net/ipv4/fib_semantics.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              IPv4 Forwarding Information Base: semantics.
   7  *
   8  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   9  *
  10  *              This program is free software; you can redistribute it and/or
  11  *              modify it under the terms of the GNU General Public License
  12  *              as published by the Free Software Foundation; either version
  13  *              2 of the License, or (at your option) any later version.
  14  */
  15
  16 #include <linux/uaccess.h>
  17 #include <linux/bitops.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/jiffies.h>
  21 #include <linux/mm.h>
  22 #include <linux/string.h>
  23 #include <linux/socket.h>
  24 #include <linux/sockios.h>
  25 #include <linux/errno.h>
  26 #include <linux/in.h>
  27 #include <linux/inet.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/init.h>
  34 #include <linux/slab.h>
  35 #include <linux/netlink.h>
  36
  37 #include <net/arp.h>
  38 #include <net/ip.h>
  39 #include <net/protocol.h>
  40 #include <net/route.h>
  41 #include <net/tcp.h>
  42 #include <net/sock.h>
  43 #include <net/ip_fib.h>
  44 #include <net/ip6_fib.h>
  45 #include <net/nexthop.h>
  46 #include <net/netlink.h>
  47 #include <net/rtnh.h>
  48 #include <net/lwtunnel.h>
  49 #include <net/fib_notifier.h>
  50 #include <net/addrconf.h>
  51
  52 #include "fib_lookup.h"
  53
  54 static DEFINE_SPINLOCK(fib_info_lock);
  55 static struct hlist_head *fib_info_hash;
  56 static struct hlist_head *fib_info_laddrhash;
  57 static unsigned int fib_info_hash_size;
  58 static unsigned int fib_info_cnt;
  59
  60 #define DEVINDEX_HASHBITS 8
  61 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  62 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  63
  64 /* for_nexthops and change_nexthops only used when nexthop object
  65  * is not set in a fib_info. The logic within can reference fib_nh.
  66  */
  67 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  68
  69 #define for_nexthops(fi) {                                              \
  70         int nhsel; const struct fib_nh *nh;                             \
  71         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  72              nhsel < fib_info_num_path((fi));                           \
  73              nh++, nhsel++)
  74
  75 #define change_nexthops(fi) {                                           \
  76         int nhsel; struct fib_nh *nexthop_nh;                           \
  77         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  78              nhsel < fib_info_num_path((fi));                           \
  79              nexthop_nh++, nhsel++)
  80
  81 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  82
  83 /* Hope, that gcc will optimize it to get rid of dummy loop */
  84
  85 #define for_nexthops(fi) {                                              \
  86         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  87         for (nhsel = 0; nhsel < 1; nhsel++)
  88
  89 #define change_nexthops(fi) {                                           \
  90         int nhsel;                                                      \
  91         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  92         for (nhsel = 0; nhsel < 1; nhsel++)
  93
  94 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  95
  96 #define endfor_nexthops(fi) }
  97
  98
  99 const struct fib_prop fib_props[RTN_MAX + 1] = {
 100         [RTN_UNSPEC] = {
 101                 .error  = 0,
 102                 .scope  = RT_SCOPE_NOWHERE,
 103         },
 104         [RTN_UNICAST] = {
 105                 .error  = 0,
 106                 .scope  = RT_SCOPE_UNIVERSE,
 107         },
 108         [RTN_LOCAL] = {
 109                 .error  = 0,
 110                 .scope  = RT_SCOPE_HOST,
 111         },
 112         [RTN_BROADCAST] = {
 113                 .error  = 0,
 114                 .scope  = RT_SCOPE_LINK,
 115         },
 116         [RTN_ANYCAST] = {
 117                 .error  = 0,
 118                 .scope  = RT_SCOPE_LINK,
 119         },
 120         [RTN_MULTICAST] = {
 121                 .error  = 0,
 122                 .scope  = RT_SCOPE_UNIVERSE,
 123         },
 124         [RTN_BLACKHOLE] = {
 125                 .error  = -EINVAL,
 126                 .scope  = RT_SCOPE_UNIVERSE,
 127         },
 128         [RTN_UNREACHABLE] = {
 129                 .error  = -EHOSTUNREACH,
 130                 .scope  = RT_SCOPE_UNIVERSE,
 131         },
 132         [RTN_PROHIBIT] = {
 133                 .error  = -EACCES,
 134                 .scope  = RT_SCOPE_UNIVERSE,
 135         },
 136         [RTN_THROW] = {
 137                 .error  = -EAGAIN,
 138                 .scope  = RT_SCOPE_UNIVERSE,
 139         },
 140         [RTN_NAT] = {
 141                 .error  = -EINVAL,
 142                 .scope  = RT_SCOPE_NOWHERE,
 143         },
 144         [RTN_XRESOLVE] = {
 145                 .error  = -EINVAL,
 146                 .scope  = RT_SCOPE_NOWHERE,
 147         },
 148 };
 149
 150 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 151 {
 152         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 153
 154         if (!rt)
 155                 return;
 156
 157         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 158          * because we waited an RCU grace period before calling
 159          * free_fib_info_rcu()
 160          */
 161
 162         dst_dev_put(&rt->dst);
 163         dst_release_immediate(&rt->dst);
 164 }
 165
 166 static void free_nh_exceptions(struct fib_nh_common *nhc)
 167 {
 168         struct fnhe_hash_bucket *hash;
 169         int i;
 170
 171         hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
 172         if (!hash)
 173                 return;
 174         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 175                 struct fib_nh_exception *fnhe;
 176
 177                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 178                 while (fnhe) {
 179                         struct fib_nh_exception *next;
 180
 181                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 182
 183                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 184                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 185
 186                         kfree(fnhe);
 187
 188                         fnhe = next;
 189                 }
 190         }
 191         kfree(hash);
 192 }
 193
 194 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 195 {
 196         int cpu;
 197
 198         if (!rtp)
 199                 return;
 200
 201         for_each_possible_cpu(cpu) {
 202                 struct rtable *rt;
 203
 204                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 205                 if (rt) {
 206                         dst_dev_put(&rt->dst);
 207                         dst_release_immediate(&rt->dst);
 208                 }
 209         }
 210         free_percpu(rtp);
 211 }
 212
 213 void fib_nh_common_release(struct fib_nh_common *nhc)
 214 {
 215         if (nhc->nhc_dev)
 216                 dev_put(nhc->nhc_dev);
 217
 218         lwtstate_put(nhc->nhc_lwtstate);
 219         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 220         rt_fibinfo_free(&nhc->nhc_rth_input);
 221         free_nh_exceptions(nhc);
 222 }
 223 EXPORT_SYMBOL_GPL(fib_nh_common_release);
 224
 225 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
 226 {
 227 #ifdef CONFIG_IP_ROUTE_CLASSID
 228         if (fib_nh->nh_tclassid)
 229                 net->ipv4.fib_num_tclassid_users--;
 230 #endif
 231         fib_nh_common_release(&fib_nh->nh_common);
 232 }
 233
 234 /* Release a nexthop info record */
 235 static void free_fib_info_rcu(struct rcu_head *head)
 236 {
 237         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 238
 239         change_nexthops(fi) {
 240                 fib_nh_release(fi->fib_net, nexthop_nh);
 241         } endfor_nexthops(fi);
 242
 243         ip_fib_metrics_put(fi->fib_metrics);
 244
 245         kfree(fi);
 246 }
 247
 248 void free_fib_info(struct fib_info *fi)
 249 {
 250         if (fi->fib_dead == 0) {
 251                 pr_warn("Freeing alive fib_info %p\n", fi);
 252                 return;
 253         }
 254         fib_info_cnt--;
 255
 256         call_rcu(&fi->rcu, free_fib_info_rcu);
 257 }
 258 EXPORT_SYMBOL_GPL(free_fib_info);
 259
 260 void fib_release_info(struct fib_info *fi)
 261 {
 262         spin_lock_bh(&fib_info_lock);
 263         if (fi && --fi->fib_treeref == 0) {
 264                 hlist_del(&fi->fib_hash);
 265                 if (fi->fib_prefsrc)
 266                         hlist_del(&fi->fib_lhash);
 267                 change_nexthops(fi) {
 268                         if (!nexthop_nh->fib_nh_dev)
 269                                 continue;
 270                         hlist_del(&nexthop_nh->nh_hash);
 271                 } endfor_nexthops(fi)
 272                 fi->fib_dead = 1;
 273                 fib_info_put(fi);
 274         }
 275         spin_unlock_bh(&fib_info_lock);
 276 }
 277
 278 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
 279 {
 280         const struct fib_nh *onh;
 281
 282         for_nexthops(fi) {
 283                 onh = fib_info_nh(ofi, nhsel);
 284
 285                 if (nh->fib_nh_oif != onh->fib_nh_oif ||
 286                     nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
 287                     nh->fib_nh_scope != onh->fib_nh_scope ||
 288 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 289                     nh->fib_nh_weight != onh->fib_nh_weight ||
 290 #endif
 291 #ifdef CONFIG_IP_ROUTE_CLASSID
 292                     nh->nh_tclassid != onh->nh_tclassid ||
 293 #endif
 294                     lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
 295                     ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
 296                         return -1;
 297
 298                 if (nh->fib_nh_gw_family == AF_INET &&
 299                     nh->fib_nh_gw4 != onh->fib_nh_gw4)
 300                         return -1;
 301
 302                 if (nh->fib_nh_gw_family == AF_INET6 &&
 303                     ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
 304                         return -1;
 305         } endfor_nexthops(fi);
 306         return 0;
 307 }
 308
 309 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 310 {
 311         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 312
 313         return (val ^
 314                 (val >> DEVINDEX_HASHBITS) ^
 315                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 316 }
 317
 318 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 319 {
 320         unsigned int mask = (fib_info_hash_size - 1);
 321         unsigned int val = fi->fib_nhs;
 322
 323         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 324         val ^= (__force u32)fi->fib_prefsrc;
 325         val ^= fi->fib_priority;
 326         for_nexthops(fi) {
 327                 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
 328         } endfor_nexthops(fi)
 329
 330         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 331 }
 332
 333 static struct fib_info *fib_find_info(struct fib_info *nfi)
 334 {
 335         struct hlist_head *head;
 336         struct fib_info *fi;
 337         unsigned int hash;
 338
 339         hash = fib_info_hashfn(nfi);
 340         head = &fib_info_hash[hash];
 341
 342         hlist_for_each_entry(fi, head, fib_hash) {
 343                 if (!net_eq(fi->fib_net, nfi->fib_net))
 344                         continue;
 345                 if (fi->fib_nhs != nfi->fib_nhs)
 346                         continue;
 347                 if (nfi->fib_protocol == fi->fib_protocol &&
 348                     nfi->fib_scope == fi->fib_scope &&
 349                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 350                     nfi->fib_priority == fi->fib_priority &&
 351                     nfi->fib_type == fi->fib_type &&
 352                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 353                            sizeof(u32) * RTAX_MAX) == 0 &&
 354                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 355                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 356                         return fi;
 357         }
 358
 359         return NULL;
 360 }
 361
 362 /* Check, that the gateway is already configured.
 363  * Used only by redirect accept routine.
 364  */
 365 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 366 {
 367         struct hlist_head *head;
 368         struct fib_nh *nh;
 369         unsigned int hash;
 370
 371         spin_lock(&fib_info_lock);
 372
 373         hash = fib_devindex_hashfn(dev->ifindex);
 374         head = &fib_info_devhash[hash];
 375         hlist_for_each_entry(nh, head, nh_hash) {
 376                 if (nh->fib_nh_dev == dev &&
 377                     nh->fib_nh_gw4 == gw &&
 378                     !(nh->fib_nh_flags & RTNH_F_DEAD)) {
 379                         spin_unlock(&fib_info_lock);
 380                         return 0;
 381                 }
 382         }
 383
 384         spin_unlock(&fib_info_lock);
 385
 386         return -1;
 387 }
 388
 389 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 390 {
 391         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 392                          + nla_total_size(4) /* RTA_TABLE */
 393                          + nla_total_size(4) /* RTA_DST */
 394                          + nla_total_size(4) /* RTA_PRIORITY */
 395                          + nla_total_size(4) /* RTA_PREFSRC */
 396                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 397         unsigned int nhs = fib_info_num_path(fi);
 398
 399         /* space for nested metrics */
 400         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 401
 402         if (nhs) {
 403                 size_t nh_encapsize = 0;
 404                 /* Also handles the special case nhs == 1 */
 405
 406                 /* each nexthop is packed in an attribute */
 407                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 408                 unsigned int i;
 409
 410                 /* may contain flow and gateway attribute */
 411                 nhsize += 2 * nla_total_size(4);
 412
 413                 /* grab encap info */
 414                 for (i = 0; i < fib_info_num_path(fi); i++) {
 415                         struct fib_nh_common *nhc = fib_info_nhc(fi, i);
 416
 417                         if (nhc->nhc_lwtstate) {
 418                                 /* RTA_ENCAP_TYPE */
 419                                 nh_encapsize += lwtunnel_get_encap_size(
 420                                                 nhc->nhc_lwtstate);
 421                                 /* RTA_ENCAP */
 422                                 nh_encapsize +=  nla_total_size(2);
 423                         }
 424                 }
 425
 426                 /* all nexthops are packed in a nested attribute */
 427                 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
 428
 429         }
 430
 431         return payload;
 432 }
 433
 434 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 435                int dst_len, u32 tb_id, const struct nl_info *info,
 436                unsigned int nlm_flags)
 437 {
 438         struct sk_buff *skb;
 439         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 440         int err = -ENOBUFS;
 441
 442         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 443         if (!skb)
 444                 goto errout;
 445
 446         err = fib_dump_info(skb, info->portid, seq, event, tb_id,
 447                             fa->fa_type, key, dst_len,
 448                             fa->fa_tos, fa->fa_info, nlm_flags);
 449         if (err < 0) {
 450                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 451                 WARN_ON(err == -EMSGSIZE);
 452                 kfree_skb(skb);
 453                 goto errout;
 454         }
 455         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 456                     info->nlh, GFP_KERNEL);
 457         return;
 458 errout:
 459         if (err < 0)
 460                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 461 }
 462
 463 static int fib_detect_death(struct fib_info *fi, int order,
 464                             struct fib_info **last_resort, int *last_idx,
 465                             int dflt)
 466 {
 467         const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
 468         struct neighbour *n;
 469         int state = NUD_NONE;
 470
 471         if (likely(nhc->nhc_gw_family == AF_INET))
 472                 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
 473         else if (nhc->nhc_gw_family == AF_INET6)
 474                 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
 475                                  nhc->nhc_dev);
 476         else
 477                 n = NULL;
 478
 479         if (n) {
 480                 state = n->nud_state;
 481                 neigh_release(n);
 482         } else {
 483                 return 0;
 484         }
 485         if (state == NUD_REACHABLE)
 486                 return 0;
 487         if ((state & NUD_VALID) && order != dflt)
 488                 return 0;
 489         if ((state & NUD_VALID) ||
 490             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 491                 *last_resort = fi;
 492                 *last_idx = order;
 493         }
 494         return 1;
 495 }
 496
 497 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
 498                        u16 encap_type, void *cfg, gfp_t gfp_flags,
 499                        struct netlink_ext_ack *extack)
 500 {
 501         int err;
 502
 503         nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
 504                                                     gfp_flags);
 505         if (!nhc->nhc_pcpu_rth_output)
 506                 return -ENOMEM;
 507
 508         if (encap) {
 509                 struct lwtunnel_state *lwtstate;
 510
 511                 if (encap_type == LWTUNNEL_ENCAP_NONE) {
 512                         NL_SET_ERR_MSG(extack, "LWT encap type not specified");
 513                         err = -EINVAL;
 514                         goto lwt_failure;
 515                 }
 516                 err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family,
 517                                            cfg, &lwtstate, extack);
 518                 if (err)
 519                         goto lwt_failure;
 520
 521                 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
 522         }
 523
 524         return 0;
 525
 526 lwt_failure:
 527         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 528         nhc->nhc_pcpu_rth_output = NULL;
 529         return err;
 530 }
 531 EXPORT_SYMBOL_GPL(fib_nh_common_init);
 532
 533 int fib_nh_init(struct net *net, struct fib_nh *nh,
 534                 struct fib_config *cfg, int nh_weight,
 535                 struct netlink_ext_ack *extack)
 536 {
 537         int err;
 538
 539         nh->fib_nh_family = AF_INET;
 540
 541         err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
 542                                  cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
 543         if (err)
 544                 return err;
 545
 546         nh->fib_nh_oif = cfg->fc_oif;
 547         nh->fib_nh_gw_family = cfg->fc_gw_family;
 548         if (cfg->fc_gw_family == AF_INET)
 549                 nh->fib_nh_gw4 = cfg->fc_gw4;
 550         else if (cfg->fc_gw_family == AF_INET6)
 551                 nh->fib_nh_gw6 = cfg->fc_gw6;
 552
 553         nh->fib_nh_flags = cfg->fc_flags;
 554
 555 #ifdef CONFIG_IP_ROUTE_CLASSID
 556         nh->nh_tclassid = cfg->fc_flow;
 557         if (nh->nh_tclassid)
 558                 net->ipv4.fib_num_tclassid_users++;
 559 #endif
 560 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 561         nh->fib_nh_weight = nh_weight;
 562 #endif
 563         return 0;
 564 }
 565
 566 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 567
 568 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
 569                               struct netlink_ext_ack *extack)
 570 {
 571         int nhs = 0;
 572
 573         while (rtnh_ok(rtnh, remaining)) {
 574                 nhs++;
 575                 rtnh = rtnh_next(rtnh, &remaining);
 576         }
 577
 578         /* leftover implies invalid nexthop configuration, discard it */
 579         if (remaining > 0) {
 580                 NL_SET_ERR_MSG(extack,
 581                                "Invalid nexthop configuration - extra data after nexthops");
 582                 nhs = 0;
 583         }
 584
 585         return nhs;
 586 }
 587
 588 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 589                        int remaining, struct fib_config *cfg,
 590                        struct netlink_ext_ack *extack)
 591 {
 592         struct net *net = fi->fib_net;
 593         struct fib_config fib_cfg;
 594         struct fib_nh *nh;
 595         int ret;
 596
 597         change_nexthops(fi) {
 598                 int attrlen;
 599
 600                 memset(&fib_cfg, 0, sizeof(fib_cfg));
 601
 602                 if (!rtnh_ok(rtnh, remaining)) {
 603                         NL_SET_ERR_MSG(extack,
 604                                        "Invalid nexthop configuration - extra data after nexthop");
 605                         return -EINVAL;
 606                 }
 607
 608                 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
 609                         NL_SET_ERR_MSG(extack,
 610                                        "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
 611                         return -EINVAL;
 612                 }
 613
 614                 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 615                 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
 616
 617                 attrlen = rtnh_attrlen(rtnh);
 618                 if (attrlen > 0) {
 619                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 620
 621                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 622                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 623                         if (nla && nlav) {
 624                                 NL_SET_ERR_MSG(extack,
 625                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 626                                 return -EINVAL;
 627                         }
 628                         if (nla) {
 629                                 fib_cfg.fc_gw4 = nla_get_in_addr(nla);
 630                                 if (fib_cfg.fc_gw4)
 631                                         fib_cfg.fc_gw_family = AF_INET;
 632                         } else if (nlav) {
 633                                 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
 634                                 if (ret)
 635                                         goto errout;
 636                         }
 637
 638                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 639                         if (nla)
 640                                 fib_cfg.fc_flow = nla_get_u32(nla);
 641
 642                         fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
 643                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
 644                         if (nla)
 645                                 fib_cfg.fc_encap_type = nla_get_u16(nla);
 646                 }
 647
 648                 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
 649                                   rtnh->rtnh_hops + 1, extack);
 650                 if (ret)
 651                         goto errout;
 652
 653                 rtnh = rtnh_next(rtnh, &remaining);
 654         } endfor_nexthops(fi);
 655
 656         ret = -EINVAL;
 657         nh = fib_info_nh(fi, 0);
 658         if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
 659                 NL_SET_ERR_MSG(extack,
 660                                "Nexthop device index does not match RTA_OIF");
 661                 goto errout;
 662         }
 663         if (cfg->fc_gw_family) {
 664                 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
 665                     (cfg->fc_gw_family == AF_INET &&
 666                      nh->fib_nh_gw4 != cfg->fc_gw4) ||
 667                     (cfg->fc_gw_family == AF_INET6 &&
 668                      ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
 669                         NL_SET_ERR_MSG(extack,
 670                                        "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
 671                         goto errout;
 672                 }
 673         }
 674 #ifdef CONFIG_IP_ROUTE_CLASSID
 675         if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
 676                 NL_SET_ERR_MSG(extack,
 677                                "Nexthop class id does not match RTA_FLOW");
 678                 goto errout;
 679         }
 680 #endif
 681         ret = 0;
 682 errout:
 683         return ret;
 684 }
 685
 686 static void fib_rebalance(struct fib_info *fi)
 687 {
 688         int total;
 689         int w;
 690
 691         if (fib_info_num_path(fi) < 2)
 692                 return;
 693
 694         total = 0;
 695         for_nexthops(fi) {
 696                 if (nh->fib_nh_flags & RTNH_F_DEAD)
 697                         continue;
 698
 699                 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
 700                     nh->fib_nh_flags & RTNH_F_LINKDOWN)
 701                         continue;
 702
 703                 total += nh->fib_nh_weight;
 704         } endfor_nexthops(fi);
 705
 706         w = 0;
 707         change_nexthops(fi) {
 708                 int upper_bound;
 709
 710                 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
 711                         upper_bound = -1;
 712                 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
 713                            nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
 714                         upper_bound = -1;
 715                 } else {
 716                         w += nexthop_nh->fib_nh_weight;
 717                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 718                                                             total) - 1;
 719                 }
 720
 721                 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
 722         } endfor_nexthops(fi);
 723 }
 724 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 725
 726 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 727                        int remaining, struct fib_config *cfg,
 728                        struct netlink_ext_ack *extack)
 729 {
 730         NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
 731
 732         return -EINVAL;
 733 }
 734
 735 #define fib_rebalance(fi) do { } while (0)
 736
 737 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 738
 739 static int fib_encap_match(u16 encap_type,
 740                            struct nlattr *encap,
 741                            const struct fib_nh *nh,
 742                            const struct fib_config *cfg,
 743                            struct netlink_ext_ack *extack)
 744 {
 745         struct lwtunnel_state *lwtstate;
 746         int ret, result = 0;
 747
 748         if (encap_type == LWTUNNEL_ENCAP_NONE)
 749                 return 0;
 750
 751         ret = lwtunnel_build_state(encap_type, encap, AF_INET,
 752                                    cfg, &lwtstate, extack);
 753         if (!ret) {
 754                 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
 755                 lwtstate_free(lwtstate);
 756         }
 757
 758         return result;
 759 }
 760
 761 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
 762                  struct netlink_ext_ack *extack)
 763 {
 764 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 765         struct rtnexthop *rtnh;
 766         int remaining;
 767 #endif
 768
 769         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 770                 return 1;
 771
 772         if (cfg->fc_oif || cfg->fc_gw_family) {
 773                 struct fib_nh *nh = fib_info_nh(fi, 0);
 774
 775                 if (cfg->fc_encap) {
 776                         if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
 777                                             nh, cfg, extack))
 778                                 return 1;
 779                 }
 780 #ifdef CONFIG_IP_ROUTE_CLASSID
 781                 if (cfg->fc_flow &&
 782                     cfg->fc_flow != nh->nh_tclassid)
 783                         return 1;
 784 #endif
 785                 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
 786                     (cfg->fc_gw_family &&
 787                      cfg->fc_gw_family != nh->fib_nh_gw_family))
 788                         return 1;
 789
 790                 if (cfg->fc_gw_family == AF_INET &&
 791                     cfg->fc_gw4 != nh->fib_nh_gw4)
 792                         return 1;
 793
 794                 if (cfg->fc_gw_family == AF_INET6 &&
 795                     ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
 796                         return 1;
 797
 798                 return 0;
 799         }
 800
 801 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 802         if (!cfg->fc_mp)
 803                 return 0;
 804
 805         rtnh = cfg->fc_mp;
 806         remaining = cfg->fc_mp_len;
 807
 808         for_nexthops(fi) {
 809                 int attrlen;
 810
 811                 if (!rtnh_ok(rtnh, remaining))
 812                         return -EINVAL;
 813
 814                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
 815                         return 1;
 816
 817                 attrlen = rtnh_attrlen(rtnh);
 818                 if (attrlen > 0) {
 819                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 820
 821                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 822                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 823                         if (nla && nlav) {
 824                                 NL_SET_ERR_MSG(extack,
 825                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 826                                 return -EINVAL;
 827                         }
 828
 829                         if (nla) {
 830                                 if (nh->fib_nh_gw_family != AF_INET ||
 831                                     nla_get_in_addr(nla) != nh->fib_nh_gw4)
 832                                         return 1;
 833                         } else if (nlav) {
 834                                 struct fib_config cfg2;
 835                                 int err;
 836
 837                                 err = fib_gw_from_via(&cfg2, nlav, extack);
 838                                 if (err)
 839                                         return err;
 840
 841                                 switch (nh->fib_nh_gw_family) {
 842                                 case AF_INET:
 843                                         if (cfg2.fc_gw_family != AF_INET ||
 844                                             cfg2.fc_gw4 != nh->fib_nh_gw4)
 845                                                 return 1;
 846                                         break;
 847                                 case AF_INET6:
 848                                         if (cfg2.fc_gw_family != AF_INET6 ||
 849                                             ipv6_addr_cmp(&cfg2.fc_gw6,
 850                                                           &nh->fib_nh_gw6))
 851                                                 return 1;
 852                                         break;
 853                                 }
 854                         }
 855
 856 #ifdef CONFIG_IP_ROUTE_CLASSID
 857                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 858                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 859                                 return 1;
 860 #endif
 861                 }
 862
 863                 rtnh = rtnh_next(rtnh, &remaining);
 864         } endfor_nexthops(fi);
 865 #endif
 866         return 0;
 867 }
 868
 869 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
 870 {
 871         struct nlattr *nla;
 872         int remaining;
 873
 874         if (!cfg->fc_mx)
 875                 return true;
 876
 877         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 878                 int type = nla_type(nla);
 879                 u32 fi_val, val;
 880
 881                 if (!type)
 882                         continue;
 883                 if (type > RTAX_MAX)
 884                         return false;
 885
 886                 if (type == RTAX_CC_ALGO) {
 887                         char tmp[TCP_CA_NAME_MAX];
 888                         bool ecn_ca = false;
 889
 890                         nla_strlcpy(tmp, nla, sizeof(tmp));
 891                         val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
 892                 } else {
 893                         if (nla_len(nla) != sizeof(u32))
 894                                 return false;
 895                         val = nla_get_u32(nla);
 896                 }
 897
 898                 fi_val = fi->fib_metrics->metrics[type - 1];
 899                 if (type == RTAX_FEATURES)
 900                         fi_val &= ~DST_FEATURE_ECN_CA;
 901
 902                 if (fi_val != val)
 903                         return false;
 904         }
 905
 906         return true;
 907 }
 908
 909 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
 910                               u32 table, struct netlink_ext_ack *extack)
 911 {
 912         struct fib6_config cfg = {
 913                 .fc_table = table,
 914                 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
 915                 .fc_ifindex = nh->fib_nh_oif,
 916                 .fc_gateway = nh->fib_nh_gw6,
 917         };
 918         struct fib6_nh fib6_nh = {};
 919         int err;
 920
 921         err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
 922         if (!err) {
 923                 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
 924                 dev_hold(nh->fib_nh_dev);
 925                 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
 926                 nh->fib_nh_scope = RT_SCOPE_LINK;
 927
 928                 ipv6_stub->fib6_nh_release(&fib6_nh);
 929         }
 930
 931         return err;
 932 }
 933
 934 /*
 935  * Picture
 936  * -------
 937  *
 938  * Semantics of nexthop is very messy by historical reasons.
 939  * We have to take into account, that:
 940  * a) gateway can be actually local interface address,
 941  *    so that gatewayed route is direct.
 942  * b) gateway must be on-link address, possibly
 943  *    described not by an ifaddr, but also by a direct route.
 944  * c) If both gateway and interface are specified, they should not
 945  *    contradict.
 946  * d) If we use tunnel routes, gateway could be not on-link.
 947  *
 948  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 949  * results in pretty ugly and hairy code with obscure logic.
 950  *
 951  * I chose to generalized it instead, so that the size
 952  * of code does not increase practically, but it becomes
 953  * much more general.
 954  * Every prefix is assigned a "scope" value: "host" is local address,
 955  * "link" is direct route,
 956  * [ ... "site" ... "interior" ... ]
 957  * and "universe" is true gateway route with global meaning.
 958  *
 959  * Every prefix refers to a set of "nexthop"s (gw, oif),
 960  * where gw must have narrower scope. This recursion stops
 961  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 962  * which means that gw is forced to be on link.
 963  *
 964  * Code is still hairy, but now it is apparently logically
 965  * consistent and very flexible. F.e. as by-product it allows
 966  * to co-exists in peace independent exterior and interior
 967  * routing processes.
 968  *
 969  * Normally it looks as following.
 970  *
 971  * {universe prefix}  -> (gw, oif) [scope link]
 972  *                |
 973  *                |-> {link prefix} -> (gw, oif) [scope local]
 974  *                                      |
 975  *                                      |-> {local prefix} (terminal node)
 976  */
 977 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
 978                               u8 scope, struct netlink_ext_ack *extack)
 979 {
 980         struct net_device *dev;
 981         struct fib_result res;
 982         int err;
 983
 984         if (nh->fib_nh_flags & RTNH_F_ONLINK) {
 985                 unsigned int addr_type;
 986
 987                 if (scope >= RT_SCOPE_LINK) {
 988                         NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
 989                         return -EINVAL;
 990                 }
 991                 dev = __dev_get_by_index(net, nh->fib_nh_oif);
 992                 if (!dev) {
 993                         NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
 994                         return -ENODEV;
 995                 }
 996                 if (!(dev->flags & IFF_UP)) {
 997                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
 998                         return -ENETDOWN;
 999                 }
1000                 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1001                 if (addr_type != RTN_UNICAST) {
1002                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1003                         return -EINVAL;
1004                 }
1005                 if (!netif_carrier_ok(dev))
1006                         nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1007                 nh->fib_nh_dev = dev;
1008                 dev_hold(dev);
1009                 nh->fib_nh_scope = RT_SCOPE_LINK;
1010                 return 0;
1011         }
1012         rcu_read_lock();
1013         {
1014                 struct fib_table *tbl = NULL;
1015                 struct flowi4 fl4 = {
1016                         .daddr = nh->fib_nh_gw4,
1017                         .flowi4_scope = scope + 1,
1018                         .flowi4_oif = nh->fib_nh_oif,
1019                         .flowi4_iif = LOOPBACK_IFINDEX,
1020                 };
1021
1022                 /* It is not necessary, but requires a bit of thinking */
1023                 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1024                         fl4.flowi4_scope = RT_SCOPE_LINK;
1025
1026                 if (table)
1027                         tbl = fib_get_table(net, table);
1028
1029                 if (tbl)
1030                         err = fib_table_lookup(tbl, &fl4, &res,
1031                                                FIB_LOOKUP_IGNORE_LINKSTATE |
1032                                                FIB_LOOKUP_NOREF);
1033
1034                 /* on error or if no table given do full lookup. This
1035                  * is needed for example when nexthops are in the local
1036                  * table rather than the given table
1037                  */
1038                 if (!tbl || err) {
1039                         err = fib_lookup(net, &fl4, &res,
1040                                          FIB_LOOKUP_IGNORE_LINKSTATE);
1041                 }
1042
1043                 if (err) {
1044                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1045                         goto out;
1046                 }
1047         }
1048
1049         err = -EINVAL;
1050         if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1051                 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1052                 goto out;
1053         }
1054         nh->fib_nh_scope = res.scope;
1055         nh->fib_nh_oif = FIB_RES_OIF(res);
1056         nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1057         if (!dev) {
1058                 NL_SET_ERR_MSG(extack,
1059                                "No egress device for nexthop gateway");
1060                 goto out;
1061         }
1062         dev_hold(dev);
1063         if (!netif_carrier_ok(dev))
1064                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1065         err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1066 out:
1067         rcu_read_unlock();
1068         return err;
1069 }
1070
1071 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1072                               struct netlink_ext_ack *extack)
1073 {
1074         struct in_device *in_dev;
1075         int err;
1076
1077         if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1078                 NL_SET_ERR_MSG(extack,
1079                                "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1080                 return -EINVAL;
1081         }
1082
1083         rcu_read_lock();
1084
1085         err = -ENODEV;
1086         in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1087         if (!in_dev)
1088                 goto out;
1089         err = -ENETDOWN;
1090         if (!(in_dev->dev->flags & IFF_UP)) {
1091                 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1092                 goto out;
1093         }
1094
1095         nh->fib_nh_dev = in_dev->dev;
1096         dev_hold(nh->fib_nh_dev);
1097         nh->fib_nh_scope = RT_SCOPE_HOST;
1098         if (!netif_carrier_ok(nh->fib_nh_dev))
1099                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1100         err = 0;
1101 out:
1102         rcu_read_unlock();
1103         return err;
1104 }
1105
1106 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1107                  struct netlink_ext_ack *extack)
1108 {
1109         int err;
1110
1111         if (nh->fib_nh_gw_family == AF_INET)
1112                 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1113         else if (nh->fib_nh_gw_family == AF_INET6)
1114                 err = fib_check_nh_v6_gw(net, nh, table, extack);
1115         else
1116                 err = fib_check_nh_nongw(net, nh, extack);
1117
1118         return err;
1119 }
1120
1121 static inline unsigned int fib_laddr_hashfn(__be32 val)
1122 {
1123         unsigned int mask = (fib_info_hash_size - 1);
1124
1125         return ((__force u32)val ^
1126                 ((__force u32)val >> 7) ^
1127                 ((__force u32)val >> 14)) & mask;
1128 }
1129
1130 static struct hlist_head *fib_info_hash_alloc(int bytes)
1131 {
1132         if (bytes <= PAGE_SIZE)
1133                 return kzalloc(bytes, GFP_KERNEL);
1134         else
1135                 return (struct hlist_head *)
1136                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1137                                          get_order(bytes));
1138 }
1139
1140 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1141 {
1142         if (!hash)
1143                 return;
1144
1145         if (bytes <= PAGE_SIZE)
1146                 kfree(hash);
1147         else
1148                 free_pages((unsigned long) hash, get_order(bytes));
1149 }
1150
1151 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1152                                struct hlist_head *new_laddrhash,
1153                                unsigned int new_size)
1154 {
1155         struct hlist_head *old_info_hash, *old_laddrhash;
1156         unsigned int old_size = fib_info_hash_size;
1157         unsigned int i, bytes;
1158
1159         spin_lock_bh(&fib_info_lock);
1160         old_info_hash = fib_info_hash;
1161         old_laddrhash = fib_info_laddrhash;
1162         fib_info_hash_size = new_size;
1163
1164         for (i = 0; i < old_size; i++) {
1165                 struct hlist_head *head = &fib_info_hash[i];
1166                 struct hlist_node *n;
1167                 struct fib_info *fi;
1168
1169                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1170                         struct hlist_head *dest;
1171                         unsigned int new_hash;
1172
1173                         new_hash = fib_info_hashfn(fi);
1174                         dest = &new_info_hash[new_hash];
1175                         hlist_add_head(&fi->fib_hash, dest);
1176                 }
1177         }
1178         fib_info_hash = new_info_hash;
1179
1180         for (i = 0; i < old_size; i++) {
1181                 struct hlist_head *lhead = &fib_info_laddrhash[i];
1182                 struct hlist_node *n;
1183                 struct fib_info *fi;
1184
1185                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1186                         struct hlist_head *ldest;
1187                         unsigned int new_hash;
1188
1189                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1190                         ldest = &new_laddrhash[new_hash];
1191                         hlist_add_head(&fi->fib_lhash, ldest);
1192                 }
1193         }
1194         fib_info_laddrhash = new_laddrhash;
1195
1196         spin_unlock_bh(&fib_info_lock);
1197
1198         bytes = old_size * sizeof(struct hlist_head *);
1199         fib_info_hash_free(old_info_hash, bytes);
1200         fib_info_hash_free(old_laddrhash, bytes);
1201 }
1202
1203 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1204                                  unsigned char scope)
1205 {
1206         struct fib_nh *nh;
1207
1208         if (nhc->nhc_family != AF_INET)
1209                 return inet_select_addr(nhc->nhc_dev, 0, scope);
1210
1211         nh = container_of(nhc, struct fib_nh, nh_common);
1212         nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1213         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1214
1215         return nh->nh_saddr;
1216 }
1217
1218 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1219 {
1220         struct fib_nh_common *nhc = res->nhc;
1221
1222         if (res->fi->fib_prefsrc)
1223                 return res->fi->fib_prefsrc;
1224
1225         if (nhc->nhc_family == AF_INET) {
1226                 struct fib_nh *nh;
1227
1228                 nh = container_of(nhc, struct fib_nh, nh_common);
1229                 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1230                         return nh->nh_saddr;
1231         }
1232
1233         return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1234 }
1235
1236 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1237 {
1238         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1239             fib_prefsrc != cfg->fc_dst) {
1240                 u32 tb_id = cfg->fc_table;
1241                 int rc;
1242
1243                 if (tb_id == RT_TABLE_MAIN)
1244                         tb_id = RT_TABLE_LOCAL;
1245
1246                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1247                                           fib_prefsrc, tb_id);
1248
1249                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1250                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1251                                                   fib_prefsrc, RT_TABLE_LOCAL);
1252                 }
1253
1254                 if (rc != RTN_LOCAL)
1255                         return false;
1256         }
1257         return true;
1258 }
1259
1260 struct fib_info *fib_create_info(struct fib_config *cfg,
1261                                  struct netlink_ext_ack *extack)
1262 {
1263         int err;
1264         struct fib_info *fi = NULL;
1265         struct fib_info *ofi;
1266         int nhs = 1;
1267         struct net *net = cfg->fc_nlinfo.nl_net;
1268
1269         if (cfg->fc_type > RTN_MAX)
1270                 goto err_inval;
1271
1272         /* Fast check to catch the most weird cases */
1273         if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1274                 NL_SET_ERR_MSG(extack, "Invalid scope");
1275                 goto err_inval;
1276         }
1277
1278         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1279                 NL_SET_ERR_MSG(extack,
1280                                "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1281                 goto err_inval;
1282         }
1283
1284 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1285         if (cfg->fc_mp) {
1286                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1287                 if (nhs == 0)
1288                         goto err_inval;
1289         }
1290 #endif
1291
1292         err = -ENOBUFS;
1293         if (fib_info_cnt >= fib_info_hash_size) {
1294                 unsigned int new_size = fib_info_hash_size << 1;
1295                 struct hlist_head *new_info_hash;
1296                 struct hlist_head *new_laddrhash;
1297                 unsigned int bytes;
1298
1299                 if (!new_size)
1300                         new_size = 16;
1301                 bytes = new_size * sizeof(struct hlist_head *);
1302                 new_info_hash = fib_info_hash_alloc(bytes);
1303                 new_laddrhash = fib_info_hash_alloc(bytes);
1304                 if (!new_info_hash || !new_laddrhash) {
1305                         fib_info_hash_free(new_info_hash, bytes);
1306                         fib_info_hash_free(new_laddrhash, bytes);
1307                 } else
1308                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1309
1310                 if (!fib_info_hash_size)
1311                         goto failure;
1312         }
1313
1314         fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1315         if (!fi)
1316                 goto failure;
1317         fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1318                                               cfg->fc_mx_len, extack);
1319         if (unlikely(IS_ERR(fi->fib_metrics))) {
1320                 err = PTR_ERR(fi->fib_metrics);
1321                 kfree(fi);
1322                 return ERR_PTR(err);
1323         }
1324
1325         fib_info_cnt++;
1326         fi->fib_net = net;
1327         fi->fib_protocol = cfg->fc_protocol;
1328         fi->fib_scope = cfg->fc_scope;
1329         fi->fib_flags = cfg->fc_flags;
1330         fi->fib_priority = cfg->fc_priority;
1331         fi->fib_prefsrc = cfg->fc_prefsrc;
1332         fi->fib_type = cfg->fc_type;
1333         fi->fib_tb_id = cfg->fc_table;
1334
1335         fi->fib_nhs = nhs;
1336         change_nexthops(fi) {
1337                 nexthop_nh->nh_parent = fi;
1338         } endfor_nexthops(fi)
1339
1340         if (cfg->fc_mp)
1341                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1342         else
1343                 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1344
1345         if (err != 0)
1346                 goto failure;
1347
1348         if (fib_props[cfg->fc_type].error) {
1349                 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1350                         NL_SET_ERR_MSG(extack,
1351                                        "Gateway, device and multipath can not be specified for this route type");
1352                         goto err_inval;
1353                 }
1354                 goto link_it;
1355         } else {
1356                 switch (cfg->fc_type) {
1357                 case RTN_UNICAST:
1358                 case RTN_LOCAL:
1359                 case RTN_BROADCAST:
1360                 case RTN_ANYCAST:
1361                 case RTN_MULTICAST:
1362                         break;
1363                 default:
1364                         NL_SET_ERR_MSG(extack, "Invalid route type");
1365                         goto err_inval;
1366                 }
1367         }
1368
1369         if (cfg->fc_scope > RT_SCOPE_HOST) {
1370                 NL_SET_ERR_MSG(extack, "Invalid scope");
1371                 goto err_inval;
1372         }
1373
1374         if (cfg->fc_scope == RT_SCOPE_HOST) {
1375                 struct fib_nh *nh = fi->fib_nh;
1376
1377                 /* Local address is added. */
1378                 if (nhs != 1) {
1379                         NL_SET_ERR_MSG(extack,
1380                                        "Route with host scope can not have multiple nexthops");
1381                         goto err_inval;
1382                 }
1383                 if (nh->fib_nh_gw_family) {
1384                         NL_SET_ERR_MSG(extack,
1385                                        "Route with host scope can not have a gateway");
1386                         goto err_inval;
1387                 }
1388                 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1389                 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1390                 err = -ENODEV;
1391                 if (!nh->fib_nh_dev)
1392                         goto failure;
1393         } else {
1394                 int linkdown = 0;
1395
1396                 change_nexthops(fi) {
1397                         err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1398                                            cfg->fc_table, cfg->fc_scope,
1399                                            extack);
1400                         if (err != 0)
1401                                 goto failure;
1402                         if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1403                                 linkdown++;
1404                 } endfor_nexthops(fi)
1405                 if (linkdown == fi->fib_nhs)
1406                         fi->fib_flags |= RTNH_F_LINKDOWN;
1407         }
1408
1409         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1410                 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1411                 goto err_inval;
1412         }
1413
1414         change_nexthops(fi) {
1415                 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1416                                           fi->fib_scope);
1417                 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1418                         fi->fib_nh_is_v6 = true;
1419         } endfor_nexthops(fi)
1420
1421         fib_rebalance(fi);
1422
1423 link_it:
1424         ofi = fib_find_info(fi);
1425         if (ofi) {
1426                 fi->fib_dead = 1;
1427                 free_fib_info(fi);
1428                 ofi->fib_treeref++;
1429                 return ofi;
1430         }
1431
1432         fi->fib_treeref++;
1433         refcount_set(&fi->fib_clntref, 1);
1434         spin_lock_bh(&fib_info_lock);
1435         hlist_add_head(&fi->fib_hash,
1436                        &fib_info_hash[fib_info_hashfn(fi)]);
1437         if (fi->fib_prefsrc) {
1438                 struct hlist_head *head;
1439
1440                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1441                 hlist_add_head(&fi->fib_lhash, head);
1442         }
1443         change_nexthops(fi) {
1444                 struct hlist_head *head;
1445                 unsigned int hash;
1446
1447                 if (!nexthop_nh->fib_nh_dev)
1448                         continue;
1449                 hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1450                 head = &fib_info_devhash[hash];
1451                 hlist_add_head(&nexthop_nh->nh_hash, head);
1452         } endfor_nexthops(fi)
1453         spin_unlock_bh(&fib_info_lock);
1454         return fi;
1455
1456 err_inval:
1457         err = -EINVAL;
1458
1459 failure:
1460         if (fi) {
1461                 fi->fib_dead = 1;
1462                 free_fib_info(fi);
1463         }
1464
1465         return ERR_PTR(err);
1466 }
1467
1468 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1469                      unsigned char *flags, bool skip_oif)
1470 {
1471         if (nhc->nhc_flags & RTNH_F_DEAD)
1472                 *flags |= RTNH_F_DEAD;
1473
1474         if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1475                 *flags |= RTNH_F_LINKDOWN;
1476
1477                 rcu_read_lock();
1478                 switch (nhc->nhc_family) {
1479                 case AF_INET:
1480                         if (ip_ignore_linkdown(nhc->nhc_dev))
1481                                 *flags |= RTNH_F_DEAD;
1482                         break;
1483                 case AF_INET6:
1484                         if (ip6_ignore_linkdown(nhc->nhc_dev))
1485                                 *flags |= RTNH_F_DEAD;
1486                         break;
1487                 }
1488                 rcu_read_unlock();
1489         }
1490
1491         switch (nhc->nhc_gw_family) {
1492         case AF_INET:
1493                 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1494                         goto nla_put_failure;
1495                 break;
1496         case AF_INET6:
1497                 /* if gateway family does not match nexthop family
1498                  * gateway is encoded as RTA_VIA
1499                  */
1500                 if (nhc->nhc_gw_family != nhc->nhc_family) {
1501                         int alen = sizeof(struct in6_addr);
1502                         struct nlattr *nla;
1503                         struct rtvia *via;
1504
1505                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
1506                         if (!nla)
1507                                 goto nla_put_failure;
1508
1509                         via = nla_data(nla);
1510                         via->rtvia_family = AF_INET6;
1511                         memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1512                 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1513                                             &nhc->nhc_gw.ipv6) < 0) {
1514                         goto nla_put_failure;
1515                 }
1516                 break;
1517         }
1518
1519         *flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1520         if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1521                 *flags |= RTNH_F_OFFLOAD;
1522
1523         if (!skip_oif && nhc->nhc_dev &&
1524             nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1525                 goto nla_put_failure;
1526
1527         if (nhc->nhc_lwtstate &&
1528             lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1529                                 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1530                 goto nla_put_failure;
1531
1532         return 0;
1533
1534 nla_put_failure:
1535         return -EMSGSIZE;
1536 }
1537 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1538
1539 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1540 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1541                     int nh_weight)
1542 {
1543         const struct net_device *dev = nhc->nhc_dev;
1544         struct rtnexthop *rtnh;
1545         unsigned char flags = 0;
1546
1547         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1548         if (!rtnh)
1549                 goto nla_put_failure;
1550
1551         rtnh->rtnh_hops = nh_weight - 1;
1552         rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1553
1554         if (fib_nexthop_info(skb, nhc, &flags, true) < 0)
1555                 goto nla_put_failure;
1556
1557         rtnh->rtnh_flags = flags;
1558
1559         /* length of rtnetlink header + attributes */
1560         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1561
1562         return 0;
1563
1564 nla_put_failure:
1565         return -EMSGSIZE;
1566 }
1567 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1568 #endif
1569
1570 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1571 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1572 {
1573         struct nlattr *mp;
1574
1575         mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1576         if (!mp)
1577                 goto nla_put_failure;
1578
1579         for_nexthops(fi) {
1580                 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight) < 0)
1581                         goto nla_put_failure;
1582 #ifdef CONFIG_IP_ROUTE_CLASSID
1583                 if (nh->nh_tclassid &&
1584                     nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1585                         goto nla_put_failure;
1586 #endif
1587         } endfor_nexthops(fi);
1588
1589         nla_nest_end(skb, mp);
1590
1591         return 0;
1592
1593 nla_put_failure:
1594         return -EMSGSIZE;
1595 }
1596 #else
1597 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1598 {
1599         return 0;
1600 }
1601 #endif
1602
1603 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1604                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1605                   struct fib_info *fi, unsigned int flags)
1606 {
1607         unsigned int nhs = fib_info_num_path(fi);
1608         struct nlmsghdr *nlh;
1609         struct rtmsg *rtm;
1610
1611         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1612         if (!nlh)
1613                 return -EMSGSIZE;
1614
1615         rtm = nlmsg_data(nlh);
1616         rtm->rtm_family = AF_INET;
1617         rtm->rtm_dst_len = dst_len;
1618         rtm->rtm_src_len = 0;
1619         rtm->rtm_tos = tos;
1620         if (tb_id < 256)
1621                 rtm->rtm_table = tb_id;
1622         else
1623                 rtm->rtm_table = RT_TABLE_COMPAT;
1624         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1625                 goto nla_put_failure;
1626         rtm->rtm_type = type;
1627         rtm->rtm_flags = fi->fib_flags;
1628         rtm->rtm_scope = fi->fib_scope;
1629         rtm->rtm_protocol = fi->fib_protocol;
1630
1631         if (rtm->rtm_dst_len &&
1632             nla_put_in_addr(skb, RTA_DST, dst))
1633                 goto nla_put_failure;
1634         if (fi->fib_priority &&
1635             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1636                 goto nla_put_failure;
1637         if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1638                 goto nla_put_failure;
1639
1640         if (fi->fib_prefsrc &&
1641             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1642                 goto nla_put_failure;
1643         if (nhs == 1) {
1644                 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1645                 unsigned char flags = 0;
1646
1647                 if (fib_nexthop_info(skb, nhc, &flags, false) < 0)
1648                         goto nla_put_failure;
1649
1650                 rtm->rtm_flags = flags;
1651 #ifdef CONFIG_IP_ROUTE_CLASSID
1652                 if (nhc->nhc_family == AF_INET) {
1653                         struct fib_nh *nh;
1654
1655                         nh = container_of(nhc, struct fib_nh, nh_common);
1656                         if (nh->nh_tclassid &&
1657                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1658                                 goto nla_put_failure;
1659                 }
1660 #endif
1661         } else {
1662                 if (fib_add_multipath(skb, fi) < 0)
1663                         goto nla_put_failure;
1664         }
1665
1666         nlmsg_end(skb, nlh);
1667         return 0;
1668
1669 nla_put_failure:
1670         nlmsg_cancel(skb, nlh);
1671         return -EMSGSIZE;
1672 }
1673
1674 /*
1675  * Update FIB if:
1676  * - local address disappeared -> we must delete all the entries
1677  *   referring to it.
1678  * - device went down -> we must shutdown all nexthops going via it.
1679  */
1680 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1681 {
1682         int ret = 0;
1683         unsigned int hash = fib_laddr_hashfn(local);
1684         struct hlist_head *head = &fib_info_laddrhash[hash];
1685         struct net *net = dev_net(dev);
1686         int tb_id = l3mdev_fib_table(dev);
1687         struct fib_info *fi;
1688
1689         if (!fib_info_laddrhash || local == 0)
1690                 return 0;
1691
1692         hlist_for_each_entry(fi, head, fib_lhash) {
1693                 if (!net_eq(fi->fib_net, net) ||
1694                     fi->fib_tb_id != tb_id)
1695                         continue;
1696                 if (fi->fib_prefsrc == local) {
1697                         fi->fib_flags |= RTNH_F_DEAD;
1698                         ret++;
1699                 }
1700         }
1701         return ret;
1702 }
1703
1704 static int call_fib_nh_notifiers(struct fib_nh *nh,
1705                                  enum fib_event_type event_type)
1706 {
1707         bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1708         struct fib_nh_notifier_info info = {
1709                 .fib_nh = nh,
1710         };
1711
1712         switch (event_type) {
1713         case FIB_EVENT_NH_ADD:
1714                 if (nh->fib_nh_flags & RTNH_F_DEAD)
1715                         break;
1716                 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1717                         break;
1718                 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1719                                            &info.info);
1720         case FIB_EVENT_NH_DEL:
1721                 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1722                     (nh->fib_nh_flags & RTNH_F_DEAD))
1723                         return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1724                                                    event_type, &info.info);
1725         default:
1726                 break;
1727         }
1728
1729         return NOTIFY_DONE;
1730 }
1731
1732 /* Update the PMTU of exceptions when:
1733  * - the new MTU of the first hop becomes smaller than the PMTU
1734  * - the old MTU was the same as the PMTU, and it limited discovery of
1735  *   larger MTUs on the path. With that limit raised, we can now
1736  *   discover larger MTUs
1737  * A special case is locked exceptions, for which the PMTU is smaller
1738  * than the minimal accepted PMTU:
1739  * - if the new MTU is greater than the PMTU, don't make any change
1740  * - otherwise, unlock and set PMTU
1741  */
1742 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1743 {
1744         struct fnhe_hash_bucket *bucket;
1745         int i;
1746
1747         bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1748         if (!bucket)
1749                 return;
1750
1751         for (i = 0; i < FNHE_HASH_SIZE; i++) {
1752                 struct fib_nh_exception *fnhe;
1753
1754                 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1755                      fnhe;
1756                      fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1757                         if (fnhe->fnhe_mtu_locked) {
1758                                 if (new <= fnhe->fnhe_pmtu) {
1759                                         fnhe->fnhe_pmtu = new;
1760                                         fnhe->fnhe_mtu_locked = false;
1761                                 }
1762                         } else if (new < fnhe->fnhe_pmtu ||
1763                                    orig == fnhe->fnhe_pmtu) {
1764                                 fnhe->fnhe_pmtu = new;
1765                         }
1766                 }
1767         }
1768 }
1769
1770 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1771 {
1772         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1773         struct hlist_head *head = &fib_info_devhash[hash];
1774         struct fib_nh *nh;
1775
1776         hlist_for_each_entry(nh, head, nh_hash) {
1777                 if (nh->fib_nh_dev == dev)
1778                         fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1779         }
1780 }
1781
1782 /* Event              force Flags           Description
1783  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1784  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1785  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1786  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1787  */
1788 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1789 {
1790         int ret = 0;
1791         int scope = RT_SCOPE_NOWHERE;
1792         struct fib_info *prev_fi = NULL;
1793         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1794         struct hlist_head *head = &fib_info_devhash[hash];
1795         struct fib_nh *nh;
1796
1797         if (force)
1798                 scope = -1;
1799
1800         hlist_for_each_entry(nh, head, nh_hash) {
1801                 struct fib_info *fi = nh->nh_parent;
1802                 int dead;
1803
1804                 BUG_ON(!fi->fib_nhs);
1805                 if (nh->fib_nh_dev != dev || fi == prev_fi)
1806                         continue;
1807                 prev_fi = fi;
1808                 dead = 0;
1809                 change_nexthops(fi) {
1810                         if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1811                                 dead++;
1812                         else if (nexthop_nh->fib_nh_dev == dev &&
1813                                  nexthop_nh->fib_nh_scope != scope) {
1814                                 switch (event) {
1815                                 case NETDEV_DOWN:
1816                                 case NETDEV_UNREGISTER:
1817                                         nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1818                                         /* fall through */
1819                                 case NETDEV_CHANGE:
1820                                         nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1821                                         break;
1822                                 }
1823                                 call_fib_nh_notifiers(nexthop_nh,
1824                                                       FIB_EVENT_NH_DEL);
1825                                 dead++;
1826                         }
1827 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1828                         if (event == NETDEV_UNREGISTER &&
1829                             nexthop_nh->fib_nh_dev == dev) {
1830                                 dead = fi->fib_nhs;
1831                                 break;
1832                         }
1833 #endif
1834                 } endfor_nexthops(fi)
1835                 if (dead == fi->fib_nhs) {
1836                         switch (event) {
1837                         case NETDEV_DOWN:
1838                         case NETDEV_UNREGISTER:
1839                                 fi->fib_flags |= RTNH_F_DEAD;
1840                                 /* fall through */
1841                         case NETDEV_CHANGE:
1842                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1843                                 break;
1844                         }
1845                         ret++;
1846                 }
1847
1848                 fib_rebalance(fi);
1849         }
1850
1851         return ret;
1852 }
1853
1854 /* Must be invoked inside of an RCU protected region.  */
1855 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1856 {
1857         struct fib_info *fi = NULL, *last_resort = NULL;
1858         struct hlist_head *fa_head = res->fa_head;
1859         struct fib_table *tb = res->table;
1860         u8 slen = 32 - res->prefixlen;
1861         int order = -1, last_idx = -1;
1862         struct fib_alias *fa, *fa1 = NULL;
1863         u32 last_prio = res->fi->fib_priority;
1864         u8 last_tos = 0;
1865
1866         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1867                 struct fib_info *next_fi = fa->fa_info;
1868                 struct fib_nh *nh;
1869
1870                 if (fa->fa_slen != slen)
1871                         continue;
1872                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1873                         continue;
1874                 if (fa->tb_id != tb->tb_id)
1875                         continue;
1876                 if (next_fi->fib_priority > last_prio &&
1877                     fa->fa_tos == last_tos) {
1878                         if (last_tos)
1879                                 continue;
1880                         break;
1881                 }
1882                 if (next_fi->fib_flags & RTNH_F_DEAD)
1883                         continue;
1884                 last_tos = fa->fa_tos;
1885                 last_prio = next_fi->fib_priority;
1886
1887                 if (next_fi->fib_scope != res->scope ||
1888                     fa->fa_type != RTN_UNICAST)
1889                         continue;
1890
1891                 nh = fib_info_nh(next_fi, 0);
1892                 if (!nh->fib_nh_gw4 || nh->fib_nh_scope != RT_SCOPE_LINK)
1893                         continue;
1894
1895                 fib_alias_accessed(fa);
1896
1897                 if (!fi) {
1898                         if (next_fi != res->fi)
1899                                 break;
1900                         fa1 = fa;
1901                 } else if (!fib_detect_death(fi, order, &last_resort,
1902                                              &last_idx, fa1->fa_default)) {
1903                         fib_result_assign(res, fi);
1904                         fa1->fa_default = order;
1905                         goto out;
1906                 }
1907                 fi = next_fi;
1908                 order++;
1909         }
1910
1911         if (order <= 0 || !fi) {
1912                 if (fa1)
1913                         fa1->fa_default = -1;
1914                 goto out;
1915         }
1916
1917         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1918                               fa1->fa_default)) {
1919                 fib_result_assign(res, fi);
1920                 fa1->fa_default = order;
1921                 goto out;
1922         }
1923
1924         if (last_idx >= 0)
1925                 fib_result_assign(res, last_resort);
1926         fa1->fa_default = last_idx;
1927 out:
1928         return;
1929 }
1930
1931 /*
1932  * Dead device goes up. We wake up dead nexthops.
1933  * It takes sense only on multipath routes.
1934  */
1935 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
1936 {
1937         struct fib_info *prev_fi;
1938         unsigned int hash;
1939         struct hlist_head *head;
1940         struct fib_nh *nh;
1941         int ret;
1942
1943         if (!(dev->flags & IFF_UP))
1944                 return 0;
1945
1946         if (nh_flags & RTNH_F_DEAD) {
1947                 unsigned int flags = dev_get_flags(dev);
1948
1949                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1950                         nh_flags |= RTNH_F_LINKDOWN;
1951         }
1952
1953         prev_fi = NULL;
1954         hash = fib_devindex_hashfn(dev->ifindex);
1955         head = &fib_info_devhash[hash];
1956         ret = 0;
1957
1958         hlist_for_each_entry(nh, head, nh_hash) {
1959                 struct fib_info *fi = nh->nh_parent;
1960                 int alive;
1961
1962                 BUG_ON(!fi->fib_nhs);
1963                 if (nh->fib_nh_dev != dev || fi == prev_fi)
1964                         continue;
1965
1966                 prev_fi = fi;
1967                 alive = 0;
1968                 change_nexthops(fi) {
1969                         if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
1970                                 alive++;
1971                                 continue;
1972                         }
1973                         if (!nexthop_nh->fib_nh_dev ||
1974                             !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
1975                                 continue;
1976                         if (nexthop_nh->fib_nh_dev != dev ||
1977                             !__in_dev_get_rtnl(dev))
1978                                 continue;
1979                         alive++;
1980                         nexthop_nh->fib_nh_flags &= ~nh_flags;
1981                         call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1982                 } endfor_nexthops(fi)
1983
1984                 if (alive > 0) {
1985                         fi->fib_flags &= ~nh_flags;
1986                         ret++;
1987                 }
1988
1989                 fib_rebalance(fi);
1990         }
1991
1992         return ret;
1993 }
1994
1995 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1996 static bool fib_good_nh(const struct fib_nh *nh)
1997 {
1998         int state = NUD_REACHABLE;
1999
2000         if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2001                 struct neighbour *n;
2002
2003                 rcu_read_lock_bh();
2004
2005                 if (likely(nh->fib_nh_gw_family == AF_INET))
2006                         n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2007                                                    (__force u32)nh->fib_nh_gw4);
2008                 else if (nh->fib_nh_gw_family == AF_INET6)
2009                         n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2010                                                            &nh->fib_nh_gw6);
2011                 else
2012                         n = NULL;
2013                 if (n)
2014                         state = n->nud_state;
2015
2016                 rcu_read_unlock_bh();
2017         }
2018
2019         return !!(state & NUD_VALID);
2020 }
2021
2022 void fib_select_multipath(struct fib_result *res, int hash)
2023 {
2024         struct fib_info *fi = res->fi;
2025         struct net *net = fi->fib_net;
2026         bool first = false;
2027
2028         change_nexthops(fi) {
2029                 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2030                         if (!fib_good_nh(nexthop_nh))
2031                                 continue;
2032                         if (!first) {
2033                                 res->nh_sel = nhsel;
2034                                 res->nhc = &nexthop_nh->nh_common;
2035                                 first = true;
2036                         }
2037                 }
2038
2039                 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2040                         continue;
2041
2042                 res->nh_sel = nhsel;
2043                 res->nhc = &nexthop_nh->nh_common;
2044                 return;
2045         } endfor_nexthops(fi);
2046 }
2047 #endif
2048
2049 void fib_select_path(struct net *net, struct fib_result *res,
2050                      struct flowi4 *fl4, const struct sk_buff *skb)
2051 {
2052         if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2053                 goto check_saddr;
2054
2055 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2056         if (fib_info_num_path(res->fi) > 1) {
2057                 int h = fib_multipath_hash(net, fl4, skb, NULL);
2058
2059                 fib_select_multipath(res, h);
2060         }
2061         else
2062 #endif
2063         if (!res->prefixlen &&
2064             res->table->tb_num_default > 1 &&
2065             res->type == RTN_UNICAST)
2066                 fib_select_default(fl4, res);
2067
2068 check_saddr:
2069         if (!fl4->saddr)
2070                 fl4->saddr = fib_result_prefsrc(net, res);
2071 }