct net *net,
4221                                      const struct in6_addr *addr,
4222                                      struct net_device *dev)
4223 {
4224         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4225         struct fib6_info *rt;
4226         struct fib6_table *table;
4227
4228         table = fib6_get_table(net, tb_id);
4229         if (!table)
4230                 return NULL;
4231
4232         rcu_read_lock();
4233         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4234                 struct fib6_nh *nh;
4235
4236                 /* RA routes do not use nexthops */
4237                 if (rt->nh)
4238                         continue;
4239
4240                 nh = rt->fib6_nh;
4241                 if (dev == nh->fib_nh_dev &&
4242                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4243                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4244                         break;
4245         }
4246         if (rt && !fib6_info_hold_safe(rt))
4247                 rt = NULL;
4248         rcu_read_unlock();
4249         return rt;
4250 }
4251
4252 struct fib6_info *rt6_add_dflt_router(struct net *net,
4253                                      const struct in6_addr *gwaddr,
4254                                      struct net_device *dev,
4255                                      unsigned int pref)
4256 {
4257         struct fib6_config cfg = {
4258                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4259                 .fc_metric      = IP6_RT_PRIO_USER,
4260                 .fc_ifindex     = dev->ifindex,
4261                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4262                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4263                 .fc_protocol = RTPROT_RA,
4264                 .fc_type = RTN_UNICAST,
4265                 .fc_nlinfo.portid = 0,
4266                 .fc_nlinfo.nlh = NULL,
4267                 .fc_nlinfo.nl_net = net,
4268         };
4269
4270         cfg.fc_gateway = *gwaddr;
4271
4272         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4273                 struct fib6_table *table;
4274
4275                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4276                 if (table)
4277                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4278         }
4279
4280         return rt6_get_dflt_router(net, gwaddr, dev);
4281 }
4282
4283 static void __rt6_purge_dflt_routers(struct net *net,
4284                                      struct fib6_table *table)
4285 {
4286         struct fib6_info *rt;
4287
4288 restart:
4289         rcu_read_lock();
4290         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4291                 struct net_device *dev = fib6_info_nh_dev(rt);
4292                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4293
4294                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4295                     (!idev || idev->cnf.accept_ra != 2) &&
4296                     fib6_info_hold_safe(rt)) {
4297                         rcu_read_unlock();
4298                         ip6_del_rt(net, rt, false);
4299                         goto restart;
4300                 }
4301         }
4302         rcu_read_unlock();
4303
4304         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4305 }
4306
4307 void rt6_purge_dflt_routers(struct net *net)
4308 {
4309         struct fib6_table *table;
4310         struct hlist_head *head;
4311         unsigned int h;
4312
4313         rcu_read_lock();
4314
4315         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4316                 head = &net->ipv6.fib_table_hash[h];
4317                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4318                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4319                                 __rt6_purge_dflt_routers(net, table);
4320                 }
4321         }
4322
4323         rcu_read_unlock();
4324 }
4325
4326 static void rtmsg_to_fib6_config(struct net *net,
4327                                  struct in6_rtmsg *rtmsg,
4328                                  struct fib6_config *cfg)
4329 {
4330         *cfg = (struct fib6_config){
4331                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4332                          : RT6_TABLE_MAIN,
4333                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4334                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4335                 .fc_expires = rtmsg->rtmsg_info,
4336                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4337                 .fc_src_len = rtmsg->rtmsg_src_len,
4338                 .fc_flags = rtmsg->rtmsg_flags,
4339                 .fc_type = rtmsg->rtmsg_type,
4340
4341                 .fc_nlinfo.nl_net = net,
4342
4343                 .fc_dst = rtmsg->rtmsg_dst,
4344                 .fc_src = rtmsg->rtmsg_src,
4345                 .fc_gateway = rtmsg->rtmsg_gateway,
4346         };
4347 }
4348
4349 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4350 {
4351         struct fib6_config cfg;
4352         int err;
4353
4354         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4355                 return -EINVAL;
4356         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4357                 return -EPERM;
4358
4359         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4360
4361         rtnl_lock();
4362         switch (cmd) {
4363         case SIOCADDRT:
4364                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4365                 break;
4366         case SIOCDELRT:
4367                 err = ip6_route_del(&cfg, NULL);
4368                 break;
4369         }
4370         rtnl_unlock();
4371         return err;
4372 }
4373
4374 /*
4375  *      Drop the packet on the floor
4376  */
4377
4378 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4379 {
4380         struct dst_entry *dst = skb_dst(skb);
4381         struct net *net = dev_net(dst->dev);
4382         struct inet6_dev *idev;
4383         int type;
4384
4385         if (netif_is_l3_master(skb->dev) &&
4386             dst->dev == net->loopback_dev)
4387                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4388         else
4389                 idev = ip6_dst_idev(dst);
4390
4391         switch (ipstats_mib_noroutes) {
4392         case IPSTATS_MIB_INNOROUTES:
4393                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4394                 if (type == IPV6_ADDR_ANY) {
4395                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4396                         break;
4397                 }
4398                 fallthrough;
4399         case IPSTATS_MIB_OUTNOROUTES:
4400                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4401                 break;
4402         }
4403
4404         /* Start over by dropping the dst for l3mdev case */
4405         if (netif_is_l3_master(skb->dev))
4406                 skb_dst_drop(skb);
4407
4408         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4409         kfree_skb(skb);
4410         return 0;
4411 }
4412
4413 static int ip6_pkt_discard(struct sk_buff *skb)
4414 {
4415         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4416 }
4417
4418 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4419 {
4420         skb->dev = skb_dst(skb)->dev;
4421         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4422 }
4423
4424 static int ip6_pkt_prohibit(struct sk_buff *skb)
4425 {
4426         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4427 }
4428
4429 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4430 {
4431         skb->dev = skb_dst(skb)->dev;
4432         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4433 }
4434
4435 /*
4436  *      Allocate a dst for local (unicast / anycast) address.
4437  */
4438
4439 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4440                                      struct inet6_dev *idev,
4441                                      const struct in6_addr *addr,
4442                                      bool anycast, gfp_t gfp_flags)
4443 {
4444         struct fib6_config cfg = {
4445                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4446                 .fc_ifindex = idev->dev->ifindex,
4447                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4448                 .fc_dst = *addr,
4449                 .fc_dst_len = 128,
4450                 .fc_protocol = RTPROT_KERNEL,
4451                 .fc_nlinfo.nl_net = net,
4452                 .fc_ignore_dev_down = true,
4453         };
4454         struct fib6_info *f6i;
4455
4456         if (anycast) {
4457                 cfg.fc_type = RTN_ANYCAST;
4458                 cfg.fc_flags |= RTF_ANYCAST;
4459         } else {
4460                 cfg.fc_type = RTN_LOCAL;
4461                 cfg.fc_flags |= RTF_LOCAL;
4462         }
4463
4464         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4465         if (!IS_ERR(f6i))
4466                 f6i->dst_nocount = true;
4467         return f6i;
4468 }
4469
4470 /* remove deleted ip from prefsrc entries */
4471 struct arg_dev_net_ip {
4472         struct net_device *dev;
4473         struct net *net;
4474         struct in6_addr *addr;
4475 };
4476
4477 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4478 {
4479         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4480         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4481         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4482
4483         if (!rt->nh &&
4484             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4485             rt != net->ipv6.fib6_null_entry &&
4486             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4487                 spin_lock_bh(&rt6_exception_lock);
4488                 /* remove prefsrc entry */
4489                 rt->fib6_prefsrc.plen = 0;
4490                 spin_unlock_bh(&rt6_exception_lock);
4491         }
4492         return 0;
4493 }
4494
4495 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4496 {
4497         struct net *net = dev_net(ifp->idev->dev);
4498         struct arg_dev_net_ip adni = {
4499                 .dev = ifp->idev->dev,
4500                 .net = net,
4501                 .addr = &ifp->addr,
4502         };
4503         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4504 }
4505
4506 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4507
4508 /* Remove routers and update dst entries when gateway turn into host. */
4509 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4510 {
4511         struct in6_addr *gateway = (struct in6_addr *)arg;
4512         struct fib6_nh *nh;
4513
4514         /* RA routes do not use nexthops */
4515         if (rt->nh)
4516                 return 0;
4517
4518         nh = rt->fib6_nh;
4519         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4520             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4521                 return -1;
4522
4523         /* Further clean up cached routes in exception table.
4524          * This is needed because cached route may have a different
4525          * gateway than its 'parent' in the case of an ip redirect.
4526          */
4527         fib6_nh_exceptions_clean_tohost(nh, gateway);
4528
4529         return 0;
4530 }
4531
4532 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4533 {
4534         fib6_clean_all(net, fib6_clean_tohost, gateway);
4535 }
4536
4537 struct arg_netdev_event {
4538         const struct net_device *dev;
4539         union {
4540                 unsigned char nh_flags;
4541                 unsigned long event;
4542         };
4543 };
4544
4545 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4546 {
4547         struct fib6_info *iter;
4548         struct fib6_node *fn;
4549
4550         fn = rcu_dereference_protected(rt->fib6_node,
4551                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4552         iter = rcu_dereference_protected(fn->leaf,
4553                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4554         while (iter) {
4555                 if (iter->fib6_metric == rt->fib6_metric &&
4556                     rt6_qualify_for_ecmp(iter))
4557                         return iter;
4558                 iter = rcu_dereference_protected(iter->fib6_next,
4559                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4560         }
4561
4562         return NULL;
4563 }
4564
4565 /* only called for fib entries with builtin fib6_nh */
4566 static bool rt6_is_dead(const struct fib6_info *rt)
4567 {
4568         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4569             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4570              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4571                 return true;
4572
4573         return false;
4574 }
4575
4576 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4577 {
4578         struct fib6_info *iter;
4579         int total = 0;
4580
4581         if (!rt6_is_dead(rt))
4582                 total += rt->fib6_nh->fib_nh_weight;
4583
4584         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4585                 if (!rt6_is_dead(iter))
4586                         total += iter->fib6_nh->fib_nh_weight;
4587         }
4588
4589         return total;
4590 }
4591
4592 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4593 {
4594         int upper_bound = -1;
4595
4596         if (!rt6_is_dead(rt)) {
4597                 *weight += rt->fib6_nh->fib_nh_weight;
4598                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4599                                                     total) - 1;
4600         }
4601         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4602 }
4603
4604 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4605 {
4606         struct fib6_info *iter;
4607         int weight = 0;
4608
4609         rt6_upper_bound_set(rt, &weight, total);
4610
4611         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4612                 rt6_upper_bound_set(iter, &weight, total);
4613 }
4614
4615 void rt6_multipath_rebalance(struct fib6_info *rt)
4616 {
4617         struct fib6_info *first;
4618         int total;
4619
4620         /* In case the entire multipath route was marked for flushing,
4621          * then there is no need to rebalance upon the removal of every
4622          * sibling route.
4623          */
4624         if (!rt->fib6_nsiblings || rt->should_flush)
4625                 return;
4626
4627         /* During lookup routes are evaluated in order, so we need to
4628          * make sure upper bounds are assigned from the first sibling
4629          * onwards.
4630          */
4631         first = rt6_multipath_first_sibling(rt);
4632         if (WARN_ON_ONCE(!first))
4633                 return;
4634
4635         total = rt6_multipath_total_weight(first);
4636         rt6_multipath_upper_bound_set(first, total);
4637 }
4638
4639 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4640 {
4641         const struct arg_netdev_event *arg = p_arg;
4642         struct net *net = dev_net(arg->dev);
4643
4644         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4645             rt->fib6_nh->fib_nh_dev == arg->dev) {
4646                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4647                 fib6_update_sernum_upto_root(net, rt);
4648                 rt6_multipath_rebalance(rt);
4649         }
4650
4651         return 0;
4652 }
4653
4654 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4655 {
4656         struct arg_netdev_event arg = {
4657                 .dev = dev,
4658                 {
4659                         .nh_flags = nh_flags,
4660                 },
4661         };
4662
4663         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4664                 arg.nh_flags |= RTNH_F_LINKDOWN;
4665
4666         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4667 }
4668
4669 /* only called for fib entries with inline fib6_nh */
4670 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4671                                    const struct net_device *dev)
4672 {
4673         struct fib6_info *iter;
4674
4675         if (rt->fib6_nh->fib_nh_dev == dev)
4676                 return true;
4677         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4678                 if (iter->fib6_nh->fib_nh_dev == dev)
4679                         return true;
4680
4681         return false;
4682 }
4683
4684 static void rt6_multipath_flush(struct fib6_info *rt)
4685 {
4686         struct fib6_info *iter;
4687
4688         rt->should_flush = 1;
4689         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4690                 iter->should_flush = 1;
4691 }
4692
4693 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4694                                              const struct net_device *down_dev)
4695 {
4696         struct fib6_info *iter;
4697         unsigned int dead = 0;
4698
4699         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4700             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4701                 dead++;
4702         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4703                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4704                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4705                         dead++;
4706
4707         return dead;
4708 }
4709
4710 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4711                                        const struct net_device *dev,
4712                                        unsigned char nh_flags)
4713 {
4714         struct fib6_info *iter;
4715
4716         if (rt->fib6_nh->fib_nh_dev == dev)
4717                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4718         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4719                 if (iter->fib6_nh->fib_nh_dev == dev)
4720                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4721 }
4722
4723 /* called with write lock held for table with rt */
4724 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4725 {
4726         const struct arg_netdev_event *arg = p_arg;
4727         const struct net_device *dev = arg->dev;
4728         struct net *net = dev_net(dev);
4729
4730         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4731                 return 0;
4732
4733         switch (arg->event) {
4734         case NETDEV_UNREGISTER:
4735                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4736         case NETDEV_DOWN:
4737                 if (rt->should_flush)
4738                         return -1;
4739                 if (!rt->fib6_nsiblings)
4740                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4741                 if (rt6_multipath_uses_dev(rt, dev)) {
4742                         unsigned int count;
4743
4744                         count = rt6_multipath_dead_count(rt, dev);
4745                         if (rt->fib6_nsiblings + 1 == count) {
4746                                 rt6_multipath_flush(rt);
4747                                 return -1;
4748                         }
4749                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4750                                                    RTNH_F_LINKDOWN);
4751                         fib6_update_sernum(net, rt);
4752                         rt6_multipath_rebalance(rt);
4753                 }
4754                 return -2;
4755         case NETDEV_CHANGE:
4756                 if (rt->fib6_nh->fib_nh_dev != dev ||
4757                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4758                         break;
4759                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4760                 rt6_multipath_rebalance(rt);
4761                 break;
4762         }
4763
4764         return 0;
4765 }
4766
4767 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4768 {
4769         struct arg_netdev_event arg = {
4770                 .dev = dev,
4771                 {
4772                         .event = event,
4773                 },
4774         };
4775         struct net *net = dev_net(dev);
4776
4777         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4778                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4779         else
4780                 fib6_clean_all(net, fib6_ifdown, &arg);
4781 }
4782
4783 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4784 {
4785         rt6_sync_down_dev(dev, event);
4786         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4787         neigh_ifdown(&nd_tbl, dev);
4788 }
4789
4790 struct rt6_mtu_change_arg {
4791         struct net_device *dev;
4792         unsigned int mtu;
4793         struct fib6_info *f6i;
4794 };
4795
4796 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4797 {
4798         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4799         struct fib6_info *f6i = arg->f6i;
4800
4801         /* For administrative MTU increase, there is no way to discover
4802          * IPv6 PMTU increase, so PMTU increase should be updated here.
4803          * Since RFC 1981 doesn't include administrative MTU increase
4804          * update PMTU increase is a MUST. (i.e. jumbo frame)
4805          */
4806         if (nh->fib_nh_dev == arg->dev) {
4807                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4808                 u32 mtu = f6i->fib6_pmtu;
4809
4810                 if (mtu >= arg->mtu ||
4811                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4812                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4813
4814                 spin_lock_bh(&rt6_exception_lock);
4815                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4816                 spin_unlock_bh(&rt6_exception_lock);
4817         }
4818
4819         return 0;
4820 }
4821
4822 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4823 {
4824         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4825         struct inet6_dev *idev;
4826
4827         /* In IPv6 pmtu discovery is not optional,
4828            so that RTAX_MTU lock cannot disable it.
4829            We still use this lock to block changes
4830            caused by addrconf/ndisc.
4831         */
4832
4833         idev = __in6_dev_get(arg->dev);
4834         if (!idev)
4835                 return 0;
4836
4837         if (fib6_metric_locked(f6i, RTAX_MTU))
4838                 return 0;
4839
4840         arg->f6i = f6i;
4841         if (f6i->nh) {
4842                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4843                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4844                                                 arg);
4845         }
4846
4847         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4848 }
4849
4850 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4851 {
4852         struct rt6_mtu_change_arg arg = {
4853                 .dev = dev,
4854                 .mtu = mtu,
4855         };
4856
4857         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4858 }
4859
4860 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4861         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4862         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4863         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4864         [RTA_OIF]               = { .type = NLA_U32 },
4865         [RTA_IIF]               = { .type = NLA_U32 },
4866         [RTA_PRIORITY]          = { .type = NLA_U32 },
4867         [RTA_METRICS]           = { .type = NLA_NESTED },
4868         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4869         [RTA_PREF]              = { .type = NLA_U8 },
4870         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4871         [RTA_ENCAP]             = { .type = NLA_NESTED },
4872         [RTA_EXPIRES]           = { .type = NLA_U32 },
4873         [RTA_UID]               = { .type = NLA_U32 },
4874         [RTA_MARK]              = { .type = NLA_U32 },
4875         [RTA_TABLE]             = { .type = NLA_U32 },
4876         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4877         [RTA_SPORT]             = { .type = NLA_U16 },
4878         [RTA_DPORT]             = { .type = NLA_U16 },
4879         [RTA_NH_ID]             = { .type = NLA_U32 },
4880 };
4881
4882 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4883                               struct fib6_config *cfg,
4884                               struct netlink_ext_ack *extack)
4885 {
4886         struct rtmsg *rtm;
4887         struct nlattr *tb[RTA_MAX+1];
4888         unsigned int pref;
4889         int err;
4890
4891         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4892                                      rtm_ipv6_policy, extack);
4893         if (err < 0)
4894                 goto errout;
4895
4896         err = -EINVAL;
4897         rtm = nlmsg_data(nlh);
4898
4899         *cfg = (struct fib6_config){
4900                 .fc_table = rtm->rtm_table,
4901                 .fc_dst_len = rtm->rtm_dst_len,
4902                 .fc_src_len = rtm->rtm_src_len,
4903                 .fc_flags = RTF_UP,
4904                 .fc_protocol = rtm->rtm_protocol,
4905                 .fc_type = rtm->rtm_type,
4906
4907                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4908                 .fc_nlinfo.nlh = nlh,
4909                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4910         };
4911
4912         if (rtm->rtm_type == RTN_UNREACHABLE ||
4913             rtm->rtm_type == RTN_BLACKHOLE ||
4914             rtm->rtm_type == RTN_PROHIBIT ||
4915             rtm->rtm_type == RTN_THROW)
4916                 cfg->fc_flags |= RTF_REJECT;
4917
4918         if (rtm->rtm_type == RTN_LOCAL)
4919                 cfg->fc_flags |= RTF_LOCAL;
4920
4921         if (rtm->rtm_flags & RTM_F_CLONED)
4922                 cfg->fc_flags |= RTF_CACHE;
4923
4924         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4925
4926         if (tb[RTA_NH_ID]) {
4927                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4928                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4929                         NL_SET_ERR_MSG(extack,
4930                                        "Nexthop specification and nexthop id are mutually exclusive");
4931                         goto errout;
4932                 }
4933                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4934         }
4935
4936         if (tb[RTA_GATEWAY]) {
4937                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4938                 cfg->fc_flags |= RTF_GATEWAY;
4939         }
4940         if (tb[RTA_VIA]) {
4941                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4942                 goto errout;
4943         }
4944
4945         if (tb[RTA_DST]) {
4946                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4947
4948                 if (nla_len(tb[RTA_DST]) < plen)
4949                         goto errout;
4950
4951                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4952         }
4953
4954         if (tb[RTA_SRC]) {
4955                 int plen = (rtm->rtm_src_len + 7) >> 3;
4956
4957                 if (nla_len(tb[RTA_SRC]) < plen)
4958                         goto errout;
4959
4960                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4961         }
4962
4963         if (tb[RTA_PREFSRC])
4964                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4965
4966         if (tb[RTA_OIF])
4967                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4968
4969         if (tb[RTA_PRIORITY])
4970                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4971
4972         if (tb[RTA_METRICS]) {
4973                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4974                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4975         }
4976
4977         if (tb[RTA_TABLE])
4978                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4979
4980         if (tb[RTA_MULTIPATH]) {
4981                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4982                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4983
4984                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4985                                                      cfg->fc_mp_len, extack);
4986                 if (err < 0)
4987                         goto errout;
4988         }
4989
4990         if (tb[RTA_PREF]) {
4991                 pref = nla_get_u8(tb[RTA_PREF]);
4992                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4993                     pref != ICMPV6_ROUTER_PREF_HIGH)
4994                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4995                 cfg->fc_flags |= RTF_PREF(pref);
4996         }
4997
4998         if (tb[RTA_ENCAP])
4999                 cfg->fc_encap = tb[RTA_ENCAP];
5000
5001         if (tb[RTA_ENCAP_TYPE]) {
5002                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5003
5004                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5005                 if (err < 0)
5006                         goto errout;
5007         }
5008
5009         if (tb[RTA_EXPIRES]) {
5010                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5011
5012                 if (addrconf_finite_timeout(timeout)) {
5013                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5014                         cfg->fc_flags |= RTF_EXPIRES;
5015                 }
5016         }
5017
5018         err = 0;
5019 errout:
5020         return err;
5021 }
5022
5023 struct rt6_nh {
5024         struct fib6_info *fib6_info;
5025         struct fib6_config r_cfg;
5026         struct list_head next;
5027 };
5028
5029 static int ip6_route_info_append(struct net *net,
5030                                  struct list_head *rt6_nh_list,
5031                                  struct fib6_info *rt,
5032                                  struct fib6_config *r_cfg)
5033 {
5034         struct rt6_nh *nh;
5035         int err = -EEXIST;
5036
5037         list_for_each_entry(nh, rt6_nh_list, next) {
5038                 /* check if fib6_info already exists */
5039                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5040                         return err;
5041         }
5042
5043         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5044         if (!nh)
5045                 return -ENOMEM;
5046         nh->fib6_info = rt;
5047         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5048         list_add_tail(&nh->next, rt6_nh_list);
5049
5050         return 0;
5051 }
5052
5053 static void ip6_route_mpath_notify(struct fib6_info *rt,
5054                                    struct fib6_info *rt_last,
5055                                    struct nl_info *info,
5056                                    __u16 nlflags)
5057 {
5058         /* if this is an APPEND route, then rt points to the first route
5059          * inserted and rt_last points to last route inserted. Userspace
5060          * wants a consistent dump of the route which starts at the first
5061          * nexthop. Since sibling routes are always added at the end of
5062          * the list, find the first sibling of the last route appended
5063          */
5064         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5065                 rt = list_first_entry(&rt_last->fib6_siblings,
5066                                       struct fib6_info,
5067                                       fib6_siblings);
5068         }
5069
5070         if (rt)
5071                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5072 }
5073
5074 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5075 {
5076         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5077         bool should_notify = false;
5078         struct fib6_info *leaf;
5079         struct fib6_node *fn;
5080
5081         rcu_read_lock();
5082         fn = rcu_dereference(rt->fib6_node);
5083         if (!fn)
5084                 goto out;
5085
5086         leaf = rcu_dereference(fn->leaf);
5087         if (!leaf)
5088                 goto out;
5089
5090         if (rt == leaf ||
5091             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5092              rt6_qualify_for_ecmp(leaf)))
5093                 should_notify = true;
5094 out:
5095         rcu_read_unlock();
5096
5097         return should_notify;
5098 }
5099
5100 static int ip6_route_multipath_add(struct fib6_config *cfg,
5101                                    struct netlink_ext_ack *extack)
5102 {
5103         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5104         struct nl_info *info = &cfg->fc_nlinfo;
5105         struct fib6_config r_cfg;
5106         struct rtnexthop *rtnh;
5107         struct fib6_info *rt;
5108         struct rt6_nh *err_nh;
5109         struct rt6_nh *nh, *nh_safe;
5110         __u16 nlflags;
5111         int remaining;
5112         int attrlen;
5113         int err = 1;
5114         int nhn = 0;
5115         int replace = (cfg->fc_nlinfo.nlh &&
5116                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5117         LIST_HEAD(rt6_nh_list);
5118
5119         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5120         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5121                 nlflags |= NLM_F_APPEND;
5122
5123         remaining = cfg->fc_mp_len;
5124         rtnh = (struct rtnexthop *)cfg->fc_mp;
5125
5126         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5127          * fib6_info structs per nexthop
5128          */
5129         while (rtnh_ok(rtnh, remaining)) {
5130                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5131                 if (rtnh->rtnh_ifindex)
5132                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5133
5134                 attrlen = rtnh_attrlen(rtnh);
5135                 if (attrlen > 0) {
5136                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5137
5138                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5139                         if (nla) {
5140                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5141                                 r_cfg.fc_flags |= RTF_GATEWAY;
5142                         }
5143                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5144                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5145                         if (nla)
5146                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5147                 }
5148
5149                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5150                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5151                 if (IS_ERR(rt)) {
5152                         err = PTR_ERR(rt);
5153                         rt = NULL;
5154                         goto cleanup;
5155                 }
5156                 if (!rt6_qualify_for_ecmp(rt)) {
5157                         err = -EINVAL;
5158                         NL_SET_ERR_MSG(extack,
5159                                        "Device only routes can not be added for IPv6 using the multipath API.");
5160                         fib6_info_release(rt);
5161                         goto cleanup;
5162                 }
5163
5164                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5165
5166                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5167                                             rt, &r_cfg);
5168                 if (err) {
5169                         fib6_info_release(rt);
5170                         goto cleanup;
5171                 }
5172
5173                 rtnh = rtnh_next(rtnh, &remaining);
5174         }
5175
5176         if (list_empty(&rt6_nh_list)) {
5177                 NL_SET_ERR_MSG(extack,
5178                                "Invalid nexthop configuration - no valid nexthops");
5179                 return -EINVAL;
5180         }
5181
5182         /* for add and replace send one notification with all nexthops.
5183          * Skip the notification in fib6_add_rt2node and send one with
5184          * the full route when done
5185          */
5186         info->skip_notify = 1;
5187
5188         /* For add and replace, send one notification with all nexthops. For
5189          * append, send one notification with all appended nexthops.
5190          */
5191         info->skip_notify_kernel = 1;
5192
5193         err_nh = NULL;
5194         list_for_each_entry(nh, &rt6_nh_list, next) {
5195                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5196                 fib6_info_release(nh->fib6_info);
5197
5198                 if (!err) {
5199                         /* save reference to last route successfully inserted */
5200                         rt_last = nh->fib6_info;
5201
5202                         /* save reference to first route for notification */
5203                         if (!rt_notif)
5204                                 rt_notif = nh->fib6_info;
5205                 }
5206
5207                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5208                 nh->fib6_info = NULL;
5209                 if (err) {
5210                         if (replace && nhn)
5211                                 NL_SET_ERR_MSG_MOD(extack,
5212                                                    "multipath route replace failed (check consistency of installed routes)");
5213                         err_nh = nh;
5214                         goto add_errout;
5215                 }
5216
5217                 /* Because each route is added like a single route we remove
5218                  * these flags after the first nexthop: if there is a collision,
5219                  * we have already failed to add the first nexthop:
5220                  * fib6_add_rt2node() has rejected it; when replacing, old
5221                  * nexthops have been replaced by first new, the rest should
5222                  * be added to it.
5223                  */
5224                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5225                                                      NLM_F_REPLACE);
5226                 cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5227                 nhn++;
5228         }
5229
5230         /* An in-kernel notification should only be sent in case the new
5231          * multipath route is added as the first route in the node, or if
5232          * it was appended to it. We pass 'rt_notif' since it is the first
5233          * sibling and might allow us to skip some checks in the replace case.
5234          */
5235         if (ip6_route_mpath_should_notify(rt_notif)) {
5236                 enum fib_event_type fib_event;
5237
5238                 if (rt_notif->fib6_nsiblings != nhn - 1)
5239                         fib_event = FIB_EVENT_ENTRY_APPEND;
5240                 else
5241                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5242
5243                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5244                                                           fib_event, rt_notif,
5245                                                           nhn - 1, extack);
5246                 if (err) {
5247                         /* Delete all the siblings that were just added */
5248                         err_nh = NULL;
5249                         goto add_errout;
5250                 }
5251         }
5252
5253         /* success ... tell user about new route */
5254         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5255         goto cleanup;
5256
5257 add_errout:
5258         /* send notification for routes that were added so that
5259          * the delete notifications sent by ip6_route_del are
5260          * coherent
5261          */
5262         if (rt_notif)
5263                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5264
5265         /* Delete routes that were already added */
5266         list_for_each_entry(nh, &rt6_nh_list, next) {
5267                 if (err_nh == nh)
5268                         break;
5269                 ip6_route_del(&nh->r_cfg, extack);
5270         }
5271
5272 cleanup:
5273         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5274                 if (nh->fib6_info)
5275                         fib6_info_release(nh->fib6_info);
5276                 list_del(&nh->next);
5277                 kfree(nh);
5278         }
5279
5280         return err;
5281 }
5282
5283 static int ip6_route_multipath_del(struct fib6_config *cfg,
5284                                    struct netlink_ext_ack *extack)
5285 {
5286         struct fib6_config r_cfg;
5287         struct rtnexthop *rtnh;
5288         int last_err = 0;
5289         int remaining;
5290         int attrlen;
5291         int err;
5292
5293         remaining = cfg->fc_mp_len;
5294         rtnh = (struct rtnexthop *)cfg->fc_mp;
5295
5296         /* Parse a Multipath Entry */
5297         while (rtnh_ok(rtnh, remaining)) {
5298                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5299                 if (rtnh->rtnh_ifindex)
5300                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5301
5302                 attrlen = rtnh_attrlen(rtnh);
5303                 if (attrlen > 0) {
5304                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5305
5306                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5307                         if (nla) {
5308                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5309                                 r_cfg.fc_flags |= RTF_GATEWAY;
5310                         }
5311                 }
5312                 err = ip6_route_del(&r_cfg, extack);
5313                 if (err)
5314                         last_err = err;
5315
5316                 rtnh = rtnh_next(rtnh, &remaining);
5317         }
5318
5319         return last_err;
5320 }
5321
5322 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5323                               struct netlink_ext_ack *extack)
5324 {
5325         struct fib6_config cfg;
5326         int err;
5327
5328         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5329         if (err < 0)
5330                 return err;
5331
5332         if (cfg.fc_nh_id &&
5333             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5334                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5335                 return -EINVAL;
5336         }
5337
5338         if (cfg.fc_mp)
5339                 return ip6_route_multipath_del(&cfg, extack);
5340         else {
5341                 cfg.fc_delete_all_nh = 1;
5342                 return ip6_route_del(&cfg, extack);
5343         }
5344 }
5345
5346 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5347                               struct netlink_ext_ack *extack)
5348 {
5349         struct fib6_config cfg;
5350         int err;
5351
5352         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5353         if (err < 0)
5354                 return err;
5355
5356         if (cfg.fc_metric == 0)
5357                 cfg.fc_metric = IP6_RT_PRIO_USER;
5358
5359         if (cfg.fc_mp)
5360                 return ip6_route_multipath_add(&cfg, extack);
5361         else
5362                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5363 }
5364
5365 /* add the overhead of this fib6_nh to nexthop_len */
5366 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5367 {
5368         int *nexthop_len = arg;
5369
5370         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5371                      + NLA_ALIGN(sizeof(struct rtnexthop))
5372                      + nla_total_size(16); /* RTA_GATEWAY */
5373
5374         if (nh->fib_nh_lws) {
5375                 /* RTA_ENCAP_TYPE */
5376                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5377                 /* RTA_ENCAP */
5378                 *nexthop_len += nla_total_size(2);
5379         }
5380
5381         return 0;
5382 }
5383
5384 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5385 {
5386         int nexthop_len;
5387
5388         if (f6i->nh) {
5389                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5390                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5391                                          &nexthop_len);
5392         } else {
5393                 struct fib6_nh *nh = f6i->fib6_nh;
5394
5395                 nexthop_len = 0;
5396                 if (f6i->fib6_nsiblings) {
5397                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5398                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5399                                     + nla_total_size(16) /* RTA_GATEWAY */
5400                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5401
5402                         nexthop_len *= f6i->fib6_nsiblings;
5403                 }
5404                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5405         }
5406
5407         return NLMSG_ALIGN(sizeof(struct rtmsg))
5408                + nla_total_size(16) /* RTA_SRC */
5409                + nla_total_size(16) /* RTA_DST */
5410                + nla_total_size(16) /* RTA_GATEWAY */
5411                + nla_total_size(16) /* RTA_PREFSRC */
5412                + nla_total_size(4) /* RTA_TABLE */
5413                + nla_total_size(4) /* RTA_IIF */
5414                + nla_total_size(4) /* RTA_OIF */
5415                + nla_total_size(4) /* RTA_PRIORITY */
5416                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5417                + nla_total_size(sizeof(struct rta_cacheinfo))
5418                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5419                + nla_total_size(1) /* RTA_PREF */
5420                + nexthop_len;
5421 }
5422
5423 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5424                                  unsigned char *flags)
5425 {
5426         if (nexthop_is_multipath(nh)) {
5427                 struct nlattr *mp;
5428
5429                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5430                 if (!mp)
5431                         goto nla_put_failure;
5432
5433                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5434                         goto nla_put_failure;
5435
5436                 nla_nest_end(skb, mp);
5437         } else {
5438                 struct fib6_nh *fib6_nh;
5439
5440                 fib6_nh = nexthop_fib6_nh(nh);
5441                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5442                                      flags, false) < 0)
5443                         goto nla_put_failure;
5444         }
5445
5446         return 0;
5447
5448 nla_put_failure:
5449         return -EMSGSIZE;
5450 }
5451
5452 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5453                          struct fib6_info *rt, struct dst_entry *dst,
5454                          struct in6_addr *dest, struct in6_addr *src,
5455                          int iif, int type, u32 portid, u32 seq,
5456                          unsigned int flags)
5457 {
5458         struct rt6_info *rt6 = (struct rt6_info *)dst;
5459         struct rt6key *rt6_dst, *rt6_src;
5460         u32 *pmetrics, table, rt6_flags;
5461         unsigned char nh_flags = 0;
5462         struct nlmsghdr *nlh;
5463         struct rtmsg *rtm;
5464         long expires = 0;
5465
5466         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5467         if (!nlh)
5468                 return -EMSGSIZE;
5469
5470         if (rt6) {
5471                 rt6_dst = &rt6->rt6i_dst;
5472                 rt6_src = &rt6->rt6i_src;
5473                 rt6_flags = rt6->rt6i_flags;
5474         } else {
5475                 rt6_dst = &rt->fib6_dst;
5476                 rt6_src = &rt->fib6_src;
5477                 rt6_flags = rt->fib6_flags;
5478         }
5479
5480         rtm = nlmsg_data(nlh);
5481         rtm->rtm_family = AF_INET6;
5482         rtm->rtm_dst_len = rt6_dst->plen;
5483         rtm->rtm_src_len = rt6_src->plen;
5484         rtm->rtm_tos = 0;
5485         if (rt->fib6_table)
5486                 table = rt->fib6_table->tb6_id;
5487         else
5488                 table = RT6_TABLE_UNSPEC;
5489         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5490         if (nla_put_u32(skb, RTA_TABLE, table))
5491                 goto nla_put_failure;
5492
5493         rtm->rtm_type = rt->fib6_type;
5494         rtm->rtm_flags = 0;
5495         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5496         rtm->rtm_protocol = rt->fib6_protocol;
5497
5498         if (rt6_flags & RTF_CACHE)
5499                 rtm->rtm_flags |= RTM_F_CLONED;
5500
5501         if (dest) {
5502                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5503                         goto nla_put_failure;
5504                 rtm->rtm_dst_len = 128;
5505         } else if (rtm->rtm_dst_len)
5506                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5507                         goto nla_put_failure;
5508 #ifdef CONFIG_IPV6_SUBTREES
5509         if (src) {
5510                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5511                         goto nla_put_failure;
5512                 rtm->rtm_src_len = 128;
5513         } else if (rtm->rtm_src_len &&
5514                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5515                 goto nla_put_failure;
5516 #endif
5517         if (iif) {
5518 #ifdef CONFIG_IPV6_MROUTE
5519                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5520                         int err = ip6mr_get_route(net, skb, rtm, portid);
5521
5522                         if (err == 0)
5523                                 return 0;
5524                         if (err < 0)
5525                                 goto nla_put_failure;
5526                 } else
5527 #endif
5528                         if (nla_put_u32(skb, RTA_IIF, iif))
5529                                 goto nla_put_failure;
5530         } else if (dest) {
5531                 struct in6_addr saddr_buf;
5532                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5533                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5534                         goto nla_put_failure;
5535         }
5536
5537         if (rt->fib6_prefsrc.plen) {
5538                 struct in6_addr saddr_buf;
5539                 saddr_buf = rt->fib6_prefsrc.addr;
5540                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5541                         goto nla_put_failure;
5542         }
5543
5544         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5545         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5546                 goto nla_put_failure;
5547
5548         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5549                 goto nla_put_failure;
5550
5551         /* For multipath routes, walk the siblings list and add
5552          * each as a nexthop within RTA_MULTIPATH.
5553          */
5554         if (rt6) {
5555                 if (rt6_flags & RTF_GATEWAY &&
5556                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5557                         goto nla_put_failure;
5558
5559                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5560                         goto nla_put_failure;
5561         } else if (rt->fib6_nsiblings) {
5562                 struct fib6_info *sibling, *next_sibling;
5563                 struct nlattr *mp;
5564
5565                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5566                 if (!mp)
5567                         goto nla_put_failure;
5568
5569                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5570                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5571                         goto nla_put_failure;
5572
5573                 list_for_each_entry_safe(sibling, next_sibling,
5574                                          &rt->fib6_siblings, fib6_siblings) {
5575                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5576                                             sibling->fib6_nh->fib_nh_weight,
5577                                             AF_INET6) < 0)
5578                                 goto nla_put_failure;
5579                 }
5580
5581                 nla_nest_end(skb, mp);
5582         } else if (rt->nh) {
5583                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5584                         goto nla_put_failure;
5585
5586                 if (nexthop_is_blackhole(rt->nh))
5587                         rtm->rtm_type = RTN_BLACKHOLE;
5588
5589                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5590                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5591                         goto nla_put_failure;
5592
5593                 rtm->rtm_flags |= nh_flags;
5594         } else {
5595                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5596                                      &nh_flags, false) < 0)
5597                         goto nla_put_failure;
5598
5599                 rtm->rtm_flags |= nh_flags;
5600         }
5601
5602         if (rt6_flags & RTF_EXPIRES) {
5603                 expires = dst ? dst->expires : rt->expires;
5604                 expires -= jiffies;
5605         }
5606
5607         if (!dst) {
5608                 if (rt->offload)
5609                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5610                 if (rt->trap)
5611                         rtm->rtm_flags |= RTM_F_TRAP;
5612         }
5613
5614         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5615                 goto nla_put_failure;
5616
5617         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5618                 goto nla_put_failure;
5619
5620
5621         nlmsg_end(skb, nlh);
5622         return 0;
5623
5624 nla_put_failure:
5625         nlmsg_cancel(skb, nlh);
5626         return -EMSGSIZE;
5627 }
5628
5629 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5630 {
5631         const struct net_device *dev = arg;
5632
5633         if (nh->fib_nh_dev == dev)
5634                 return 1;
5635
5636         return 0;
5637 }
5638
5639 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5640                                const struct net_device *dev)
5641 {
5642         if (f6i->nh) {
5643                 struct net_device *_dev = (struct net_device *)dev;
5644
5645                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5646                                                   fib6_info_nh_uses_dev,
5647                                                   _dev);
5648         }
5649
5650         if (f6i->fib6_nh->fib_nh_dev == dev)
5651                 return true;
5652
5653         if (f6i->fib6_nsiblings) {
5654                 struct fib6_info *sibling, *next_sibling;
5655
5656                 list_for_each_entry_safe(sibling, next_sibling,
5657                                          &f6i->fib6_siblings, fib6_siblings) {
5658                         if (sibling->fib6_nh->fib_nh_dev == dev)
5659                                 return true;
5660                 }
5661         }
5662
5663         return false;
5664 }
5665
5666 struct fib6_nh_exception_dump_walker {
5667         struct rt6_rtnl_dump_arg *dump;
5668         struct fib6_info *rt;
5669         unsigned int flags;
5670         unsigned int skip;
5671         unsigned int count;
5672 };
5673
5674 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5675 {
5676         struct fib6_nh_exception_dump_walker *w = arg;
5677         struct rt6_rtnl_dump_arg *dump = w->dump;
5678         struct rt6_exception_bucket *bucket;
5679         struct rt6_exception *rt6_ex;
5680         int i, err;
5681
5682         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5683         if (!bucket)
5684                 return 0;
5685
5686         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5687                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5688                         if (w->skip) {
5689                                 w->skip--;
5690                                 continue;
5691                         }
5692
5693                         /* Expiration of entries doesn't bump sernum, insertion
5694                          * does. Removal is triggered by insertion, so we can
5695                          * rely on the fact that if entries change between two
5696                          * partial dumps, this node is scanned again completely,
5697                          * see rt6_insert_exception() and fib6_dump_table().
5698                          *
5699                          * Count expired entries we go through as handled
5700                          * entries that we'll skip next time, in case of partial
5701                          * node dump. Otherwise, if entries expire meanwhile,
5702                          * we'll skip the wrong amount.
5703                          */
5704                         if (rt6_check_expired(rt6_ex->rt6i)) {
5705                                 w->count++;
5706                                 continue;
5707                         }
5708
5709                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5710                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5711                                             RTM_NEWROUTE,
5712                                             NETLINK_CB(dump->cb->skb).portid,
5713                                             dump->cb->nlh->nlmsg_seq, w->flags);
5714                         if (err)
5715                                 return err;
5716
5717                         w->count++;
5718                 }
5719                 bucket++;
5720         }
5721
5722         return 0;
5723 }
5724
5725 /* Return -1 if done with node, number of handled routes on partial dump */
5726 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5727 {
5728         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5729         struct fib_dump_filter *filter = &arg->filter;
5730         unsigned int flags = NLM_F_MULTI;
5731         struct net *net = arg->net;
5732         int count = 0;
5733
5734         if (rt == net->ipv6.fib6_null_entry)
5735                 return -1;
5736
5737         if ((filter->flags & RTM_F_PREFIX) &&
5738             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5739                 /* success since this is not a prefix route */
5740                 return -1;
5741         }
5742         if (filter->filter_set &&
5743             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5744              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5745              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5746                 return -1;
5747         }
5748
5749         if (filter->filter_set ||
5750             !filter->dump_routes || !filter->dump_exceptions) {
5751                 flags |= NLM_F_DUMP_FILTERED;
5752         }
5753
5754         if (filter->dump_routes) {
5755                 if (skip) {
5756                         skip--;
5757                 } else {
5758                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5759                                           0, RTM_NEWROUTE,
5760                                           NETLINK_CB(arg->cb->skb).portid,
5761                                           arg->cb->nlh->nlmsg_seq, flags)) {
5762                                 return 0;
5763                         }
5764                         count++;
5765                 }
5766         }
5767
5768         if (filter->dump_exceptions) {
5769                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5770                                                            .rt = rt,
5771                                                            .flags = flags,
5772                                                            .skip = skip,
5773                                                            .count = 0 };
5774                 int err;
5775
5776                 rcu_read_lock();
5777                 if (rt->nh) {
5778                         err = nexthop_for_each_fib6_nh(rt->nh,
5779                                                        rt6_nh_dump_exceptions,
5780                                                        &w);
5781                 } else {
5782                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5783                 }
5784                 rcu_read_unlock();
5785
5786                 if (err)
5787                         return count += w.count;
5788         }
5789
5790         return -1;
5791 }
5792
5793 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5794                                         const struct nlmsghdr *nlh,
5795                                         struct nlattr **tb,
5796                                         struct netlink_ext_ack *extack)
5797 {
5798         struct rtmsg *rtm;
5799         int i, err;
5800
5801         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5802                 NL_SET_ERR_MSG_MOD(extack,
5803                                    "Invalid header for get route request");
5804                 return -EINVAL;
5805         }
5806
5807         if (!netlink_strict_get_check(skb))
5808                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5809                                               rtm_ipv6_policy, extack);
5810
5811         rtm = nlmsg_data(nlh);
5812         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5813             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5814             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5815             rtm->rtm_type) {
5816                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5817                 return -EINVAL;
5818         }
5819         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5820                 NL_SET_ERR_MSG_MOD(extack,
5821                                    "Invalid flags for get route request");
5822                 return -EINVAL;
5823         }
5824
5825         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5826                                             rtm_ipv6_policy, extack);
5827         if (err)
5828                 return err;
5829
5830         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5831             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5832                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5833                 return -EINVAL;
5834         }
5835
5836         for (i = 0; i <= RTA_MAX; i++) {
5837                 if (!tb[i])
5838                         continue;
5839
5840                 switch (i) {
5841                 case RTA_SRC:
5842                 case RTA_DST:
5843                 case RTA_IIF:
5844                 case RTA_OIF:
5845                 case RTA_MARK:
5846                 case RTA_UID:
5847                 case RTA_SPORT:
5848                 case RTA_DPORT:
5849                 case RTA_IP_PROTO:
5850                         break;
5851                 default:
5852                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5853                         return -EINVAL;
5854                 }
5855         }
5856
5857         return 0;
5858 }
5859
5860 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5861                               struct netlink_ext_ack *extack)
5862 {
5863         struct net *net = sock_net(in_skb->sk);
5864         struct nlattr *tb[RTA_MAX+1];
5865         int err, iif = 0, oif = 0;
5866         struct fib6_info *from;
5867         struct dst_entry *dst;
5868         struct rt6_info *rt;
5869         struct sk_buff *skb;
5870         struct rtmsg *rtm;
5871         struct flowi6 fl6 = {};
5872         bool fibmatch;
5873
5874         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5875         if (err < 0)
5876                 goto errout;
5877
5878         err = -EINVAL;
5879         rtm = nlmsg_data(nlh);
5880         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5881         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5882
5883         if (tb[RTA_SRC]) {
5884                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5885                         goto errout;
5886
5887                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5888         }
5889
5890         if (tb[RTA_DST]) {
5891                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5892                         goto errout;
5893
5894                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5895         }
5896
5897         if (tb[RTA_IIF])
5898                 iif = nla_get_u32(tb[RTA_IIF]);
5899
5900         if (tb[RTA_OIF])
5901                 oif = nla_get_u32(tb[RTA_OIF]);
5902
5903         if (tb[RTA_MARK])
5904                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5905
5906         if (tb[RTA_UID])
5907                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5908                                            nla_get_u32(tb[RTA_UID]));
5909         else
5910                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5911
5912         if (tb[RTA_SPORT])
5913                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5914
5915         if (tb[RTA_DPORT])
5916                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5917
5918         if (tb[RTA_IP_PROTO]) {
5919                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5920                                                   &fl6.flowi6_proto, AF_INET6,
5921                                                   extack);
5922                 if (err)
5923                         goto errout;
5924         }
5925
5926         if (iif) {
5927                 struct net_device *dev;
5928                 int flags = 0;
5929
5930                 rcu_read_lock();
5931
5932                 dev = dev_get_by_index_rcu(net, iif);
5933                 if (!dev) {
5934                         rcu_read_unlock();
5935                         err = -ENODEV;
5936                         goto errout;
5937                 }
5938
5939                 fl6.flowi6_iif = iif;
5940
5941                 if (!ipv6_addr_any(&fl6.saddr))
5942                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5943
5944                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5945
5946                 rcu_read_unlock();
5947         } else {
5948                 fl6.flowi6_oif = oif;
5949
5950                 dst = ip6_route_output(net, NULL, &fl6);
5951         }
5952
5953
5954         rt = container_of(dst, struct rt6_info, dst);
5955         if (rt->dst.error) {
5956                 err = rt->dst.error;
5957                 ip6_rt_put(rt);
5958                 goto errout;
5959         }
5960
5961         if (rt == net->ipv6.ip6_null_entry) {
5962                 err = rt->dst.error;
5963                 ip6_rt_put(rt);
5964                 goto errout;
5965         }
5966
5967         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5968         if (!skb) {
5969                 ip6_rt_put(rt);
5970                 err = -ENOBUFS;
5971                 goto errout;
5972         }
5973
5974         skb_dst_set(skb, &rt->dst);
5975
5976         rcu_read_lock();
5977         from = rcu_dereference(rt->from);
5978         if (from) {
5979                 if (fibmatch)
5980                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5981                                             iif, RTM_NEWROUTE,
5982                                             NETLINK_CB(in_skb).portid,
5983                                             nlh->nlmsg_seq, 0);
5984                 else
5985                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5986                                             &fl6.saddr, iif, RTM_NEWROUTE,
5987                                             NETLINK_CB(in_skb).portid,
5988                                             nlh->nlmsg_seq, 0);
5989         } else {
5990                 err = -ENETUNREACH;
5991         }
5992         rcu_read_unlock();
5993
5994         if (err < 0) {
5995                 kfree_skb(skb);
5996                 goto errout;
5997         }
5998
5999         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6000 errout:
6001         return err;
6002 }
6003
6004 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6005                      unsigned int nlm_flags)
6006 {
6007         struct sk_buff *skb;
6008         struct net *net = info->nl_net;
6009         u32 seq;
6010         int err;
6011
6012         err = -ENOBUFS;
6013         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6014
6015         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6016         if (!skb)
6017                 goto errout;
6018
6019         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6020                             event, info->portid, seq, nlm_flags);
6021         if (err < 0) {
6022                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6023                 WARN_ON(err == -EMSGSIZE);
6024                 kfree_skb(skb);
6025                 goto errout;
6026         }
6027         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6028                     info->nlh, gfp_any());
6029         return;
6030 errout:
6031         if (err < 0)
6032                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6033 }
6034
6035 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6036                     struct nl_info *info)
6037 {
6038         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6039         struct sk_buff *skb;
6040         int err = -ENOBUFS;
6041
6042         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
6043          * is implemented and supported for nexthop objects
6044          */
6045         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
6046
6047         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6048         if (!skb)
6049                 goto errout;
6050
6051         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6052                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6053         if (err < 0) {
6054                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6055                 WARN_ON(err == -EMSGSIZE);
6056                 kfree_skb(skb);
6057                 goto errout;
6058         }
6059         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6060                     info->nlh, gfp_any());
6061         return;
6062 errout:
6063         if (err < 0)
6064                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6065 }
6066
6067 static int ip6_route_dev_notify(struct notifier_block *this,
6068                                 unsigned long event, void *ptr)
6069 {
6070         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6071         struct net *net = dev_net(dev);
6072
6073         if (!(dev->flags & IFF_LOOPBACK))
6074                 return NOTIFY_OK;
6075
6076         if (event == NETDEV_REGISTER) {
6077                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6078                 net->ipv6.ip6_null_entry->dst.dev = dev;
6079                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6080 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6081                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6082                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6083                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6084                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6085 #endif
6086          } else if (event == NETDEV_UNREGISTER &&
6087                     dev->reg_state != NETREG_UNREGISTERED) {
6088                 /* NETDEV_UNREGISTER could be fired for multiple times by
6089                  * netdev_wait_allrefs(). Make sure we only call this once.
6090                  */
6091                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6092 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6093                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6094                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6095 #endif
6096         }
6097
6098         return NOTIFY_OK;
6099 }
6100
6101 /*
6102  *      /proc
6103  */
6104
6105 #ifdef CONFIG_PROC_FS
6106 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6107 {
6108         struct net *net = (struct net *)seq->private;
6109         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6110                    net->ipv6.rt6_stats->fib_nodes,
6111                    net->ipv6.rt6_stats->fib_route_nodes,
6112                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6113                    net->ipv6.rt6_stats->fib_rt_entries,
6114                    net->ipv6.rt6_stats->fib_rt_cache,
6115                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6116                    net->ipv6.rt6_stats->fib_discarded_routes);
6117
6118         return 0;
6119 }
6120 #endif  /* CONFIG_PROC_FS */
6121
6122 #ifdef CONFIG_SYSCTL
6123
6124 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6125                               void *buffer, size_t *lenp, loff_t *ppos)
6126 {
6127         struct net *net;
6128         int delay;
6129         int ret;
6130         if (!write)
6131                 return -EINVAL;
6132
6133         net = (struct net *)ctl->extra1;
6134         delay = net->ipv6.sysctl.flush_delay;
6135         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6136         if (ret)
6137                 return ret;
6138
6139         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6140         return 0;
6141 }
6142
6143 static struct ctl_table ipv6_route_table_template[] = {
6144         {
6145                 .procname       =       "flush",
6146                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6147                 .maxlen         =       sizeof(int),
6148                 .mode           =       0200,
6149                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6150         },
6151         {
6152                 .procname       =       "gc_thresh",
6153                 .data           =       &ip6_dst_ops_template.gc_thresh,
6154                 .maxlen         =       sizeof(int),
6155                 .mode           =       0644,
6156                 .proc_handler   =       proc_dointvec,
6157         },
6158         {
6159                 .procname       =       "max_size",
6160                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6161                 .maxlen         =       sizeof(int),
6162                 .mode           =       0644,
6163                 .proc_handler   =       proc_dointvec,
6164         },
6165         {
6166                 .procname       =       "gc_min_interval",
6167                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6168                 .maxlen         =       sizeof(int),
6169                 .mode           =       0644,
6170                 .proc_handler   =       proc_dointvec_jiffies,
6171         },
6172         {
6173                 .procname       =       "gc_timeout",
6174                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6175                 .maxlen         =       sizeof(int),
6176                 .mode           =       0644,
6177                 .proc_handler   =       proc_dointvec_jiffies,
6178         },
6179         {
6180                 .procname       =       "gc_interval",
6181                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6182                 .maxlen         =       sizeof(int),
6183                 .mode           =       0644,
6184                 .proc_handler   =       proc_dointvec_jiffies,
6185         },
6186         {
6187                 .procname       =       "gc_elasticity",
6188                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6189                 .maxlen         =       sizeof(int),
6190                 .mode           =       0644,
6191                 .proc_handler   =       proc_dointvec,
6192         },
6193         {
6194                 .procname       =       "mtu_expires",
6195                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6196                 .maxlen         =       sizeof(int),
6197                 .mode           =       0644,
6198                 .proc_handler   =       proc_dointvec_jiffies,
6199         },
6200         {
6201                 .procname       =       "min_adv_mss",
6202                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6203                 .maxlen         =       sizeof(int),
6204                 .mode           =       0644,
6205                 .proc_handler   =       proc_dointvec,
6206         },
6207         {
6208                 .procname       =       "gc_min_interval_ms",
6209                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6210                 .maxlen         =       sizeof(int),
6211                 .mode           =       0644,
6212                 .proc_handler   =       proc_dointvec_ms_jiffies,
6213         },
6214         {
6215                 .procname       =       "skip_notify_on_dev_down",
6216                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6217                 .maxlen         =       sizeof(int),
6218                 .mode           =       0644,
6219                 .proc_handler   =       proc_dointvec_minmax,
6220                 .extra1         =       SYSCTL_ZERO,
6221                 .extra2         =       SYSCTL_ONE,
6222         },
6223         { }
6224 };
6225
6226 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6227 {
6228         struct ctl_table *table;
6229
6230         table = kmemdup(ipv6_route_table_template,
6231                         sizeof(ipv6_route_table_template),
6232                         GFP_KERNEL);
6233
6234         if (table) {
6235                 table[0].data = &net->ipv6.sysctl.flush_delay;
6236                 table[0].extra1 = net;
6237                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6238                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6239                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6240                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6241                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6242                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6243                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6244                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6245                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6246                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6247
6248                 /* Don't export sysctls to unprivileged users */
6249                 if (net->user_ns != &init_user_ns)
6250                         table[0].procname = NULL;
6251         }
6252
6253         return table;
6254 }
6255 #endif
6256
6257 static int __net_init ip6_route_net_init(struct net *net)
6258 {
6259         int ret = -ENOMEM;
6260
6261         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6262                sizeof(net->ipv6.ip6_dst_ops));
6263
6264         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6265                 goto out_ip6_dst_ops;
6266
6267         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6268         if (!net->ipv6.fib6_null_entry)
6269                 goto out_ip6_dst_entries;
6270         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6271                sizeof(*net->ipv6.fib6_null_entry));
6272
6273         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6274                                            sizeof(*net->ipv6.ip6_null_entry),
6275                                            GFP_KERNEL);
6276         if (!net->ipv6.ip6_null_entry)
6277                 goto out_fib6_null_entry;
6278         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6279         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6280                          ip6_template_metrics, true);
6281         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6282
6283 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6284         net->ipv6.fib6_has_custom_rules = false;
6285         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6286                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6287                                                GFP_KERNEL);
6288         if (!net->ipv6.ip6_prohibit_entry)
6289                 goto out_ip6_null_entry;
6290         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6291         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6292                          ip6_template_metrics, true);
6293         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6294
6295         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6296                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6297                                                GFP_KERNEL);
6298         if (!net->ipv6.ip6_blk_hole_entry)
6299                 goto out_ip6_prohibit_entry;
6300         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6301         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6302                          ip6_template_metrics, true);
6303         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6304 #ifdef CONFIG_IPV6_SUBTREES
6305         net->ipv6.fib6_routes_require_src = 0;
6306 #endif
6307 #endif
6308
6309         net->ipv6.sysctl.flush_delay = 0;
6310         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6311         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6312         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6313         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6314         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6315         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6316         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6317         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6318
6319         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6320
6321         ret = 0;
6322 out:
6323         return ret;
6324
6325 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6326 out_ip6_prohibit_entry:
6327         kfree(net->ipv6.ip6_prohibit_entry);
6328 out_ip6_null_entry:
6329         kfree(net->ipv6.ip6_null_entry);
6330 #endif
6331 out_fib6_null_entry:
6332         kfree(net->ipv6.fib6_null_entry);
6333 out_ip6_dst_entries:
6334         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6335 out_ip6_dst_ops:
6336         goto out;
6337 }
6338
6339 static void __net_exit ip6_route_net_exit(struct net *net)
6340 {
6341         kfree(net->ipv6.fib6_null_entry);
6342         kfree(net->ipv6.ip6_null_entry);
6343 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6344         kfree(net->ipv6.ip6_prohibit_entry);
6345         kfree(net->ipv6.ip6_blk_hole_entry);
6346 #endif
6347         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6348 }
6349
6350 static int __net_init ip6_route_net_init_late(struct net *net)
6351 {
6352 #ifdef CONFIG_PROC_FS
6353         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6354                         sizeof(struct ipv6_route_iter));
6355         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6356                         rt6_stats_seq_show, NULL);
6357 #endif
6358         return 0;
6359 }
6360
6361 static void __net_exit ip6_route_net_exit_late(struct net *net)
6362 {
6363 #ifdef CONFIG_PROC_FS
6364         remove_proc_entry("ipv6_route", net->proc_net);
6365         remove_proc_entry("rt6_stats", net->proc_net);
6366 #endif
6367 }
6368
6369 static struct pernet_operations ip6_route_net_ops = {
6370         .init = ip6_route_net_init,
6371         .exit = ip6_route_net_exit,
6372 };
6373
6374 static int __net_init ipv6_inetpeer_init(struct net *net)
6375 {
6376         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6377
6378         if (!bp)
6379                 return -ENOMEM;
6380         inet_peer_base_init(bp);
6381         net->ipv6.peers = bp;
6382         return 0;
6383 }
6384
6385 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6386 {
6387         struct inet_peer_base *bp = net->ipv6.peers;
6388
6389         net->ipv6.peers = NULL;
6390         inetpeer_invalidate_tree(bp);
6391         kfree(bp);
6392 }
6393
6394 static struct pernet_operations ipv6_inetpeer_ops = {
6395         .init   =       ipv6_inetpeer_init,
6396         .exit   =       ipv6_inetpeer_exit,
6397 };
6398
6399 static struct pernet_operations ip6_route_net_late_ops = {
6400         .init = ip6_route_net_init_late,
6401         .exit = ip6_route_net_exit_late,
6402 };
6403
6404 static struct notifier_block ip6_route_dev_notifier = {
6405         .notifier_call = ip6_route_dev_notify,
6406         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6407 };
6408
6409 void __init ip6_route_init_special_entries(void)
6410 {
6411         /* Registering of the loopback is done before this portion of code,
6412          * the loopback reference in rt6_info will not be taken, do it
6413          * manually for init_net */
6414         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6415         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6416         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6417   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6418         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6419         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6420         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6421         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6422   #endif
6423 }
6424
6425 #if IS_BUILTIN(CONFIG_IPV6)
6426 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6427 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6428
6429 BTF_ID_LIST(btf_fib6_info_id)
6430 BTF_ID(struct, fib6_info)
6431
6432 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6433         .seq_ops                = &ipv6_route_seq_ops,
6434         .init_seq_private       = bpf_iter_init_seq_net,
6435         .fini_seq_private       = bpf_iter_fini_seq_net,
6436         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6437 };
6438
6439 static struct bpf_iter_reg ipv6_route_reg_info = {
6440         .target                 = "ipv6_route",
6441         .ctx_arg_info_size      = 1,
6442         .ctx_arg_info           = {
6443                 { offsetof(struct bpf_iter__ipv6_route, rt),
6444                   PTR_TO_BTF_ID_OR_NULL },
6445         },
6446         .seq_info               = &ipv6_route_seq_info,
6447 };
6448
6449 static int __init bpf_iter_register(void)
6450 {
6451         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6452         return bpf_iter_reg_target(&ipv6_route_reg_info);
6453 }
6454
6455 static void bpf_iter_unregister(void)
6456 {
6457         bpf_iter_unreg_target(&ipv6_route_reg_info);
6458 }
6459 #endif
6460 #endif
6461
6462 int __init ip6_route_init(void)
6463 {
6464         int ret;
6465         int cpu;
6466
6467         ret = -ENOMEM;
6468         ip6_dst_ops_template.kmem_cachep =
6469                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6470                                   SLAB_HWCACHE_ALIGN, NULL);
6471         if (!ip6_dst_ops_template.kmem_cachep)
6472                 goto out;
6473
6474         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6475         if (ret)
6476                 goto out_kmem_cache;
6477
6478         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6479         if (ret)
6480                 goto out_dst_entries;
6481
6482         ret = register_pernet_subsys(&ip6_route_net_ops);
6483         if (ret)
6484                 goto out_register_inetpeer;
6485
6486         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6487
6488         ret = fib6_init();
6489         if (ret)
6490                 goto out_register_subsys;
6491
6492         ret = xfrm6_init();
6493         if (ret)
6494                 goto out_fib6_init;
6495
6496         ret = fib6_rules_init();
6497         if (ret)
6498                 goto xfrm6_init;
6499
6500         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6501         if (ret)
6502                 goto fib6_rules_init;
6503
6504         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6505                                    inet6_rtm_newroute, NULL, 0);
6506         if (ret < 0)
6507                 goto out_register_late_subsys;
6508
6509         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6510                                    inet6_rtm_delroute, NULL, 0);
6511         if (ret < 0)
6512                 goto out_register_late_subsys;
6513
6514         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6515                                    inet6_rtm_getroute, NULL,
6516                                    RTNL_FLAG_DOIT_UNLOCKED);
6517         if (ret < 0)
6518                 goto out_register_late_subsys;
6519
6520         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6521         if (ret)
6522                 goto out_register_late_subsys;
6523
6524 #if IS_BUILTIN(CONFIG_IPV6)
6525 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6526         ret = bpf_iter_register();
6527         if (ret)
6528                 goto out_register_late_subsys;
6529 #endif
6530 #endif
6531
6532         for_each_possible_cpu(cpu) {
6533                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6534
6535                 INIT_LIST_HEAD(&ul->head);
6536                 spin_lock_init(&ul->lock);
6537         }
6538
6539 out:
6540         return ret;
6541
6542 out_register_late_subsys:
6543         rtnl_unregister_all(PF_INET6);
6544         unregister_pernet_subsys(&ip6_route_net_late_ops);
6545 fib6_rules_init:
6546         fib6_rules_cleanup();
6547 xfrm6_init:
6548         xfrm6_fini();
6549 out_fib6_init:
6550         fib6_gc_cleanup();
6551 out_register_subsys:
6552         unregister_pernet_subsys(&ip6_route_net_ops);
6553 out_register_inetpeer:
6554         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6555 out_dst_entries:
6556         dst_entries_destroy(&ip6_dst_blackhole_ops);
6557 out_kmem_cache:
6558         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6559         goto out;
6560 }
6561
6562 void ip6_route_cleanup(void)
6563 {
6564 #if IS_BUILTIN(CONFIG_IPV6)
6565 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6566         bpf_iter_unregister();
6567 #endif
6568 #endif
6569         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6570         unregister_pernet_subsys(&ip6_route_net_late_ops);
6571         fib6_rules_cleanup();
6572         xfrm6_fini();
6573         fib6_gc_cleanup();
6574         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6575         unregister_pernet_subsys(&ip6_route_net_ops);
6576         dst_entries_destroy(&ip6_dst_blackhole_ops);
6577         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6578 }