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/netlink.h>
  45 #include <net/nexthop.h>
  46 #include <net/lwtunnel.h>
  47 #include <net/fib_notifier.h>
  48
  49 #include "fib_lookup.h"
  50
  51 static DEFINE_SPINLOCK(fib_info_lock);
  52 static struct hlist_head *fib_info_hash;
  53 static struct hlist_head *fib_info_laddrhash;
  54 static unsigned int fib_info_hash_size;
  55 static unsigned int fib_info_cnt;
  56
  57 #define DEVINDEX_HASHBITS 8
  58 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  59 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  60
  61 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  62
  63 #define for_nexthops(fi) {                                              \
  64         int nhsel; const struct fib_nh *nh;                             \
  65         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  66              nhsel < (fi)->fib_nhs;                                     \
  67              nh++, nhsel++)
  68
  69 #define change_nexthops(fi) {                                           \
  70         int nhsel; struct fib_nh *nexthop_nh;                           \
  71         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  72              nhsel < (fi)->fib_nhs;                                     \
  73              nexthop_nh++, nhsel++)
  74
  75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  76
  77 /* Hope, that gcc will optimize it to get rid of dummy loop */
  78
  79 #define for_nexthops(fi) {                                              \
  80         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  81         for (nhsel = 0; nhsel < 1; nhsel++)
  82
  83 #define change_nexthops(fi) {                                           \
  84         int nhsel;                                                      \
  85         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  86         for (nhsel = 0; nhsel < 1; nhsel++)
  87
  88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  89
  90 #define endfor_nexthops(fi) }
  91
  92
  93 const struct fib_prop fib_props[RTN_MAX + 1] = {
  94         [RTN_UNSPEC] = {
  95                 .error  = 0,
  96                 .scope  = RT_SCOPE_NOWHERE,
  97         },
  98         [RTN_UNICAST] = {
  99                 .error  = 0,
 100                 .scope  = RT_SCOPE_UNIVERSE,
 101         },
 102         [RTN_LOCAL] = {
 103                 .error  = 0,
 104                 .scope  = RT_SCOPE_HOST,
 105         },
 106         [RTN_BROADCAST] = {
 107                 .error  = 0,
 108                 .scope  = RT_SCOPE_LINK,
 109         },
 110         [RTN_ANYCAST] = {
 111                 .error  = 0,
 112                 .scope  = RT_SCOPE_LINK,
 113         },
 114         [RTN_MULTICAST] = {
 115                 .error  = 0,
 116                 .scope  = RT_SCOPE_UNIVERSE,
 117         },
 118         [RTN_BLACKHOLE] = {
 119                 .error  = -EINVAL,
 120                 .scope  = RT_SCOPE_UNIVERSE,
 121         },
 122         [RTN_UNREACHABLE] = {
 123                 .error  = -EHOSTUNREACH,
 124                 .scope  = RT_SCOPE_UNIVERSE,
 125         },
 126         [RTN_PROHIBIT] = {
 127                 .error  = -EACCES,
 128                 .scope  = RT_SCOPE_UNIVERSE,
 129         },
 130         [RTN_THROW] = {
 131                 .error  = -EAGAIN,
 132                 .scope  = RT_SCOPE_UNIVERSE,
 133         },
 134         [RTN_NAT] = {
 135                 .error  = -EINVAL,
 136                 .scope  = RT_SCOPE_NOWHERE,
 137         },
 138         [RTN_XRESOLVE] = {
 139                 .error  = -EINVAL,
 140                 .scope  = RT_SCOPE_NOWHERE,
 141         },
 142 };
 143
 144 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 145 {
 146         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 147
 148         if (!rt)
 149                 return;
 150
 151         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 152          * because we waited an RCU grace period before calling
 153          * free_fib_info_rcu()
 154          */
 155
 156         dst_dev_put(&rt->dst);
 157         dst_release_immediate(&rt->dst);
 158 }
 159
 160 static void free_nh_exceptions(struct fib_nh *nh)
 161 {
 162         struct fnhe_hash_bucket *hash;
 163         int i;
 164
 165         hash = rcu_dereference_protected(nh->nh_exceptions, 1);
 166         if (!hash)
 167                 return;
 168         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 169                 struct fib_nh_exception *fnhe;
 170
 171                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 172                 while (fnhe) {
 173                         struct fib_nh_exception *next;
 174
 175                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 176
 177                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 178                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 179
 180                         kfree(fnhe);
 181
 182                         fnhe = next;
 183                 }
 184         }
 185         kfree(hash);
 186 }
 187
 188 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 189 {
 190         int cpu;
 191
 192         if (!rtp)
 193                 return;
 194
 195         for_each_possible_cpu(cpu) {
 196                 struct rtable *rt;
 197
 198                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 199                 if (rt) {
 200                         dst_dev_put(&rt->dst);
 201                         dst_release_immediate(&rt->dst);
 202                 }
 203         }
 204         free_percpu(rtp);
 205 }
 206
 207 /* Release a nexthop info record */
 208 static void free_fib_info_rcu(struct rcu_head *head)
 209 {
 210         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 211         struct dst_metrics *m;
 212
 213         change_nexthops(fi) {
 214                 if (nexthop_nh->nh_dev)
 215                         dev_put(nexthop_nh->nh_dev);
 216                 lwtstate_put(nexthop_nh->nh_lwtstate);
 217                 free_nh_exceptions(nexthop_nh);
 218                 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 219                 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
 220         } endfor_nexthops(fi);
 221
 222         m = fi->fib_metrics;
 223         if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
 224                 kfree(m);
 225         kfree(fi);
 226 }
 227
 228 void free_fib_info(struct fib_info *fi)
 229 {
 230         if (fi->fib_dead == 0) {
 231                 pr_warn("Freeing alive fib_info %p\n", fi);
 232                 return;
 233         }
 234         fib_info_cnt--;
 235 #ifdef CONFIG_IP_ROUTE_CLASSID
 236         change_nexthops(fi) {
 237                 if (nexthop_nh->nh_tclassid)
 238                         fi->fib_net->ipv4.fib_num_tclassid_users--;
 239         } endfor_nexthops(fi);
 240 #endif
 241         call_rcu(&fi->rcu, free_fib_info_rcu);
 242 }
 243 EXPORT_SYMBOL_GPL(free_fib_info);
 244
 245 void fib_release_info(struct fib_info *fi)
 246 {
 247         spin_lock_bh(&fib_info_lock);
 248         if (fi && --fi->fib_treeref == 0) {
 249                 hlist_del(&fi->fib_hash);
 250                 if (fi->fib_prefsrc)
 251                         hlist_del(&fi->fib_lhash);
 252                 change_nexthops(fi) {
 253                         if (!nexthop_nh->nh_dev)
 254                                 continue;
 255                         hlist_del(&nexthop_nh->nh_hash);
 256                 } endfor_nexthops(fi)
 257                 fi->fib_dead = 1;
 258                 fib_info_put(fi);
 259         }
 260         spin_unlock_bh(&fib_info_lock);
 261 }
 262
 263 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 264 {
 265         const struct fib_nh *onh = ofi->fib_nh;
 266
 267         for_nexthops(fi) {
 268                 if (nh->nh_oif != onh->nh_oif ||
 269                     nh->nh_gw  != onh->nh_gw ||
 270                     nh->nh_scope != onh->nh_scope ||
 271 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 272                     nh->nh_weight != onh->nh_weight ||
 273 #endif
 274 #ifdef CONFIG_IP_ROUTE_CLASSID
 275                     nh->nh_tclassid != onh->nh_tclassid ||
 276 #endif
 277                     lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
 278                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
 279                         return -1;
 280                 onh++;
 281         } endfor_nexthops(fi);
 282         return 0;
 283 }
 284
 285 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 286 {
 287         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 288
 289         return (val ^
 290                 (val >> DEVINDEX_HASHBITS) ^
 291                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 292 }
 293
 294 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 295 {
 296         unsigned int mask = (fib_info_hash_size - 1);
 297         unsigned int val = fi->fib_nhs;
 298
 299         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 300         val ^= (__force u32)fi->fib_prefsrc;
 301         val ^= fi->fib_priority;
 302         for_nexthops(fi) {
 303                 val ^= fib_devindex_hashfn(nh->nh_oif);
 304         } endfor_nexthops(fi)
 305
 306         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 307 }
 308
 309 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 310 {
 311         struct hlist_head *head;
 312         struct fib_info *fi;
 313         unsigned int hash;
 314
 315         hash = fib_info_hashfn(nfi);
 316         head = &fib_info_hash[hash];
 317
 318         hlist_for_each_entry(fi, head, fib_hash) {
 319                 if (!net_eq(fi->fib_net, nfi->fib_net))
 320                         continue;
 321                 if (fi->fib_nhs != nfi->fib_nhs)
 322                         continue;
 323                 if (nfi->fib_protocol == fi->fib_protocol &&
 324                     nfi->fib_scope == fi->fib_scope &&
 325                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 326                     nfi->fib_priority == fi->fib_priority &&
 327                     nfi->fib_type == fi->fib_type &&
 328                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 329                            sizeof(u32) * RTAX_MAX) == 0 &&
 330                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 331                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 332                         return fi;
 333         }
 334
 335         return NULL;
 336 }
 337
 338 /* Check, that the gateway is already configured.
 339  * Used only by redirect accept routine.
 340  */
 341 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 342 {
 343         struct hlist_head *head;
 344         struct fib_nh *nh;
 345         unsigned int hash;
 346
 347         spin_lock(&fib_info_lock);
 348
 349         hash = fib_devindex_hashfn(dev->ifindex);
 350         head = &fib_info_devhash[hash];
 351         hlist_for_each_entry(nh, head, nh_hash) {
 352                 if (nh->nh_dev == dev &&
 353                     nh->nh_gw == gw &&
 354                     !(nh->nh_flags & RTNH_F_DEAD)) {
 355                         spin_unlock(&fib_info_lock);
 356                         return 0;
 357                 }
 358         }
 359
 360         spin_unlock(&fib_info_lock);
 361
 362         return -1;
 363 }
 364
 365 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 366 {
 367         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 368                          + nla_total_size(4) /* RTA_TABLE */
 369                          + nla_total_size(4) /* RTA_DST */
 370                          + nla_total_size(4) /* RTA_PRIORITY */
 371                          + nla_total_size(4) /* RTA_PREFSRC */
 372                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 373
 374         /* space for nested metrics */
 375         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 376
 377         if (fi->fib_nhs) {
 378                 size_t nh_encapsize = 0;
 379                 /* Also handles the special case fib_nhs == 1 */
 380
 381                 /* each nexthop is packed in an attribute */
 382                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 383
 384                 /* may contain flow and gateway attribute */
 385                 nhsize += 2 * nla_total_size(4);
 386
 387                 /* grab encap info */
 388                 for_nexthops(fi) {
 389                         if (nh->nh_lwtstate) {
 390                                 /* RTA_ENCAP_TYPE */
 391                                 nh_encapsize += lwtunnel_get_encap_size(
 392                                                 nh->nh_lwtstate);
 393                                 /* RTA_ENCAP */
 394                                 nh_encapsize +=  nla_total_size(2);
 395                         }
 396                 } endfor_nexthops(fi);
 397
 398                 /* all nexthops are packed in a nested attribute */
 399                 payload += nla_total_size((fi->fib_nhs * nhsize) +
 400                                           nh_encapsize);
 401
 402         }
 403
 404         return payload;
 405 }
 406
 407 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 408                int dst_len, u32 tb_id, const struct nl_info *info,
 409                unsigned int nlm_flags)
 410 {
 411         struct sk_buff *skb;
 412         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 413         int err = -ENOBUFS;
 414
 415         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 416         if (!skb)
 417                 goto errout;
 418
 419         err = fib_dump_info(skb, info->portid, seq, event, tb_id,
 420                             fa->fa_type, key, dst_len,
 421                             fa->fa_tos, fa->fa_info, nlm_flags);
 422         if (err < 0) {
 423                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 424                 WARN_ON(err == -EMSGSIZE);
 425                 kfree_skb(skb);
 426                 goto errout;
 427         }
 428         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 429                     info->nlh, GFP_KERNEL);
 430         return;
 431 errout:
 432         if (err < 0)
 433                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 434 }
 435
 436 static int fib_detect_death(struct fib_info *fi, int order,
 437                             struct fib_info **last_resort, int *last_idx,
 438                             int dflt)
 439 {
 440         struct neighbour *n;
 441         int state = NUD_NONE;
 442
 443         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 444         if (n) {
 445                 state = n->nud_state;
 446                 neigh_release(n);
 447         } else {
 448                 return 0;
 449         }
 450         if (state == NUD_REACHABLE)
 451                 return 0;
 452         if ((state & NUD_VALID) && order != dflt)
 453                 return 0;
 454         if ((state & NUD_VALID) ||
 455             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 456                 *last_resort = fi;
 457                 *last_idx = order;
 458         }
 459         return 1;
 460 }
 461
 462 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 463
 464 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
 465                               struct netlink_ext_ack *extack)
 466 {
 467         int nhs = 0;
 468
 469         while (rtnh_ok(rtnh, remaining)) {
 470                 nhs++;
 471                 rtnh = rtnh_next(rtnh, &remaining);
 472         }
 473
 474         /* leftover implies invalid nexthop configuration, discard it */
 475         if (remaining > 0) {
 476                 NL_SET_ERR_MSG(extack,
 477                                "Invalid nexthop configuration - extra data after nexthops");
 478                 nhs = 0;
 479         }
 480
 481         return nhs;
 482 }
 483
 484 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 485                        int remaining, struct fib_config *cfg,
 486                        struct netlink_ext_ack *extack)
 487 {
 488         int ret;
 489
 490         change_nexthops(fi) {
 491                 int attrlen;
 492
 493                 if (!rtnh_ok(rtnh, remaining)) {
 494                         NL_SET_ERR_MSG(extack,
 495                                        "Invalid nexthop configuration - extra data after nexthop");
 496                         return -EINVAL;
 497                 }
 498
 499                 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
 500                         NL_SET_ERR_MSG(extack,
 501                                        "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
 502                         return -EINVAL;
 503                 }
 504
 505                 nexthop_nh->nh_flags =
 506                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 507                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 508                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 509
 510                 attrlen = rtnh_attrlen(rtnh);
 511                 if (attrlen > 0) {
 512                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 513
 514                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 515                         nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
 516 #ifdef CONFIG_IP_ROUTE_CLASSID
 517                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 518                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 519                         if (nexthop_nh->nh_tclassid)
 520                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 521 #endif
 522                         nla = nla_find(attrs, attrlen, RTA_ENCAP);
 523                         if (nla) {
 524                                 struct lwtunnel_state *lwtstate;
 525                                 struct nlattr *nla_entype;
 526
 527                                 nla_entype = nla_find(attrs, attrlen,
 528                                                       RTA_ENCAP_TYPE);
 529                                 if (!nla_entype) {
 530                                         NL_SET_BAD_ATTR(extack, nla);
 531                                         NL_SET_ERR_MSG(extack,
 532                                                        "Encap type is missing");
 533                                         goto err_inval;
 534                                 }
 535
 536                                 ret = lwtunnel_build_state(nla_get_u16(
 537                                                            nla_entype),
 538                                                            nla,  AF_INET, cfg,
 539                                                            &lwtstate, extack);
 540                                 if (ret)
 541                                         goto errout;
 542                                 nexthop_nh->nh_lwtstate =
 543                                         lwtstate_get(lwtstate);
 544                         }
 545                 }
 546
 547                 rtnh = rtnh_next(rtnh, &remaining);
 548         } endfor_nexthops(fi);
 549
 550         return 0;
 551
 552 err_inval:
 553         ret = -EINVAL;
 554
 555 errout:
 556         return ret;
 557 }
 558
 559 static void fib_rebalance(struct fib_info *fi)
 560 {
 561         int total;
 562         int w;
 563         struct in_device *in_dev;
 564
 565         if (fi->fib_nhs < 2)
 566                 return;
 567
 568         total = 0;
 569         for_nexthops(fi) {
 570                 if (nh->nh_flags & RTNH_F_DEAD)
 571                         continue;
 572
 573                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
 574
 575                 if (in_dev &&
 576                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
 577                     nh->nh_flags & RTNH_F_LINKDOWN)
 578                         continue;
 579
 580                 total += nh->nh_weight;
 581         } endfor_nexthops(fi);
 582
 583         w = 0;
 584         change_nexthops(fi) {
 585                 int upper_bound;
 586
 587                 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
 588
 589                 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
 590                         upper_bound = -1;
 591                 } else if (in_dev &&
 592                            IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
 593                            nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
 594                         upper_bound = -1;
 595                 } else {
 596                         w += nexthop_nh->nh_weight;
 597                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 598                                                             total) - 1;
 599                 }
 600
 601                 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
 602         } endfor_nexthops(fi);
 603 }
 604 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 605
 606 #define fib_rebalance(fi) do { } while (0)
 607
 608 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 609
 610 static int fib_encap_match(u16 encap_type,
 611                            struct nlattr *encap,
 612                            const struct fib_nh *nh,
 613                            const struct fib_config *cfg,
 614                            struct netlink_ext_ack *extack)
 615 {
 616         struct lwtunnel_state *lwtstate;
 617         int ret, result = 0;
 618
 619         if (encap_type == LWTUNNEL_ENCAP_NONE)
 620                 return 0;
 621
 622         ret = lwtunnel_build_state(encap_type, encap, AF_INET,
 623                                    cfg, &lwtstate, extack);
 624         if (!ret) {
 625                 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
 626                 lwtstate_free(lwtstate);
 627         }
 628
 629         return result;
 630 }
 631
 632 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
 633                  struct netlink_ext_ack *extack)
 634 {
 635 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 636         struct rtnexthop *rtnh;
 637         int remaining;
 638 #endif
 639
 640         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 641                 return 1;
 642
 643         if (cfg->fc_oif || cfg->fc_gw) {
 644                 if (cfg->fc_encap) {
 645                         if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
 646                                             fi->fib_nh, cfg, extack))
 647                                 return 1;
 648                 }
 649                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 650                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 651                         return 0;
 652                 return 1;
 653         }
 654
 655 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 656         if (!cfg->fc_mp)
 657                 return 0;
 658
 659         rtnh = cfg->fc_mp;
 660         remaining = cfg->fc_mp_len;
 661
 662         for_nexthops(fi) {
 663                 int attrlen;
 664
 665                 if (!rtnh_ok(rtnh, remaining))
 666                         return -EINVAL;
 667
 668                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 669                         return 1;
 670
 671                 attrlen = rtnh_attrlen(rtnh);
 672                 if (attrlen > 0) {
 673                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 674
 675                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 676                         if (nla && nla_get_in_addr(nla) != nh->nh_gw)
 677                                 return 1;
 678 #ifdef CONFIG_IP_ROUTE_CLASSID
 679                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 680                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 681                                 return 1;
 682 #endif
 683                 }
 684
 685                 rtnh = rtnh_next(rtnh, &remaining);
 686         } endfor_nexthops(fi);
 687 #endif
 688         return 0;
 689 }
 690
 691 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
 692 {
 693         struct nlattr *nla;
 694         int remaining;
 695
 696         if (!cfg->fc_mx)
 697                 return true;
 698
 699         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 700                 int type = nla_type(nla);
 701                 u32 val;
 702
 703                 if (!type)
 704                         continue;
 705                 if (type > RTAX_MAX)
 706                         return false;
 707
 708                 if (type == RTAX_CC_ALGO) {
 709                         char tmp[TCP_CA_NAME_MAX];
 710                         bool ecn_ca = false;
 711
 712                         nla_strlcpy(tmp, nla, sizeof(tmp));
 713                         val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
 714                 } else {
 715                         val = nla_get_u32(nla);
 716                 }
 717
 718                 if (fi->fib_metrics->metrics[type - 1] != val)
 719                         return false;
 720         }
 721
 722         return true;
 723 }
 724
 725
 726 /*
 727  * Picture
 728  * -------
 729  *
 730  * Semantics of nexthop is very messy by historical reasons.
 731  * We have to take into account, that:
 732  * a) gateway can be actually local interface address,
 733  *    so that gatewayed route is direct.
 734  * b) gateway must be on-link address, possibly
 735  *    described not by an ifaddr, but also by a direct route.
 736  * c) If both gateway and interface are specified, they should not
 737  *    contradict.
 738  * d) If we use tunnel routes, gateway could be not on-link.
 739  *
 740  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 741  * results in pretty ugly and hairy code with obscure logic.
 742  *
 743  * I chose to generalized it instead, so that the size
 744  * of code does not increase practically, but it becomes
 745  * much more general.
 746  * Every prefix is assigned a "scope" value: "host" is local address,
 747  * "link" is direct route,
 748  * [ ... "site" ... "interior" ... ]
 749  * and "universe" is true gateway route with global meaning.
 750  *
 751  * Every prefix refers to a set of "nexthop"s (gw, oif),
 752  * where gw must have narrower scope. This recursion stops
 753  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 754  * which means that gw is forced to be on link.
 755  *
 756  * Code is still hairy, but now it is apparently logically
 757  * consistent and very flexible. F.e. as by-product it allows
 758  * to co-exists in peace independent exterior and interior
 759  * routing processes.
 760  *
 761  * Normally it looks as following.
 762  *
 763  * {universe prefix}  -> (gw, oif) [scope link]
 764  *                |
 765  *                |-> {link prefix} -> (gw, oif) [scope local]
 766  *                                      |
 767  *                                      |-> {local prefix} (terminal node)
 768  */
 769 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
 770                         struct netlink_ext_ack *extack)
 771 {
 772         int err = 0;
 773         struct net *net;
 774         struct net_device *dev;
 775
 776         net = cfg->fc_nlinfo.nl_net;
 777         if (nh->nh_gw) {
 778                 struct fib_result res;
 779
 780                 if (nh->nh_flags & RTNH_F_ONLINK) {
 781                         unsigned int addr_type;
 782
 783                         if (cfg->fc_scope >= RT_SCOPE_LINK) {
 784                                 NL_SET_ERR_MSG(extack,
 785                                                "Nexthop has invalid scope");
 786                                 return -EINVAL;
 787                         }
 788                         dev = __dev_get_by_index(net, nh->nh_oif);
 789                         if (!dev)
 790                                 return -ENODEV;
 791                         if (!(dev->flags & IFF_UP)) {
 792                                 NL_SET_ERR_MSG(extack,
 793                                                "Nexthop device is not up");
 794                                 return -ENETDOWN;
 795                         }
 796                         addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
 797                         if (addr_type != RTN_UNICAST) {
 798                                 NL_SET_ERR_MSG(extack,
 799                                                "Nexthop has invalid gateway");
 800                                 return -EINVAL;
 801                         }
 802                         if (!netif_carrier_ok(dev))
 803                                 nh->nh_flags |= RTNH_F_LINKDOWN;
 804                         nh->nh_dev = dev;
 805                         dev_hold(dev);
 806                         nh->nh_scope = RT_SCOPE_LINK;
 807                         return 0;
 808                 }
 809                 rcu_read_lock();
 810                 {
 811                         struct fib_table *tbl = NULL;
 812                         struct flowi4 fl4 = {
 813                                 .daddr = nh->nh_gw,
 814                                 .flowi4_scope = cfg->fc_scope + 1,
 815                                 .flowi4_oif = nh->nh_oif,
 816                                 .flowi4_iif = LOOPBACK_IFINDEX,
 817                         };
 818
 819                         /* It is not necessary, but requires a bit of thinking */
 820                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 821                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 822
 823                         if (cfg->fc_table)
 824                                 tbl = fib_get_table(net, cfg->fc_table);
 825
 826                         if (tbl)
 827                                 err = fib_table_lookup(tbl, &fl4, &res,
 828                                                        FIB_LOOKUP_IGNORE_LINKSTATE |
 829                                                        FIB_LOOKUP_NOREF);
 830
 831                         /* on error or if no table given do full lookup. This
 832                          * is needed for example when nexthops are in the local
 833                          * table rather than the given table
 834                          */
 835                         if (!tbl || err) {
 836                                 err = fib_lookup(net, &fl4, &res,
 837                                                  FIB_LOOKUP_IGNORE_LINKSTATE);
 838                         }
 839
 840                         if (err) {
 841                                 NL_SET_ERR_MSG(extack,
 842                                                "Nexthop has invalid gateway");
 843                                 rcu_read_unlock();
 844                                 return err;
 845                         }
 846                 }
 847                 err = -EINVAL;
 848                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
 849                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
 850                         goto out;
 851                 }
 852                 nh->nh_scope = res.scope;
 853                 nh->nh_oif = FIB_RES_OIF(res);
 854                 nh->nh_dev = dev = FIB_RES_DEV(res);
 855                 if (!dev) {
 856                         NL_SET_ERR_MSG(extack,
 857                                        "No egress device for nexthop gateway");
 858                         goto out;
 859                 }
 860                 dev_hold(dev);
 861                 if (!netif_carrier_ok(dev))
 862                         nh->nh_flags |= RTNH_F_LINKDOWN;
 863                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 864         } else {
 865                 struct in_device *in_dev;
 866
 867                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
 868                         NL_SET_ERR_MSG(extack,
 869                                        "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
 870                         return -EINVAL;
 871                 }
 872                 rcu_read_lock();
 873                 err = -ENODEV;
 874                 in_dev = inetdev_by_index(net, nh->nh_oif);
 875                 if (!in_dev)
 876                         goto out;
 877                 err = -ENETDOWN;
 878                 if (!(in_dev->dev->flags & IFF_UP)) {
 879                         NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
 880                         goto out;
 881                 }
 882                 nh->nh_dev = in_dev->dev;
 883                 dev_hold(nh->nh_dev);
 884                 nh->nh_scope = RT_SCOPE_HOST;
 885                 if (!netif_carrier_ok(nh->nh_dev))
 886                         nh->nh_flags |= RTNH_F_LINKDOWN;
 887                 err = 0;
 888         }
 889 out:
 890         rcu_read_unlock();
 891         return err;
 892 }
 893
 894 static inline unsigned int fib_laddr_hashfn(__be32 val)
 895 {
 896         unsigned int mask = (fib_info_hash_size - 1);
 897
 898         return ((__force u32)val ^
 899                 ((__force u32)val >> 7) ^
 900                 ((__force u32)val >> 14)) & mask;
 901 }
 902
 903 static struct hlist_head *fib_info_hash_alloc(int bytes)
 904 {
 905         if (bytes <= PAGE_SIZE)
 906                 return kzalloc(bytes, GFP_KERNEL);
 907         else
 908                 return (struct hlist_head *)
 909                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 910                                          get_order(bytes));
 911 }
 912
 913 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 914 {
 915         if (!hash)
 916                 return;
 917
 918         if (bytes <= PAGE_SIZE)
 919                 kfree(hash);
 920         else
 921                 free_pages((unsigned long) hash, get_order(bytes));
 922 }
 923
 924 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 925                                struct hlist_head *new_laddrhash,
 926                                unsigned int new_size)
 927 {
 928         struct hlist_head *old_info_hash, *old_laddrhash;
 929         unsigned int old_size = fib_info_hash_size;
 930         unsigned int i, bytes;
 931
 932         spin_lock_bh(&fib_info_lock);
 933         old_info_hash = fib_info_hash;
 934         old_laddrhash = fib_info_laddrhash;
 935         fib_info_hash_size = new_size;
 936
 937         for (i = 0; i < old_size; i++) {
 938                 struct hlist_head *head = &fib_info_hash[i];
 939                 struct hlist_node *n;
 940                 struct fib_info *fi;
 941
 942                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 943                         struct hlist_head *dest;
 944                         unsigned int new_hash;
 945
 946                         new_hash = fib_info_hashfn(fi);
 947                         dest = &new_info_hash[new_hash];
 948                         hlist_add_head(&fi->fib_hash, dest);
 949                 }
 950         }
 951         fib_info_hash = new_info_hash;
 952
 953         for (i = 0; i < old_size; i++) {
 954                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 955                 struct hlist_node *n;
 956                 struct fib_info *fi;
 957
 958                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 959                         struct hlist_head *ldest;
 960                         unsigned int new_hash;
 961
 962                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 963                         ldest = &new_laddrhash[new_hash];
 964                         hlist_add_head(&fi->fib_lhash, ldest);
 965                 }
 966         }
 967         fib_info_laddrhash = new_laddrhash;
 968
 969         spin_unlock_bh(&fib_info_lock);
 970
 971         bytes = old_size * sizeof(struct hlist_head *);
 972         fib_info_hash_free(old_info_hash, bytes);
 973         fib_info_hash_free(old_laddrhash, bytes);
 974 }
 975
 976 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 977 {
 978         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 979                                         nh->nh_gw,
 980                                         nh->nh_parent->fib_scope);
 981         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 982
 983         return nh->nh_saddr;
 984 }
 985
 986 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 987 {
 988         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 989             fib_prefsrc != cfg->fc_dst) {
 990                 u32 tb_id = cfg->fc_table;
 991                 int rc;
 992
 993                 if (tb_id == RT_TABLE_MAIN)
 994                         tb_id = RT_TABLE_LOCAL;
 995
 996                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
 997                                           fib_prefsrc, tb_id);
 998
 999                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1000                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1001                                                   fib_prefsrc, RT_TABLE_LOCAL);
1002                 }
1003
1004                 if (rc != RTN_LOCAL)
1005                         return false;
1006         }
1007         return true;
1008 }
1009
1010 static int
1011 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
1012 {
1013         bool ecn_ca = false;
1014         struct nlattr *nla;
1015         int remaining;
1016
1017         if (!cfg->fc_mx)
1018                 return 0;
1019
1020         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1021                 int type = nla_type(nla);
1022                 u32 val;
1023
1024                 if (!type)
1025                         continue;
1026                 if (type > RTAX_MAX)
1027                         return -EINVAL;
1028
1029                 if (type == RTAX_CC_ALGO) {
1030                         char tmp[TCP_CA_NAME_MAX];
1031
1032                         nla_strlcpy(tmp, nla, sizeof(tmp));
1033                         val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
1034                         if (val == TCP_CA_UNSPEC)
1035                                 return -EINVAL;
1036                 } else {
1037                         val = nla_get_u32(nla);
1038                 }
1039                 if (type == RTAX_ADVMSS && val > 65535 - 40)
1040                         val = 65535 - 40;
1041                 if (type == RTAX_MTU && val > 65535 - 15)
1042                         val = 65535 - 15;
1043                 if (type == RTAX_HOPLIMIT && val > 255)
1044                         val = 255;
1045                 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
1046                         return -EINVAL;
1047                 fi->fib_metrics->metrics[type - 1] = val;
1048         }
1049
1050         if (ecn_ca)
1051                 fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
1052
1053         return 0;
1054 }
1055
1056 struct fib_info *fib_create_info(struct fib_config *cfg,
1057                                  struct netlink_ext_ack *extack)
1058 {
1059         int err;
1060         struct fib_info *fi = NULL;
1061         struct fib_info *ofi;
1062         int nhs = 1;
1063         struct net *net = cfg->fc_nlinfo.nl_net;
1064
1065         if (cfg->fc_type > RTN_MAX)
1066                 goto err_inval;
1067
1068         /* Fast check to catch the most weird cases */
1069         if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1070                 NL_SET_ERR_MSG(extack, "Invalid scope");
1071                 goto err_inval;
1072         }
1073
1074         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1075                 NL_SET_ERR_MSG(extack,
1076                                "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1077                 goto err_inval;
1078         }
1079
1080 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1081         if (cfg->fc_mp) {
1082                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1083                 if (nhs == 0)
1084                         goto err_inval;
1085         }
1086 #endif
1087
1088         err = -ENOBUFS;
1089         if (fib_info_cnt >= fib_info_hash_size) {
1090                 unsigned int new_size = fib_info_hash_size << 1;
1091                 struct hlist_head *new_info_hash;
1092                 struct hlist_head *new_laddrhash;
1093                 unsigned int bytes;
1094
1095                 if (!new_size)
1096                         new_size = 16;
1097                 bytes = new_size * sizeof(struct hlist_head *);
1098                 new_info_hash = fib_info_hash_alloc(bytes);
1099                 new_laddrhash = fib_info_hash_alloc(bytes);
1100                 if (!new_info_hash || !new_laddrhash) {
1101                         fib_info_hash_free(new_info_hash, bytes);
1102                         fib_info_hash_free(new_laddrhash, bytes);
1103                 } else
1104                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1105
1106                 if (!fib_info_hash_size)
1107                         goto failure;
1108         }
1109
1110         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1111         if (!fi)
1112                 goto failure;
1113         if (cfg->fc_mx) {
1114                 fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1115                 if (unlikely(!fi->fib_metrics)) {
1116                         kfree(fi);
1117                         return ERR_PTR(err);
1118                 }
1119                 refcount_set(&fi->fib_metrics->refcnt, 1);
1120         } else {
1121                 fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1122         }
1123         fib_info_cnt++;
1124         fi->fib_net = net;
1125         fi->fib_protocol = cfg->fc_protocol;
1126         fi->fib_scope = cfg->fc_scope;
1127         fi->fib_flags = cfg->fc_flags;
1128         fi->fib_priority = cfg->fc_priority;
1129         fi->fib_prefsrc = cfg->fc_prefsrc;
1130         fi->fib_type = cfg->fc_type;
1131         fi->fib_tb_id = cfg->fc_table;
1132
1133         fi->fib_nhs = nhs;
1134         change_nexthops(fi) {
1135                 nexthop_nh->nh_parent = fi;
1136                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1137                 if (!nexthop_nh->nh_pcpu_rth_output)
1138                         goto failure;
1139         } endfor_nexthops(fi)
1140
1141         err = fib_convert_metrics(fi, cfg);
1142         if (err)
1143                 goto failure;
1144
1145         if (cfg->fc_mp) {
1146 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1147                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1148                 if (err != 0)
1149                         goto failure;
1150                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1151                         NL_SET_ERR_MSG(extack,
1152                                        "Nexthop device index does not match RTA_OIF");
1153                         goto err_inval;
1154                 }
1155                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1156                         NL_SET_ERR_MSG(extack,
1157                                        "Nexthop gateway does not match RTA_GATEWAY");
1158                         goto err_inval;
1159                 }
1160 #ifdef CONFIG_IP_ROUTE_CLASSID
1161                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1162                         NL_SET_ERR_MSG(extack,
1163                                        "Nexthop class id does not match RTA_FLOW");
1164                         goto err_inval;
1165                 }
1166 #endif
1167 #else
1168                 NL_SET_ERR_MSG(extack,
1169                                "Multipath support not enabled in kernel");
1170                 goto err_inval;
1171 #endif
1172         } else {
1173                 struct fib_nh *nh = fi->fib_nh;
1174
1175                 if (cfg->fc_encap) {
1176                         struct lwtunnel_state *lwtstate;
1177
1178                         if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1179                                 NL_SET_ERR_MSG(extack,
1180                                                "LWT encap type not specified");
1181                                 goto err_inval;
1182                         }
1183                         err = lwtunnel_build_state(cfg->fc_encap_type,
1184                                                    cfg->fc_encap, AF_INET, cfg,
1185                                                    &lwtstate, extack);
1186                         if (err)
1187                                 goto failure;
1188
1189                         nh->nh_lwtstate = lwtstate_get(lwtstate);
1190                 }
1191                 nh->nh_oif = cfg->fc_oif;
1192                 nh->nh_gw = cfg->fc_gw;
1193                 nh->nh_flags = cfg->fc_flags;
1194 #ifdef CONFIG_IP_ROUTE_CLASSID
1195                 nh->nh_tclassid = cfg->fc_flow;
1196                 if (nh->nh_tclassid)
1197                         fi->fib_net->ipv4.fib_num_tclassid_users++;
1198 #endif
1199 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1200                 nh->nh_weight = 1;
1201 #endif
1202         }
1203
1204         if (fib_props[cfg->fc_type].error) {
1205                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1206                         NL_SET_ERR_MSG(extack,
1207                                        "Gateway, device and multipath can not be specified for this route type");
1208                         goto err_inval;
1209                 }
1210                 goto link_it;
1211         } else {
1212                 switch (cfg->fc_type) {
1213                 case RTN_UNICAST:
1214                 case RTN_LOCAL:
1215                 case RTN_BROADCAST:
1216                 case RTN_ANYCAST:
1217                 case RTN_MULTICAST:
1218                         break;
1219                 default:
1220                         NL_SET_ERR_MSG(extack, "Invalid route type");
1221                         goto err_inval;
1222                 }
1223         }
1224
1225         if (cfg->fc_scope > RT_SCOPE_HOST) {
1226                 NL_SET_ERR_MSG(extack, "Invalid scope");
1227                 goto err_inval;
1228         }
1229
1230         if (cfg->fc_scope == RT_SCOPE_HOST) {
1231                 struct fib_nh *nh = fi->fib_nh;
1232
1233                 /* Local address is added. */
1234                 if (nhs != 1) {
1235                         NL_SET_ERR_MSG(extack,
1236                                        "Route with host scope can not have multiple nexthops");
1237                         goto err_inval;
1238                 }
1239                 if (nh->nh_gw) {
1240                         NL_SET_ERR_MSG(extack,
1241                                        "Route with host scope can not have a gateway");
1242                         goto err_inval;
1243                 }
1244                 nh->nh_scope = RT_SCOPE_NOWHERE;
1245                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1246                 err = -ENODEV;
1247                 if (!nh->nh_dev)
1248                         goto failure;
1249         } else {
1250                 int linkdown = 0;
1251
1252                 change_nexthops(fi) {
1253                         err = fib_check_nh(cfg, nexthop_nh, extack);
1254                         if (err != 0)
1255                                 goto failure;
1256                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1257                                 linkdown++;
1258                 } endfor_nexthops(fi)
1259                 if (linkdown == fi->fib_nhs)
1260                         fi->fib_flags |= RTNH_F_LINKDOWN;
1261         }
1262
1263         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1264                 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1265                 goto err_inval;
1266         }
1267
1268         change_nexthops(fi) {
1269                 fib_info_update_nh_saddr(net, nexthop_nh);
1270         } endfor_nexthops(fi)
1271
1272         fib_rebalance(fi);
1273
1274 link_it:
1275         ofi = fib_find_info(fi);
1276         if (ofi) {
1277                 fi->fib_dead = 1;
1278                 free_fib_info(fi);
1279                 ofi->fib_treeref++;
1280                 return ofi;
1281         }
1282
1283         fi->fib_treeref++;
1284         refcount_set(&fi->fib_clntref, 1);
1285         spin_lock_bh(&fib_info_lock);
1286         hlist_add_head(&fi->fib_hash,
1287                        &fib_info_hash[fib_info_hashfn(fi)]);
1288         if (fi->fib_prefsrc) {
1289                 struct hlist_head *head;
1290
1291                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1292                 hlist_add_head(&fi->fib_lhash, head);
1293         }
1294         change_nexthops(fi) {
1295                 struct hlist_head *head;
1296                 unsigned int hash;
1297
1298                 if (!nexthop_nh->nh_dev)
1299                         continue;
1300                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1301                 head = &fib_info_devhash[hash];
1302                 hlist_add_head(&nexthop_nh->nh_hash, head);
1303         } endfor_nexthops(fi)
1304         spin_unlock_bh(&fib_info_lock);
1305         return fi;
1306
1307 err_inval:
1308         err = -EINVAL;
1309
1310 failure:
1311         if (fi) {
1312                 fi->fib_dead = 1;
1313                 free_fib_info(fi);
1314         }
1315
1316         return ERR_PTR(err);
1317 }
1318
1319 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1320                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1321                   struct fib_info *fi, unsigned int flags)
1322 {
1323         struct nlmsghdr *nlh;
1324         struct rtmsg *rtm;
1325
1326         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1327         if (!nlh)
1328                 return -EMSGSIZE;
1329
1330         rtm = nlmsg_data(nlh);
1331         rtm->rtm_family = AF_INET;
1332         rtm->rtm_dst_len = dst_len;
1333         rtm->rtm_src_len = 0;
1334         rtm->rtm_tos = tos;
1335         if (tb_id < 256)
1336                 rtm->rtm_table = tb_id;
1337         else
1338                 rtm->rtm_table = RT_TABLE_COMPAT;
1339         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1340                 goto nla_put_failure;
1341         rtm->rtm_type = type;
1342         rtm->rtm_flags = fi->fib_flags;
1343         rtm->rtm_scope = fi->fib_scope;
1344         rtm->rtm_protocol = fi->fib_protocol;
1345
1346         if (rtm->rtm_dst_len &&
1347             nla_put_in_addr(skb, RTA_DST, dst))
1348                 goto nla_put_failure;
1349         if (fi->fib_priority &&
1350             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1351                 goto nla_put_failure;
1352         if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1353                 goto nla_put_failure;
1354
1355         if (fi->fib_prefsrc &&
1356             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1357                 goto nla_put_failure;
1358         if (fi->fib_nhs == 1) {
1359                 if (fi->fib_nh->nh_gw &&
1360                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1361                         goto nla_put_failure;
1362                 if (fi->fib_nh->nh_oif &&
1363                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1364                         goto nla_put_failure;
1365                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1366                         struct in_device *in_dev;
1367
1368                         rcu_read_lock();
1369                         in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev);
1370                         if (in_dev &&
1371                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1372                                 rtm->rtm_flags |= RTNH_F_DEAD;
1373                         rcu_read_unlock();
1374                 }
1375                 if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1376                         rtm->rtm_flags |= RTNH_F_OFFLOAD;
1377 #ifdef CONFIG_IP_ROUTE_CLASSID
1378                 if (fi->fib_nh[0].nh_tclassid &&
1379                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1380                         goto nla_put_failure;
1381 #endif
1382                 if (fi->fib_nh->nh_lwtstate &&
1383                     lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1384                         goto nla_put_failure;
1385         }
1386 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1387         if (fi->fib_nhs > 1) {
1388                 struct rtnexthop *rtnh;
1389                 struct nlattr *mp;
1390
1391                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1392                 if (!mp)
1393                         goto nla_put_failure;
1394
1395                 for_nexthops(fi) {
1396                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1397                         if (!rtnh)
1398                                 goto nla_put_failure;
1399
1400                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1401                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1402                                 struct in_device *in_dev;
1403
1404                                 rcu_read_lock();
1405                                 in_dev = __in_dev_get_rcu(nh->nh_dev);
1406                                 if (in_dev &&
1407                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1408                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1409                                 rcu_read_unlock();
1410                         }
1411                         rtnh->rtnh_hops = nh->nh_weight - 1;
1412                         rtnh->rtnh_ifindex = nh->nh_oif;
1413
1414                         if (nh->nh_gw &&
1415                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1416                                 goto nla_put_failure;
1417 #ifdef CONFIG_IP_ROUTE_CLASSID
1418                         if (nh->nh_tclassid &&
1419                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1420                                 goto nla_put_failure;
1421 #endif
1422                         if (nh->nh_lwtstate &&
1423                             lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1424                                 goto nla_put_failure;
1425
1426                         /* length of rtnetlink header + attributes */
1427                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1428                 } endfor_nexthops(fi);
1429
1430                 nla_nest_end(skb, mp);
1431         }
1432 #endif
1433         nlmsg_end(skb, nlh);
1434         return 0;
1435
1436 nla_put_failure:
1437         nlmsg_cancel(skb, nlh);
1438         return -EMSGSIZE;
1439 }
1440
1441 /*
1442  * Update FIB if:
1443  * - local address disappeared -> we must delete all the entries
1444  *   referring to it.
1445  * - device went down -> we must shutdown all nexthops going via it.
1446  */
1447 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1448 {
1449         int ret = 0;
1450         unsigned int hash = fib_laddr_hashfn(local);
1451         struct hlist_head *head = &fib_info_laddrhash[hash];
1452         struct net *net = dev_net(dev);
1453         int tb_id = l3mdev_fib_table(dev);
1454         struct fib_info *fi;
1455
1456         if (!fib_info_laddrhash || local == 0)
1457                 return 0;
1458
1459         hlist_for_each_entry(fi, head, fib_lhash) {
1460                 if (!net_eq(fi->fib_net, net) ||
1461                     fi->fib_tb_id != tb_id)
1462                         continue;
1463                 if (fi->fib_prefsrc == local) {
1464                         fi->fib_flags |= RTNH_F_DEAD;
1465                         ret++;
1466                 }
1467         }
1468         return ret;
1469 }
1470
1471 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1472                                  enum fib_event_type event_type)
1473 {
1474         struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1475         struct fib_nh_notifier_info info = {
1476                 .fib_nh = fib_nh,
1477         };
1478
1479         switch (event_type) {
1480         case FIB_EVENT_NH_ADD:
1481                 if (fib_nh->nh_flags & RTNH_F_DEAD)
1482                         break;
1483                 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1484                     fib_nh->nh_flags & RTNH_F_LINKDOWN)
1485                         break;
1486                 return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1487                                            &info.info);
1488         case FIB_EVENT_NH_DEL:
1489                 if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1490                      fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1491                     (fib_nh->nh_flags & RTNH_F_DEAD))
1492                         return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1493                                                    event_type, &info.info);
1494         default:
1495                 break;
1496         }
1497
1498         return NOTIFY_DONE;
1499 }
1500
1501 /* Event              force Flags           Description
1502  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1503  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1504  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1505  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1506  */
1507 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1508 {
1509         int ret = 0;
1510         int scope = RT_SCOPE_NOWHERE;
1511         struct fib_info *prev_fi = NULL;
1512         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1513         struct hlist_head *head = &fib_info_devhash[hash];
1514         struct fib_nh *nh;
1515
1516         if (force)
1517                 scope = -1;
1518
1519         hlist_for_each_entry(nh, head, nh_hash) {
1520                 struct fib_info *fi = nh->nh_parent;
1521                 int dead;
1522
1523                 BUG_ON(!fi->fib_nhs);
1524                 if (nh->nh_dev != dev || fi == prev_fi)
1525                         continue;
1526                 prev_fi = fi;
1527                 dead = 0;
1528                 change_nexthops(fi) {
1529                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1530                                 dead++;
1531                         else if (nexthop_nh->nh_dev == dev &&
1532                                  nexthop_nh->nh_scope != scope) {
1533                                 switch (event) {
1534                                 case NETDEV_DOWN:
1535                                 case NETDEV_UNREGISTER:
1536                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1537                                         /* fall through */
1538                                 case NETDEV_CHANGE:
1539                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1540                                         break;
1541                                 }
1542                                 call_fib_nh_notifiers(nexthop_nh,
1543                                                       FIB_EVENT_NH_DEL);
1544                                 dead++;
1545                         }
1546 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1547                         if (event == NETDEV_UNREGISTER &&
1548                             nexthop_nh->nh_dev == dev) {
1549                                 dead = fi->fib_nhs;
1550                                 break;
1551                         }
1552 #endif
1553                 } endfor_nexthops(fi)
1554                 if (dead == fi->fib_nhs) {
1555                         switch (event) {
1556                         case NETDEV_DOWN:
1557                         case NETDEV_UNREGISTER:
1558                                 fi->fib_flags |= RTNH_F_DEAD;
1559                                 /* fall through */
1560                         case NETDEV_CHANGE:
1561                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1562                                 break;
1563                         }
1564                         ret++;
1565                 }
1566
1567                 fib_rebalance(fi);
1568         }
1569
1570         return ret;
1571 }
1572
1573 /* Must be invoked inside of an RCU protected region.  */
1574 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1575 {
1576         struct fib_info *fi = NULL, *last_resort = NULL;
1577         struct hlist_head *fa_head = res->fa_head;
1578         struct fib_table *tb = res->table;
1579         u8 slen = 32 - res->prefixlen;
1580         int order = -1, last_idx = -1;
1581         struct fib_alias *fa, *fa1 = NULL;
1582         u32 last_prio = res->fi->fib_priority;
1583         u8 last_tos = 0;
1584
1585         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1586                 struct fib_info *next_fi = fa->fa_info;
1587
1588                 if (fa->fa_slen != slen)
1589                         continue;
1590                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1591                         continue;
1592                 if (fa->tb_id != tb->tb_id)
1593                         continue;
1594                 if (next_fi->fib_priority > last_prio &&
1595                     fa->fa_tos == last_tos) {
1596                         if (last_tos)
1597                                 continue;
1598                         break;
1599                 }
1600                 if (next_fi->fib_flags & RTNH_F_DEAD)
1601                         continue;
1602                 last_tos = fa->fa_tos;
1603                 last_prio = next_fi->fib_priority;
1604
1605                 if (next_fi->fib_scope != res->scope ||
1606                     fa->fa_type != RTN_UNICAST)
1607                         continue;
1608                 if (!next_fi->fib_nh[0].nh_gw ||
1609                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1610                         continue;
1611
1612                 fib_alias_accessed(fa);
1613
1614                 if (!fi) {
1615                         if (next_fi != res->fi)
1616                                 break;
1617                         fa1 = fa;
1618                 } else if (!fib_detect_death(fi, order, &last_resort,
1619                                              &last_idx, fa1->fa_default)) {
1620                         fib_result_assign(res, fi);
1621                         fa1->fa_default = order;
1622                         goto out;
1623                 }
1624                 fi = next_fi;
1625                 order++;
1626         }
1627
1628         if (order <= 0 || !fi) {
1629                 if (fa1)
1630                         fa1->fa_default = -1;
1631                 goto out;
1632         }
1633
1634         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1635                               fa1->fa_default)) {
1636                 fib_result_assign(res, fi);
1637                 fa1->fa_default = order;
1638                 goto out;
1639         }
1640
1641         if (last_idx >= 0)
1642                 fib_result_assign(res, last_resort);
1643         fa1->fa_default = last_idx;
1644 out:
1645         return;
1646 }
1647
1648 /*
1649  * Dead device goes up. We wake up dead nexthops.
1650  * It takes sense only on multipath routes.
1651  */
1652 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1653 {
1654         struct fib_info *prev_fi;
1655         unsigned int hash;
1656         struct hlist_head *head;
1657         struct fib_nh *nh;
1658         int ret;
1659
1660         if (!(dev->flags & IFF_UP))
1661                 return 0;
1662
1663         if (nh_flags & RTNH_F_DEAD) {
1664                 unsigned int flags = dev_get_flags(dev);
1665
1666                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1667                         nh_flags |= RTNH_F_LINKDOWN;
1668         }
1669
1670         prev_fi = NULL;
1671         hash = fib_devindex_hashfn(dev->ifindex);
1672         head = &fib_info_devhash[hash];
1673         ret = 0;
1674
1675         hlist_for_each_entry(nh, head, nh_hash) {
1676                 struct fib_info *fi = nh->nh_parent;
1677                 int alive;
1678
1679                 BUG_ON(!fi->fib_nhs);
1680                 if (nh->nh_dev != dev || fi == prev_fi)
1681                         continue;
1682
1683                 prev_fi = fi;
1684                 alive = 0;
1685                 change_nexthops(fi) {
1686                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1687                                 alive++;
1688                                 continue;
1689                         }
1690                         if (!nexthop_nh->nh_dev ||
1691                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1692                                 continue;
1693                         if (nexthop_nh->nh_dev != dev ||
1694                             !__in_dev_get_rtnl(dev))
1695                                 continue;
1696                         alive++;
1697                         nexthop_nh->nh_flags &= ~nh_flags;
1698                         call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1699                 } endfor_nexthops(fi)
1700
1701                 if (alive > 0) {
1702                         fi->fib_flags &= ~nh_flags;
1703                         ret++;
1704                 }
1705
1706                 fib_rebalance(fi);
1707         }
1708
1709         return ret;
1710 }
1711
1712 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1713 static bool fib_good_nh(const struct fib_nh *nh)
1714 {
1715         int state = NUD_REACHABLE;
1716
1717         if (nh->nh_scope == RT_SCOPE_LINK) {
1718                 struct neighbour *n;
1719
1720                 rcu_read_lock_bh();
1721
1722                 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1723                                               (__force u32)nh->nh_gw);
1724                 if (n)
1725                         state = n->nud_state;
1726
1727                 rcu_read_unlock_bh();
1728         }
1729
1730         return !!(state & NUD_VALID);
1731 }
1732
1733 void fib_select_multipath(struct fib_result *res, int hash)
1734 {
1735         struct fib_info *fi = res->fi;
1736         struct net *net = fi->fib_net;
1737         bool first = false;
1738
1739         for_nexthops(fi) {
1740                 if (hash > atomic_read(&nh->nh_upper_bound))
1741                         continue;
1742
1743                 if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1744                     fib_good_nh(nh)) {
1745                         res->nh_sel = nhsel;
1746                         return;
1747                 }
1748                 if (!first) {
1749                         res->nh_sel = nhsel;
1750                         first = true;
1751                 }
1752         } endfor_nexthops(fi);
1753 }
1754 #endif
1755
1756 void fib_select_path(struct net *net, struct fib_result *res,
1757                      struct flowi4 *fl4, const struct sk_buff *skb)
1758 {
1759         bool oif_check;
1760
1761         oif_check = (fl4->flowi4_oif == 0 ||
1762                      fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1763
1764 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1765         if (res->fi->fib_nhs > 1 && oif_check) {
1766                 int h = fib_multipath_hash(res->fi, fl4, skb);
1767
1768                 fib_select_multipath(res, h);
1769         }
1770         else
1771 #endif
1772         if (!res->prefixlen &&
1773             res->table->tb_num_default > 1 &&
1774             res->type == RTN_UNICAST && oif_check)
1775                 fib_select_default(fl4, res);
1776
1777         if (!fl4->saddr)
1778                 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1779 }
1780 EXPORT_SYMBOL_GPL(fib_select_path);