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
  36 #include <net/arp.h>
  37 #include <net/ip.h>
  38 #include <net/protocol.h>
  39 #include <net/route.h>
  40 #include <net/tcp.h>
  41 #include <net/sock.h>
  42 #include <net/ip_fib.h>
  43 #include <net/netlink.h>
  44 #include <net/nexthop.h>
  45 #include <net/lwtunnel.h>
  46
  47 #include "fib_lookup.h"
  48
  49 static DEFINE_SPINLOCK(fib_info_lock);
  50 static struct hlist_head *fib_info_hash;
  51 static struct hlist_head *fib_info_laddrhash;
  52 static unsigned int fib_info_hash_size;
  53 static unsigned int fib_info_cnt;
  54
  55 #define DEVINDEX_HASHBITS 8
  56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  58
  59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  60
  61 #define for_nexthops(fi) {                                              \
  62         int nhsel; const struct fib_nh *nh;                             \
  63         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  64              nhsel < (fi)->fib_nhs;                                     \
  65              nh++, nhsel++)
  66
  67 #define change_nexthops(fi) {                                           \
  68         int nhsel; struct fib_nh *nexthop_nh;                           \
  69         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  70              nhsel < (fi)->fib_nhs;                                     \
  71              nexthop_nh++, nhsel++)
  72
  73 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  74
  75 /* Hope, that gcc will optimize it to get rid of dummy loop */
  76
  77 #define for_nexthops(fi) {                                              \
  78         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  79         for (nhsel = 0; nhsel < 1; nhsel++)
  80
  81 #define change_nexthops(fi) {                                           \
  82         int nhsel;                                                      \
  83         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  84         for (nhsel = 0; nhsel < 1; nhsel++)
  85
  86 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  87
  88 #define endfor_nexthops(fi) }
  89
  90
  91 const struct fib_prop fib_props[RTN_MAX + 1] = {
  92         [RTN_UNSPEC] = {
  93                 .error  = 0,
  94                 .scope  = RT_SCOPE_NOWHERE,
  95         },
  96         [RTN_UNICAST] = {
  97                 .error  = 0,
  98                 .scope  = RT_SCOPE_UNIVERSE,
  99         },
 100         [RTN_LOCAL] = {
 101                 .error  = 0,
 102                 .scope  = RT_SCOPE_HOST,
 103         },
 104         [RTN_BROADCAST] = {
 105                 .error  = 0,
 106                 .scope  = RT_SCOPE_LINK,
 107         },
 108         [RTN_ANYCAST] = {
 109                 .error  = 0,
 110                 .scope  = RT_SCOPE_LINK,
 111         },
 112         [RTN_MULTICAST] = {
 113                 .error  = 0,
 114                 .scope  = RT_SCOPE_UNIVERSE,
 115         },
 116         [RTN_BLACKHOLE] = {
 117                 .error  = -EINVAL,
 118                 .scope  = RT_SCOPE_UNIVERSE,
 119         },
 120         [RTN_UNREACHABLE] = {
 121                 .error  = -EHOSTUNREACH,
 122                 .scope  = RT_SCOPE_UNIVERSE,
 123         },
 124         [RTN_PROHIBIT] = {
 125                 .error  = -EACCES,
 126                 .scope  = RT_SCOPE_UNIVERSE,
 127         },
 128         [RTN_THROW] = {
 129                 .error  = -EAGAIN,
 130                 .scope  = RT_SCOPE_UNIVERSE,
 131         },
 132         [RTN_NAT] = {
 133                 .error  = -EINVAL,
 134                 .scope  = RT_SCOPE_NOWHERE,
 135         },
 136         [RTN_XRESOLVE] = {
 137                 .error  = -EINVAL,
 138                 .scope  = RT_SCOPE_NOWHERE,
 139         },
 140 };
 141
 142 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 143 {
 144         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 145
 146         if (!rt)
 147                 return;
 148
 149         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 150          * because we waited an RCU grace period before calling
 151          * free_fib_info_rcu()
 152          */
 153
 154         dst_free(&rt->dst);
 155 }
 156
 157 static void free_nh_exceptions(struct fib_nh *nh)
 158 {
 159         struct fnhe_hash_bucket *hash;
 160         int i;
 161
 162         hash = rcu_dereference_protected(nh->nh_exceptions, 1);
 163         if (!hash)
 164                 return;
 165         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 166                 struct fib_nh_exception *fnhe;
 167
 168                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 169                 while (fnhe) {
 170                         struct fib_nh_exception *next;
 171
 172                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 173
 174                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 175                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 176
 177                         kfree(fnhe);
 178
 179                         fnhe = next;
 180                 }
 181         }
 182         kfree(hash);
 183 }
 184
 185 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 186 {
 187         int cpu;
 188
 189         if (!rtp)
 190                 return;
 191
 192         for_each_possible_cpu(cpu) {
 193                 struct rtable *rt;
 194
 195                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 196                 if (rt)
 197                         dst_free(&rt->dst);
 198         }
 199         free_percpu(rtp);
 200 }
 201
 202 /* Release a nexthop info record */
 203 static void free_fib_info_rcu(struct rcu_head *head)
 204 {
 205         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 206
 207         change_nexthops(fi) {
 208                 if (nexthop_nh->nh_dev)
 209                         dev_put(nexthop_nh->nh_dev);
 210                 lwtstate_put(nexthop_nh->nh_lwtstate);
 211                 free_nh_exceptions(nexthop_nh);
 212                 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 213                 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
 214         } endfor_nexthops(fi);
 215
 216         if (fi->fib_metrics != (u32 *) dst_default_metrics)
 217                 kfree(fi->fib_metrics);
 218         kfree(fi);
 219 }
 220
 221 void free_fib_info(struct fib_info *fi)
 222 {
 223         if (fi->fib_dead == 0) {
 224                 pr_warn("Freeing alive fib_info %p\n", fi);
 225                 return;
 226         }
 227         fib_info_cnt--;
 228 #ifdef CONFIG_IP_ROUTE_CLASSID
 229         change_nexthops(fi) {
 230                 if (nexthop_nh->nh_tclassid)
 231                         fi->fib_net->ipv4.fib_num_tclassid_users--;
 232         } endfor_nexthops(fi);
 233 #endif
 234         call_rcu(&fi->rcu, free_fib_info_rcu);
 235 }
 236 EXPORT_SYMBOL_GPL(free_fib_info);
 237
 238 void fib_release_info(struct fib_info *fi)
 239 {
 240         spin_lock_bh(&fib_info_lock);
 241         if (fi && --fi->fib_treeref == 0) {
 242                 hlist_del(&fi->fib_hash);
 243                 if (fi->fib_prefsrc)
 244                         hlist_del(&fi->fib_lhash);
 245                 change_nexthops(fi) {
 246                         if (!nexthop_nh->nh_dev)
 247                                 continue;
 248                         hlist_del(&nexthop_nh->nh_hash);
 249                 } endfor_nexthops(fi)
 250                 fi->fib_dead = 1;
 251                 fib_info_put(fi);
 252         }
 253         spin_unlock_bh(&fib_info_lock);
 254 }
 255
 256 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 257 {
 258         const struct fib_nh *onh = ofi->fib_nh;
 259
 260         for_nexthops(fi) {
 261                 if (nh->nh_oif != onh->nh_oif ||
 262                     nh->nh_gw  != onh->nh_gw ||
 263                     nh->nh_scope != onh->nh_scope ||
 264 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 265                     nh->nh_weight != onh->nh_weight ||
 266 #endif
 267 #ifdef CONFIG_IP_ROUTE_CLASSID
 268                     nh->nh_tclassid != onh->nh_tclassid ||
 269 #endif
 270                     lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
 271                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
 272                         return -1;
 273                 onh++;
 274         } endfor_nexthops(fi);
 275         return 0;
 276 }
 277
 278 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 279 {
 280         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 281
 282         return (val ^
 283                 (val >> DEVINDEX_HASHBITS) ^
 284                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 285 }
 286
 287 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 288 {
 289         unsigned int mask = (fib_info_hash_size - 1);
 290         unsigned int val = fi->fib_nhs;
 291
 292         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 293         val ^= (__force u32)fi->fib_prefsrc;
 294         val ^= fi->fib_priority;
 295         for_nexthops(fi) {
 296                 val ^= fib_devindex_hashfn(nh->nh_oif);
 297         } endfor_nexthops(fi)
 298
 299         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 300 }
 301
 302 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 303 {
 304         struct hlist_head *head;
 305         struct fib_info *fi;
 306         unsigned int hash;
 307
 308         hash = fib_info_hashfn(nfi);
 309         head = &fib_info_hash[hash];
 310
 311         hlist_for_each_entry(fi, head, fib_hash) {
 312                 if (!net_eq(fi->fib_net, nfi->fib_net))
 313                         continue;
 314                 if (fi->fib_nhs != nfi->fib_nhs)
 315                         continue;
 316                 if (nfi->fib_protocol == fi->fib_protocol &&
 317                     nfi->fib_scope == fi->fib_scope &&
 318                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 319                     nfi->fib_priority == fi->fib_priority &&
 320                     nfi->fib_type == fi->fib_type &&
 321                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 322                            sizeof(u32) * RTAX_MAX) == 0 &&
 323                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 324                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 325                         return fi;
 326         }
 327
 328         return NULL;
 329 }
 330
 331 /* Check, that the gateway is already configured.
 332  * Used only by redirect accept routine.
 333  */
 334 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 335 {
 336         struct hlist_head *head;
 337         struct fib_nh *nh;
 338         unsigned int hash;
 339
 340         spin_lock(&fib_info_lock);
 341
 342         hash = fib_devindex_hashfn(dev->ifindex);
 343         head = &fib_info_devhash[hash];
 344         hlist_for_each_entry(nh, head, nh_hash) {
 345                 if (nh->nh_dev == dev &&
 346                     nh->nh_gw == gw &&
 347                     !(nh->nh_flags & RTNH_F_DEAD)) {
 348                         spin_unlock(&fib_info_lock);
 349                         return 0;
 350                 }
 351         }
 352
 353         spin_unlock(&fib_info_lock);
 354
 355         return -1;
 356 }
 357
 358 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 359 {
 360         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 361                          + nla_total_size(4) /* RTA_TABLE */
 362                          + nla_total_size(4) /* RTA_DST */
 363                          + nla_total_size(4) /* RTA_PRIORITY */
 364                          + nla_total_size(4) /* RTA_PREFSRC */
 365                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 366
 367         /* space for nested metrics */
 368         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 369
 370         if (fi->fib_nhs) {
 371                 size_t nh_encapsize = 0;
 372                 /* Also handles the special case fib_nhs == 1 */
 373
 374                 /* each nexthop is packed in an attribute */
 375                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 376
 377                 /* may contain flow and gateway attribute */
 378                 nhsize += 2 * nla_total_size(4);
 379
 380                 /* grab encap info */
 381                 for_nexthops(fi) {
 382                         if (nh->nh_lwtstate) {
 383                                 /* RTA_ENCAP_TYPE */
 384                                 nh_encapsize += lwtunnel_get_encap_size(
 385                                                 nh->nh_lwtstate);
 386                                 /* RTA_ENCAP */
 387                                 nh_encapsize +=  nla_total_size(2);
 388                         }
 389                 } endfor_nexthops(fi);
 390
 391                 /* all nexthops are packed in a nested attribute */
 392                 payload += nla_total_size((fi->fib_nhs * nhsize) +
 393                                           nh_encapsize);
 394
 395         }
 396
 397         return payload;
 398 }
 399
 400 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 401                int dst_len, u32 tb_id, const struct nl_info *info,
 402                unsigned int nlm_flags)
 403 {
 404         struct sk_buff *skb;
 405         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 406         int err = -ENOBUFS;
 407
 408         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 409         if (!skb)
 410                 goto errout;
 411
 412         err = fib_dump_info(skb, info->portid, seq, event, tb_id,
 413                             fa->fa_type, key, dst_len,
 414                             fa->fa_tos, fa->fa_info, nlm_flags);
 415         if (err < 0) {
 416                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 417                 WARN_ON(err == -EMSGSIZE);
 418                 kfree_skb(skb);
 419                 goto errout;
 420         }
 421         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 422                     info->nlh, GFP_KERNEL);
 423         return;
 424 errout:
 425         if (err < 0)
 426                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 427 }
 428
 429 static int fib_detect_death(struct fib_info *fi, int order,
 430                             struct fib_info **last_resort, int *last_idx,
 431                             int dflt)
 432 {
 433         struct neighbour *n;
 434         int state = NUD_NONE;
 435
 436         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 437         if (n) {
 438                 state = n->nud_state;
 439                 neigh_release(n);
 440         } else {
 441                 return 0;
 442         }
 443         if (state == NUD_REACHABLE)
 444                 return 0;
 445         if ((state & NUD_VALID) && order != dflt)
 446                 return 0;
 447         if ((state & NUD_VALID) ||
 448             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 449                 *last_resort = fi;
 450                 *last_idx = order;
 451         }
 452         return 1;
 453 }
 454
 455 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 456
 457 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
 458                               struct netlink_ext_ack *extack)
 459 {
 460         int nhs = 0;
 461
 462         while (rtnh_ok(rtnh, remaining)) {
 463                 nhs++;
 464                 rtnh = rtnh_next(rtnh, &remaining);
 465         }
 466
 467         /* leftover implies invalid nexthop configuration, discard it */
 468         return remaining > 0 ? 0 : nhs;
 469 }
 470
 471 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 472                        int remaining, struct fib_config *cfg,
 473                        struct netlink_ext_ack *extack)
 474 {
 475         int ret;
 476
 477         change_nexthops(fi) {
 478                 int attrlen;
 479
 480                 if (!rtnh_ok(rtnh, remaining))
 481                         return -EINVAL;
 482
 483                 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
 484                         return -EINVAL;
 485
 486                 nexthop_nh->nh_flags =
 487                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 488                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 489                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 490
 491                 attrlen = rtnh_attrlen(rtnh);
 492                 if (attrlen > 0) {
 493                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 494
 495                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 496                         nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
 497 #ifdef CONFIG_IP_ROUTE_CLASSID
 498                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 499                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 500                         if (nexthop_nh->nh_tclassid)
 501                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 502 #endif
 503                         nla = nla_find(attrs, attrlen, RTA_ENCAP);
 504                         if (nla) {
 505                                 struct lwtunnel_state *lwtstate;
 506                                 struct nlattr *nla_entype;
 507
 508                                 nla_entype = nla_find(attrs, attrlen,
 509                                                       RTA_ENCAP_TYPE);
 510                                 if (!nla_entype)
 511                                         goto err_inval;
 512
 513                                 ret = lwtunnel_build_state(nla_get_u16(
 514                                                            nla_entype),
 515                                                            nla,  AF_INET, cfg,
 516                                                            &lwtstate);
 517                                 if (ret)
 518                                         goto errout;
 519                                 nexthop_nh->nh_lwtstate =
 520                                         lwtstate_get(lwtstate);
 521                         }
 522                 }
 523
 524                 rtnh = rtnh_next(rtnh, &remaining);
 525         } endfor_nexthops(fi);
 526
 527         return 0;
 528
 529 err_inval:
 530         ret = -EINVAL;
 531
 532 errout:
 533         return ret;
 534 }
 535
 536 static void fib_rebalance(struct fib_info *fi)
 537 {
 538         int total;
 539         int w;
 540         struct in_device *in_dev;
 541
 542         if (fi->fib_nhs < 2)
 543                 return;
 544
 545         total = 0;
 546         for_nexthops(fi) {
 547                 if (nh->nh_flags & RTNH_F_DEAD)
 548                         continue;
 549
 550                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
 551
 552                 if (in_dev &&
 553                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
 554                     nh->nh_flags & RTNH_F_LINKDOWN)
 555                         continue;
 556
 557                 total += nh->nh_weight;
 558         } endfor_nexthops(fi);
 559
 560         w = 0;
 561         change_nexthops(fi) {
 562                 int upper_bound;
 563
 564                 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
 565
 566                 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
 567                         upper_bound = -1;
 568                 } else if (in_dev &&
 569                            IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
 570                            nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
 571                         upper_bound = -1;
 572                 } else {
 573                         w += nexthop_nh->nh_weight;
 574                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 575                                                             total) - 1;
 576                 }
 577
 578                 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
 579         } endfor_nexthops(fi);
 580 }
 581
 582 static inline void fib_add_weight(struct fib_info *fi,
 583                                   const struct fib_nh *nh)
 584 {
 585         fi->fib_weight += nh->nh_weight;
 586 }
 587
 588 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 589
 590 #define fib_rebalance(fi) do { } while (0)
 591 #define fib_add_weight(fi, nh) do { } while (0)
 592
 593 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 594
 595 static int fib_encap_match(u16 encap_type,
 596                            struct nlattr *encap,
 597                            const struct fib_nh *nh,
 598                            const struct fib_config *cfg)
 599 {
 600         struct lwtunnel_state *lwtstate;
 601         int ret, result = 0;
 602
 603         if (encap_type == LWTUNNEL_ENCAP_NONE)
 604                 return 0;
 605
 606         ret = lwtunnel_build_state(encap_type, encap,
 607                                    AF_INET, cfg, &lwtstate);
 608         if (!ret) {
 609                 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
 610                 lwtstate_free(lwtstate);
 611         }
 612
 613         return result;
 614 }
 615
 616 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 617 {
 618 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 619         struct rtnexthop *rtnh;
 620         int remaining;
 621 #endif
 622
 623         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 624                 return 1;
 625
 626         if (cfg->fc_oif || cfg->fc_gw) {
 627                 if (cfg->fc_encap) {
 628                         if (fib_encap_match(cfg->fc_encap_type,
 629                                             cfg->fc_encap, fi->fib_nh, cfg))
 630                             return 1;
 631                 }
 632                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 633                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 634                         return 0;
 635                 return 1;
 636         }
 637
 638 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 639         if (!cfg->fc_mp)
 640                 return 0;
 641
 642         rtnh = cfg->fc_mp;
 643         remaining = cfg->fc_mp_len;
 644
 645         for_nexthops(fi) {
 646                 int attrlen;
 647
 648                 if (!rtnh_ok(rtnh, remaining))
 649                         return -EINVAL;
 650
 651                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 652                         return 1;
 653
 654                 attrlen = rtnh_attrlen(rtnh);
 655                 if (attrlen > 0) {
 656                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 657
 658                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 659                         if (nla && nla_get_in_addr(nla) != nh->nh_gw)
 660                                 return 1;
 661 #ifdef CONFIG_IP_ROUTE_CLASSID
 662                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 663                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 664                                 return 1;
 665 #endif
 666                 }
 667
 668                 rtnh = rtnh_next(rtnh, &remaining);
 669         } endfor_nexthops(fi);
 670 #endif
 671         return 0;
 672 }
 673
 674
 675 /*
 676  * Picture
 677  * -------
 678  *
 679  * Semantics of nexthop is very messy by historical reasons.
 680  * We have to take into account, that:
 681  * a) gateway can be actually local interface address,
 682  *    so that gatewayed route is direct.
 683  * b) gateway must be on-link address, possibly
 684  *    described not by an ifaddr, but also by a direct route.
 685  * c) If both gateway and interface are specified, they should not
 686  *    contradict.
 687  * d) If we use tunnel routes, gateway could be not on-link.
 688  *
 689  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 690  * results in pretty ugly and hairy code with obscure logic.
 691  *
 692  * I chose to generalized it instead, so that the size
 693  * of code does not increase practically, but it becomes
 694  * much more general.
 695  * Every prefix is assigned a "scope" value: "host" is local address,
 696  * "link" is direct route,
 697  * [ ... "site" ... "interior" ... ]
 698  * and "universe" is true gateway route with global meaning.
 699  *
 700  * Every prefix refers to a set of "nexthop"s (gw, oif),
 701  * where gw must have narrower scope. This recursion stops
 702  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 703  * which means that gw is forced to be on link.
 704  *
 705  * Code is still hairy, but now it is apparently logically
 706  * consistent and very flexible. F.e. as by-product it allows
 707  * to co-exists in peace independent exterior and interior
 708  * routing processes.
 709  *
 710  * Normally it looks as following.
 711  *
 712  * {universe prefix}  -> (gw, oif) [scope link]
 713  *                |
 714  *                |-> {link prefix} -> (gw, oif) [scope local]
 715  *                                      |
 716  *                                      |-> {local prefix} (terminal node)
 717  */
 718 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 719                         struct fib_nh *nh, struct netlink_ext_ack *extack)
 720 {
 721         int err = 0;
 722         struct net *net;
 723         struct net_device *dev;
 724
 725         net = cfg->fc_nlinfo.nl_net;
 726         if (nh->nh_gw) {
 727                 struct fib_result res;
 728
 729                 if (nh->nh_flags & RTNH_F_ONLINK) {
 730                         unsigned int addr_type;
 731
 732                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 733                                 return -EINVAL;
 734                         dev = __dev_get_by_index(net, nh->nh_oif);
 735                         if (!dev)
 736                                 return -ENODEV;
 737                         if (!(dev->flags & IFF_UP))
 738                                 return -ENETDOWN;
 739                         addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
 740                         if (addr_type != RTN_UNICAST)
 741                                 return -EINVAL;
 742                         if (!netif_carrier_ok(dev))
 743                                 nh->nh_flags |= RTNH_F_LINKDOWN;
 744                         nh->nh_dev = dev;
 745                         dev_hold(dev);
 746                         nh->nh_scope = RT_SCOPE_LINK;
 747                         return 0;
 748                 }
 749                 rcu_read_lock();
 750                 {
 751                         struct fib_table *tbl = NULL;
 752                         struct flowi4 fl4 = {
 753                                 .daddr = nh->nh_gw,
 754                                 .flowi4_scope = cfg->fc_scope + 1,
 755                                 .flowi4_oif = nh->nh_oif,
 756                                 .flowi4_iif = LOOPBACK_IFINDEX,
 757                         };
 758
 759                         /* It is not necessary, but requires a bit of thinking */
 760                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 761                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 762
 763                         if (cfg->fc_table)
 764                                 tbl = fib_get_table(net, cfg->fc_table);
 765
 766                         if (tbl)
 767                                 err = fib_table_lookup(tbl, &fl4, &res,
 768                                                        FIB_LOOKUP_IGNORE_LINKSTATE |
 769                                                        FIB_LOOKUP_NOREF);
 770
 771                         /* on error or if no table given do full lookup. This
 772                          * is needed for example when nexthops are in the local
 773                          * table rather than the given table
 774                          */
 775                         if (!tbl || err) {
 776                                 err = fib_lookup(net, &fl4, &res,
 777                                                  FIB_LOOKUP_IGNORE_LINKSTATE);
 778                         }
 779
 780                         if (err) {
 781                                 rcu_read_unlock();
 782                                 return err;
 783                         }
 784                 }
 785                 err = -EINVAL;
 786                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 787                         goto out;
 788                 nh->nh_scope = res.scope;
 789                 nh->nh_oif = FIB_RES_OIF(res);
 790                 nh->nh_dev = dev = FIB_RES_DEV(res);
 791                 if (!dev)
 792                         goto out;
 793                 dev_hold(dev);
 794                 if (!netif_carrier_ok(dev))
 795                         nh->nh_flags |= RTNH_F_LINKDOWN;
 796                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 797         } else {
 798                 struct in_device *in_dev;
 799
 800                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 801                         return -EINVAL;
 802                 rcu_read_lock();
 803                 err = -ENODEV;
 804                 in_dev = inetdev_by_index(net, nh->nh_oif);
 805                 if (!in_dev)
 806                         goto out;
 807                 err = -ENETDOWN;
 808                 if (!(in_dev->dev->flags & IFF_UP))
 809                         goto out;
 810                 nh->nh_dev = in_dev->dev;
 811                 dev_hold(nh->nh_dev);
 812                 nh->nh_scope = RT_SCOPE_HOST;
 813                 if (!netif_carrier_ok(nh->nh_dev))
 814                         nh->nh_flags |= RTNH_F_LINKDOWN;
 815                 err = 0;
 816         }
 817 out:
 818         rcu_read_unlock();
 819         return err;
 820 }
 821
 822 static inline unsigned int fib_laddr_hashfn(__be32 val)
 823 {
 824         unsigned int mask = (fib_info_hash_size - 1);
 825
 826         return ((__force u32)val ^
 827                 ((__force u32)val >> 7) ^
 828                 ((__force u32)val >> 14)) & mask;
 829 }
 830
 831 static struct hlist_head *fib_info_hash_alloc(int bytes)
 832 {
 833         if (bytes <= PAGE_SIZE)
 834                 return kzalloc(bytes, GFP_KERNEL);
 835         else
 836                 return (struct hlist_head *)
 837                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 838                                          get_order(bytes));
 839 }
 840
 841 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 842 {
 843         if (!hash)
 844                 return;
 845
 846         if (bytes <= PAGE_SIZE)
 847                 kfree(hash);
 848         else
 849                 free_pages((unsigned long) hash, get_order(bytes));
 850 }
 851
 852 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 853                                struct hlist_head *new_laddrhash,
 854                                unsigned int new_size)
 855 {
 856         struct hlist_head *old_info_hash, *old_laddrhash;
 857         unsigned int old_size = fib_info_hash_size;
 858         unsigned int i, bytes;
 859
 860         spin_lock_bh(&fib_info_lock);
 861         old_info_hash = fib_info_hash;
 862         old_laddrhash = fib_info_laddrhash;
 863         fib_info_hash_size = new_size;
 864
 865         for (i = 0; i < old_size; i++) {
 866                 struct hlist_head *head = &fib_info_hash[i];
 867                 struct hlist_node *n;
 868                 struct fib_info *fi;
 869
 870                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 871                         struct hlist_head *dest;
 872                         unsigned int new_hash;
 873
 874                         new_hash = fib_info_hashfn(fi);
 875                         dest = &new_info_hash[new_hash];
 876                         hlist_add_head(&fi->fib_hash, dest);
 877                 }
 878         }
 879         fib_info_hash = new_info_hash;
 880
 881         for (i = 0; i < old_size; i++) {
 882                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 883                 struct hlist_node *n;
 884                 struct fib_info *fi;
 885
 886                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 887                         struct hlist_head *ldest;
 888                         unsigned int new_hash;
 889
 890                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 891                         ldest = &new_laddrhash[new_hash];
 892                         hlist_add_head(&fi->fib_lhash, ldest);
 893                 }
 894         }
 895         fib_info_laddrhash = new_laddrhash;
 896
 897         spin_unlock_bh(&fib_info_lock);
 898
 899         bytes = old_size * sizeof(struct hlist_head *);
 900         fib_info_hash_free(old_info_hash, bytes);
 901         fib_info_hash_free(old_laddrhash, bytes);
 902 }
 903
 904 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 905 {
 906         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 907                                         nh->nh_gw,
 908                                         nh->nh_parent->fib_scope);
 909         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 910
 911         return nh->nh_saddr;
 912 }
 913
 914 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 915 {
 916         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 917             fib_prefsrc != cfg->fc_dst) {
 918                 u32 tb_id = cfg->fc_table;
 919                 int rc;
 920
 921                 if (tb_id == RT_TABLE_MAIN)
 922                         tb_id = RT_TABLE_LOCAL;
 923
 924                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
 925                                           fib_prefsrc, tb_id);
 926
 927                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
 928                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
 929                                                   fib_prefsrc, RT_TABLE_LOCAL);
 930                 }
 931
 932                 if (rc != RTN_LOCAL)
 933                         return false;
 934         }
 935         return true;
 936 }
 937
 938 static int
 939 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
 940 {
 941         bool ecn_ca = false;
 942         struct nlattr *nla;
 943         int remaining;
 944
 945         if (!cfg->fc_mx)
 946                 return 0;
 947
 948         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 949                 int type = nla_type(nla);
 950                 u32 val;
 951
 952                 if (!type)
 953                         continue;
 954                 if (type > RTAX_MAX)
 955                         return -EINVAL;
 956
 957                 if (type == RTAX_CC_ALGO) {
 958                         char tmp[TCP_CA_NAME_MAX];
 959
 960                         nla_strlcpy(tmp, nla, sizeof(tmp));
 961                         val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
 962                         if (val == TCP_CA_UNSPEC)
 963                                 return -EINVAL;
 964                 } else {
 965                         val = nla_get_u32(nla);
 966                 }
 967                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 968                         val = 65535 - 40;
 969                 if (type == RTAX_MTU && val > 65535 - 15)
 970                         val = 65535 - 15;
 971                 if (type == RTAX_HOPLIMIT && val > 255)
 972                         val = 255;
 973                 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
 974                         return -EINVAL;
 975                 fi->fib_metrics[type - 1] = val;
 976         }
 977
 978         if (ecn_ca)
 979                 fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
 980
 981         return 0;
 982 }
 983
 984 struct fib_info *fib_create_info(struct fib_config *cfg,
 985                                  struct netlink_ext_ack *extack)
 986 {
 987         int err;
 988         struct fib_info *fi = NULL;
 989         struct fib_info *ofi;
 990         int nhs = 1;
 991         struct net *net = cfg->fc_nlinfo.nl_net;
 992
 993         if (cfg->fc_type > RTN_MAX)
 994                 goto err_inval;
 995
 996         /* Fast check to catch the most weird cases */
 997         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 998                 goto err_inval;
 999
1000         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
1001                 goto err_inval;
1002
1003 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1004         if (cfg->fc_mp) {
1005                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1006                 if (nhs == 0)
1007                         goto err_inval;
1008         }
1009 #endif
1010
1011         err = -ENOBUFS;
1012         if (fib_info_cnt >= fib_info_hash_size) {
1013                 unsigned int new_size = fib_info_hash_size << 1;
1014                 struct hlist_head *new_info_hash;
1015                 struct hlist_head *new_laddrhash;
1016                 unsigned int bytes;
1017
1018                 if (!new_size)
1019                         new_size = 16;
1020                 bytes = new_size * sizeof(struct hlist_head *);
1021                 new_info_hash = fib_info_hash_alloc(bytes);
1022                 new_laddrhash = fib_info_hash_alloc(bytes);
1023                 if (!new_info_hash || !new_laddrhash) {
1024                         fib_info_hash_free(new_info_hash, bytes);
1025                         fib_info_hash_free(new_laddrhash, bytes);
1026                 } else
1027                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1028
1029                 if (!fib_info_hash_size)
1030                         goto failure;
1031         }
1032
1033         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1034         if (!fi)
1035                 goto failure;
1036         fib_info_cnt++;
1037         if (cfg->fc_mx) {
1038                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1039                 if (!fi->fib_metrics)
1040                         goto failure;
1041         } else
1042                 fi->fib_metrics = (u32 *) dst_default_metrics;
1043
1044         fi->fib_net = net;
1045         fi->fib_protocol = cfg->fc_protocol;
1046         fi->fib_scope = cfg->fc_scope;
1047         fi->fib_flags = cfg->fc_flags;
1048         fi->fib_priority = cfg->fc_priority;
1049         fi->fib_prefsrc = cfg->fc_prefsrc;
1050         fi->fib_type = cfg->fc_type;
1051         fi->fib_tb_id = cfg->fc_table;
1052
1053         fi->fib_nhs = nhs;
1054         change_nexthops(fi) {
1055                 nexthop_nh->nh_parent = fi;
1056                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1057                 if (!nexthop_nh->nh_pcpu_rth_output)
1058                         goto failure;
1059         } endfor_nexthops(fi)
1060
1061         err = fib_convert_metrics(fi, cfg);
1062         if (err)
1063                 goto failure;
1064
1065         if (cfg->fc_mp) {
1066 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1067                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1068                 if (err != 0)
1069                         goto failure;
1070                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1071                         goto err_inval;
1072                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1073                         goto err_inval;
1074 #ifdef CONFIG_IP_ROUTE_CLASSID
1075                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1076                         goto err_inval;
1077 #endif
1078 #else
1079                 goto err_inval;
1080 #endif
1081         } else {
1082                 struct fib_nh *nh = fi->fib_nh;
1083
1084                 if (cfg->fc_encap) {
1085                         struct lwtunnel_state *lwtstate;
1086
1087                         if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1088                                 goto err_inval;
1089                         err = lwtunnel_build_state(cfg->fc_encap_type,
1090                                                    cfg->fc_encap, AF_INET, cfg,
1091                                                    &lwtstate);
1092                         if (err)
1093                                 goto failure;
1094
1095                         nh->nh_lwtstate = lwtstate_get(lwtstate);
1096                 }
1097                 nh->nh_oif = cfg->fc_oif;
1098                 nh->nh_gw = cfg->fc_gw;
1099                 nh->nh_flags = cfg->fc_flags;
1100 #ifdef CONFIG_IP_ROUTE_CLASSID
1101                 nh->nh_tclassid = cfg->fc_flow;
1102                 if (nh->nh_tclassid)
1103                         fi->fib_net->ipv4.fib_num_tclassid_users++;
1104 #endif
1105 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1106                 nh->nh_weight = 1;
1107 #endif
1108         }
1109
1110         if (fib_props[cfg->fc_type].error) {
1111                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1112                         goto err_inval;
1113                 goto link_it;
1114         } else {
1115                 switch (cfg->fc_type) {
1116                 case RTN_UNICAST:
1117                 case RTN_LOCAL:
1118                 case RTN_BROADCAST:
1119                 case RTN_ANYCAST:
1120                 case RTN_MULTICAST:
1121                         break;
1122                 default:
1123                         goto err_inval;
1124                 }
1125         }
1126
1127         if (cfg->fc_scope > RT_SCOPE_HOST)
1128                 goto err_inval;
1129
1130         if (cfg->fc_scope == RT_SCOPE_HOST) {
1131                 struct fib_nh *nh = fi->fib_nh;
1132
1133                 /* Local address is added. */
1134                 if (nhs != 1)
1135                         goto err_inval;
1136                 if (nh->nh_gw)
1137                         goto err_inval;
1138                 nh->nh_scope = RT_SCOPE_NOWHERE;
1139                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1140                 err = -ENODEV;
1141                 if (!nh->nh_dev)
1142                         goto failure;
1143         } else {
1144                 int linkdown = 0;
1145
1146                 change_nexthops(fi) {
1147                         err = fib_check_nh(cfg, fi, nexthop_nh, extack);
1148                         if (err != 0)
1149                                 goto failure;
1150                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1151                                 linkdown++;
1152                 } endfor_nexthops(fi)
1153                 if (linkdown == fi->fib_nhs)
1154                         fi->fib_flags |= RTNH_F_LINKDOWN;
1155         }
1156
1157         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1158                 goto err_inval;
1159
1160         change_nexthops(fi) {
1161                 fib_info_update_nh_saddr(net, nexthop_nh);
1162                 fib_add_weight(fi, nexthop_nh);
1163         } endfor_nexthops(fi)
1164
1165         fib_rebalance(fi);
1166
1167 link_it:
1168         ofi = fib_find_info(fi);
1169         if (ofi) {
1170                 fi->fib_dead = 1;
1171                 free_fib_info(fi);
1172                 ofi->fib_treeref++;
1173                 return ofi;
1174         }
1175
1176         fi->fib_treeref++;
1177         atomic_inc(&fi->fib_clntref);
1178         spin_lock_bh(&fib_info_lock);
1179         hlist_add_head(&fi->fib_hash,
1180                        &fib_info_hash[fib_info_hashfn(fi)]);
1181         if (fi->fib_prefsrc) {
1182                 struct hlist_head *head;
1183
1184                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1185                 hlist_add_head(&fi->fib_lhash, head);
1186         }
1187         change_nexthops(fi) {
1188                 struct hlist_head *head;
1189                 unsigned int hash;
1190
1191                 if (!nexthop_nh->nh_dev)
1192                         continue;
1193                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1194                 head = &fib_info_devhash[hash];
1195                 hlist_add_head(&nexthop_nh->nh_hash, head);
1196         } endfor_nexthops(fi)
1197         spin_unlock_bh(&fib_info_lock);
1198         return fi;
1199
1200 err_inval:
1201         err = -EINVAL;
1202
1203 failure:
1204         if (fi) {
1205                 fi->fib_dead = 1;
1206                 free_fib_info(fi);
1207         }
1208
1209         return ERR_PTR(err);
1210 }
1211
1212 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1213                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1214                   struct fib_info *fi, unsigned int flags)
1215 {
1216         struct nlmsghdr *nlh;
1217         struct rtmsg *rtm;
1218
1219         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1220         if (!nlh)
1221                 return -EMSGSIZE;
1222
1223         rtm = nlmsg_data(nlh);
1224         rtm->rtm_family = AF_INET;
1225         rtm->rtm_dst_len = dst_len;
1226         rtm->rtm_src_len = 0;
1227         rtm->rtm_tos = tos;
1228         if (tb_id < 256)
1229                 rtm->rtm_table = tb_id;
1230         else
1231                 rtm->rtm_table = RT_TABLE_COMPAT;
1232         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1233                 goto nla_put_failure;
1234         rtm->rtm_type = type;
1235         rtm->rtm_flags = fi->fib_flags;
1236         rtm->rtm_scope = fi->fib_scope;
1237         rtm->rtm_protocol = fi->fib_protocol;
1238
1239         if (rtm->rtm_dst_len &&
1240             nla_put_in_addr(skb, RTA_DST, dst))
1241                 goto nla_put_failure;
1242         if (fi->fib_priority &&
1243             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1244                 goto nla_put_failure;
1245         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1246                 goto nla_put_failure;
1247
1248         if (fi->fib_prefsrc &&
1249             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1250                 goto nla_put_failure;
1251         if (fi->fib_nhs == 1) {
1252                 struct in_device *in_dev;
1253
1254                 if (fi->fib_nh->nh_gw &&
1255                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1256                         goto nla_put_failure;
1257                 if (fi->fib_nh->nh_oif &&
1258                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1259                         goto nla_put_failure;
1260                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1261                         in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1262                         if (in_dev &&
1263                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1264                                 rtm->rtm_flags |= RTNH_F_DEAD;
1265                 }
1266 #ifdef CONFIG_IP_ROUTE_CLASSID
1267                 if (fi->fib_nh[0].nh_tclassid &&
1268                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1269                         goto nla_put_failure;
1270 #endif
1271                 if (fi->fib_nh->nh_lwtstate &&
1272                     lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1273                         goto nla_put_failure;
1274         }
1275 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1276         if (fi->fib_nhs > 1) {
1277                 struct rtnexthop *rtnh;
1278                 struct nlattr *mp;
1279
1280                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1281                 if (!mp)
1282                         goto nla_put_failure;
1283
1284                 for_nexthops(fi) {
1285                         struct in_device *in_dev;
1286
1287                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1288                         if (!rtnh)
1289                                 goto nla_put_failure;
1290
1291                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1292                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1293                                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1294                                 if (in_dev &&
1295                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1296                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1297                         }
1298                         rtnh->rtnh_hops = nh->nh_weight - 1;
1299                         rtnh->rtnh_ifindex = nh->nh_oif;
1300
1301                         if (nh->nh_gw &&
1302                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1303                                 goto nla_put_failure;
1304 #ifdef CONFIG_IP_ROUTE_CLASSID
1305                         if (nh->nh_tclassid &&
1306                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1307                                 goto nla_put_failure;
1308 #endif
1309                         if (nh->nh_lwtstate &&
1310                             lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1311                                 goto nla_put_failure;
1312
1313                         /* length of rtnetlink header + attributes */
1314                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1315                 } endfor_nexthops(fi);
1316
1317                 nla_nest_end(skb, mp);
1318         }
1319 #endif
1320         nlmsg_end(skb, nlh);
1321         return 0;
1322
1323 nla_put_failure:
1324         nlmsg_cancel(skb, nlh);
1325         return -EMSGSIZE;
1326 }
1327
1328 /*
1329  * Update FIB if:
1330  * - local address disappeared -> we must delete all the entries
1331  *   referring to it.
1332  * - device went down -> we must shutdown all nexthops going via it.
1333  */
1334 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1335 {
1336         int ret = 0;
1337         unsigned int hash = fib_laddr_hashfn(local);
1338         struct hlist_head *head = &fib_info_laddrhash[hash];
1339         struct net *net = dev_net(dev);
1340         int tb_id = l3mdev_fib_table(dev);
1341         struct fib_info *fi;
1342
1343         if (!fib_info_laddrhash || local == 0)
1344                 return 0;
1345
1346         hlist_for_each_entry(fi, head, fib_lhash) {
1347                 if (!net_eq(fi->fib_net, net) ||
1348                     fi->fib_tb_id != tb_id)
1349                         continue;
1350                 if (fi->fib_prefsrc == local) {
1351                         fi->fib_flags |= RTNH_F_DEAD;
1352                         ret++;
1353                 }
1354         }
1355         return ret;
1356 }
1357
1358 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1359                                  enum fib_event_type event_type)
1360 {
1361         struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1362         struct fib_nh_notifier_info info = {
1363                 .fib_nh = fib_nh,
1364         };
1365
1366         switch (event_type) {
1367         case FIB_EVENT_NH_ADD:
1368                 if (fib_nh->nh_flags & RTNH_F_DEAD)
1369                         break;
1370                 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1371                     fib_nh->nh_flags & RTNH_F_LINKDOWN)
1372                         break;
1373                 return call_fib_notifiers(dev_net(fib_nh->nh_dev), event_type,
1374                                           &info.info);
1375         case FIB_EVENT_NH_DEL:
1376                 if ((IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1377                      fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1378                     (fib_nh->nh_flags & RTNH_F_DEAD))
1379                         return call_fib_notifiers(dev_net(fib_nh->nh_dev),
1380                                                   event_type, &info.info);
1381         default:
1382                 break;
1383         }
1384
1385         return NOTIFY_DONE;
1386 }
1387
1388 /* Event              force Flags           Description
1389  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1390  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1391  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1392  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1393  */
1394 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1395 {
1396         int ret = 0;
1397         int scope = RT_SCOPE_NOWHERE;
1398         struct fib_info *prev_fi = NULL;
1399         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1400         struct hlist_head *head = &fib_info_devhash[hash];
1401         struct fib_nh *nh;
1402
1403         if (force)
1404                 scope = -1;
1405
1406         hlist_for_each_entry(nh, head, nh_hash) {
1407                 struct fib_info *fi = nh->nh_parent;
1408                 int dead;
1409
1410                 BUG_ON(!fi->fib_nhs);
1411                 if (nh->nh_dev != dev || fi == prev_fi)
1412                         continue;
1413                 prev_fi = fi;
1414                 dead = 0;
1415                 change_nexthops(fi) {
1416                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1417                                 dead++;
1418                         else if (nexthop_nh->nh_dev == dev &&
1419                                  nexthop_nh->nh_scope != scope) {
1420                                 switch (event) {
1421                                 case NETDEV_DOWN:
1422                                 case NETDEV_UNREGISTER:
1423                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1424                                         /* fall through */
1425                                 case NETDEV_CHANGE:
1426                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1427                                         break;
1428                                 }
1429                                 call_fib_nh_notifiers(nexthop_nh,
1430                                                       FIB_EVENT_NH_DEL);
1431                                 dead++;
1432                         }
1433 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1434                         if (event == NETDEV_UNREGISTER &&
1435                             nexthop_nh->nh_dev == dev) {
1436                                 dead = fi->fib_nhs;
1437                                 break;
1438                         }
1439 #endif
1440                 } endfor_nexthops(fi)
1441                 if (dead == fi->fib_nhs) {
1442                         switch (event) {
1443                         case NETDEV_DOWN:
1444                         case NETDEV_UNREGISTER:
1445                                 fi->fib_flags |= RTNH_F_DEAD;
1446                                 /* fall through */
1447                         case NETDEV_CHANGE:
1448                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1449                                 break;
1450                         }
1451                         ret++;
1452                 }
1453
1454                 fib_rebalance(fi);
1455         }
1456
1457         return ret;
1458 }
1459
1460 /* Must be invoked inside of an RCU protected region.  */
1461 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1462 {
1463         struct fib_info *fi = NULL, *last_resort = NULL;
1464         struct hlist_head *fa_head = res->fa_head;
1465         struct fib_table *tb = res->table;
1466         u8 slen = 32 - res->prefixlen;
1467         int order = -1, last_idx = -1;
1468         struct fib_alias *fa, *fa1 = NULL;
1469         u32 last_prio = res->fi->fib_priority;
1470         u8 last_tos = 0;
1471
1472         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1473                 struct fib_info *next_fi = fa->fa_info;
1474
1475                 if (fa->fa_slen != slen)
1476                         continue;
1477                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1478                         continue;
1479                 if (fa->tb_id != tb->tb_id)
1480                         continue;
1481                 if (next_fi->fib_priority > last_prio &&
1482                     fa->fa_tos == last_tos) {
1483                         if (last_tos)
1484                                 continue;
1485                         break;
1486                 }
1487                 if (next_fi->fib_flags & RTNH_F_DEAD)
1488                         continue;
1489                 last_tos = fa->fa_tos;
1490                 last_prio = next_fi->fib_priority;
1491
1492                 if (next_fi->fib_scope != res->scope ||
1493                     fa->fa_type != RTN_UNICAST)
1494                         continue;
1495                 if (!next_fi->fib_nh[0].nh_gw ||
1496                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1497                         continue;
1498
1499                 fib_alias_accessed(fa);
1500
1501                 if (!fi) {
1502                         if (next_fi != res->fi)
1503                                 break;
1504                         fa1 = fa;
1505                 } else if (!fib_detect_death(fi, order, &last_resort,
1506                                              &last_idx, fa1->fa_default)) {
1507                         fib_result_assign(res, fi);
1508                         fa1->fa_default = order;
1509                         goto out;
1510                 }
1511                 fi = next_fi;
1512                 order++;
1513         }
1514
1515         if (order <= 0 || !fi) {
1516                 if (fa1)
1517                         fa1->fa_default = -1;
1518                 goto out;
1519         }
1520
1521         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1522                               fa1->fa_default)) {
1523                 fib_result_assign(res, fi);
1524                 fa1->fa_default = order;
1525                 goto out;
1526         }
1527
1528         if (last_idx >= 0)
1529                 fib_result_assign(res, last_resort);
1530         fa1->fa_default = last_idx;
1531 out:
1532         return;
1533 }
1534
1535 /*
1536  * Dead device goes up. We wake up dead nexthops.
1537  * It takes sense only on multipath routes.
1538  */
1539 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1540 {
1541         struct fib_info *prev_fi;
1542         unsigned int hash;
1543         struct hlist_head *head;
1544         struct fib_nh *nh;
1545         int ret;
1546
1547         if (!(dev->flags & IFF_UP))
1548                 return 0;
1549
1550         if (nh_flags & RTNH_F_DEAD) {
1551                 unsigned int flags = dev_get_flags(dev);
1552
1553                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1554                         nh_flags |= RTNH_F_LINKDOWN;
1555         }
1556
1557         prev_fi = NULL;
1558         hash = fib_devindex_hashfn(dev->ifindex);
1559         head = &fib_info_devhash[hash];
1560         ret = 0;
1561
1562         hlist_for_each_entry(nh, head, nh_hash) {
1563                 struct fib_info *fi = nh->nh_parent;
1564                 int alive;
1565
1566                 BUG_ON(!fi->fib_nhs);
1567                 if (nh->nh_dev != dev || fi == prev_fi)
1568                         continue;
1569
1570                 prev_fi = fi;
1571                 alive = 0;
1572                 change_nexthops(fi) {
1573                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1574                                 alive++;
1575                                 continue;
1576                         }
1577                         if (!nexthop_nh->nh_dev ||
1578                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1579                                 continue;
1580                         if (nexthop_nh->nh_dev != dev ||
1581                             !__in_dev_get_rtnl(dev))
1582                                 continue;
1583                         alive++;
1584                         nexthop_nh->nh_flags &= ~nh_flags;
1585                         call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1586                 } endfor_nexthops(fi)
1587
1588                 if (alive > 0) {
1589                         fi->fib_flags &= ~nh_flags;
1590                         ret++;
1591                 }
1592
1593                 fib_rebalance(fi);
1594         }
1595
1596         return ret;
1597 }
1598
1599 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1600 static bool fib_good_nh(const struct fib_nh *nh)
1601 {
1602         int state = NUD_REACHABLE;
1603
1604         if (nh->nh_scope == RT_SCOPE_LINK) {
1605                 struct neighbour *n;
1606
1607                 rcu_read_lock_bh();
1608
1609                 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1610                                               (__force u32)nh->nh_gw);
1611                 if (n)
1612                         state = n->nud_state;
1613
1614                 rcu_read_unlock_bh();
1615         }
1616
1617         return !!(state & NUD_VALID);
1618 }
1619
1620 void fib_select_multipath(struct fib_result *res, int hash)
1621 {
1622         struct fib_info *fi = res->fi;
1623         struct net *net = fi->fib_net;
1624         bool first = false;
1625
1626         for_nexthops(fi) {
1627                 if (hash > atomic_read(&nh->nh_upper_bound))
1628                         continue;
1629
1630                 if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1631                     fib_good_nh(nh)) {
1632                         res->nh_sel = nhsel;
1633                         return;
1634                 }
1635                 if (!first) {
1636                         res->nh_sel = nhsel;
1637                         first = true;
1638                 }
1639         } endfor_nexthops(fi);
1640 }
1641 #endif
1642
1643 void fib_select_path(struct net *net, struct fib_result *res,
1644                      struct flowi4 *fl4, const struct sk_buff *skb)
1645 {
1646         bool oif_check;
1647
1648         oif_check = (fl4->flowi4_oif == 0 ||
1649                      fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1650
1651 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1652         if (res->fi->fib_nhs > 1 && oif_check) {
1653                 int h = fib_multipath_hash(res->fi, fl4, skb);
1654
1655                 fib_select_multipath(res, h);
1656         }
1657         else
1658 #endif
1659         if (!res->prefixlen &&
1660             res->table->tb_num_default > 1 &&
1661             res->type == RTN_UNICAST && oif_check)
1662                 fib_select_default(fl4, res);
1663
1664         if (!fl4->saddr)
1665                 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1666 }
1667 EXPORT_SYMBOL_GPL(fib_select_path);