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 #ifdef CONFIG_IP_ROUTE_CLASSID
 650                 if (cfg->fc_flow &&
 651                     cfg->fc_flow != fi->fib_nh->nh_tclassid)
 652                         return 1;
 653 #endif
 654                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 655                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 656                         return 0;
 657                 return 1;
 658         }
 659
 660 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 661         if (!cfg->fc_mp)
 662                 return 0;
 663
 664         rtnh = cfg->fc_mp;
 665         remaining = cfg->fc_mp_len;
 666
 667         for_nexthops(fi) {
 668                 int attrlen;
 669
 670                 if (!rtnh_ok(rtnh, remaining))
 671                         return -EINVAL;
 672
 673                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 674                         return 1;
 675
 676                 attrlen = rtnh_attrlen(rtnh);
 677                 if (attrlen > 0) {
 678                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 679
 680                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 681                         if (nla && nla_get_in_addr(nla) != nh->nh_gw)
 682                                 return 1;
 683 #ifdef CONFIG_IP_ROUTE_CLASSID
 684                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 685                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 686                                 return 1;
 687 #endif
 688                 }
 689
 690                 rtnh = rtnh_next(rtnh, &remaining);
 691         } endfor_nexthops(fi);
 692 #endif
 693         return 0;
 694 }
 695
 696 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
 697 {
 698         struct nlattr *nla;
 699         int remaining;
 700
 701         if (!cfg->fc_mx)
 702                 return true;
 703
 704         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 705                 int type = nla_type(nla);
 706                 u32 fi_val, val;
 707
 708                 if (!type)
 709                         continue;
 710                 if (type > RTAX_MAX)
 711                         return false;
 712
 713                 if (type == RTAX_CC_ALGO) {
 714                         char tmp[TCP_CA_NAME_MAX];
 715                         bool ecn_ca = false;
 716
 717                         nla_strlcpy(tmp, nla, sizeof(tmp));
 718                         val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
 719                 } else {
 720                         if (nla_len(nla) != sizeof(u32))
 721                                 return false;
 722                         val = nla_get_u32(nla);
 723                 }
 724
 725                 fi_val = fi->fib_metrics->metrics[type - 1];
 726                 if (type == RTAX_FEATURES)
 727                         fi_val &= ~DST_FEATURE_ECN_CA;
 728
 729                 if (fi_val != val)
 730                         return false;
 731         }
 732
 733         return true;
 734 }
 735
 736
 737 /*
 738  * Picture
 739  * -------
 740  *
 741  * Semantics of nexthop is very messy by historical reasons.
 742  * We have to take into account, that:
 743  * a) gateway can be actually local interface address,
 744  *    so that gatewayed route is direct.
 745  * b) gateway must be on-link address, possibly
 746  *    described not by an ifaddr, but also by a direct route.
 747  * c) If both gateway and interface are specified, they should not
 748  *    contradict.
 749  * d) If we use tunnel routes, gateway could be not on-link.
 750  *
 751  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 752  * results in pretty ugly and hairy code with obscure logic.
 753  *
 754  * I chose to generalized it instead, so that the size
 755  * of code does not increase practically, but it becomes
 756  * much more general.
 757  * Every prefix is assigned a "scope" value: "host" is local address,
 758  * "link" is direct route,
 759  * [ ... "site" ... "interior" ... ]
 760  * and "universe" is true gateway route with global meaning.
 761  *
 762  * Every prefix refers to a set of "nexthop"s (gw, oif),
 763  * where gw must have narrower scope. This recursion stops
 764  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 765  * which means that gw is forced to be on link.
 766  *
 767  * Code is still hairy, but now it is apparently logically
 768  * consistent and very flexible. F.e. as by-product it allows
 769  * to co-exists in peace independent exterior and interior
 770  * routing processes.
 771  *
 772  * Normally it looks as following.
 773  *
 774  * {universe prefix}  -> (gw, oif) [scope link]
 775  *                |
 776  *                |-> {link prefix} -> (gw, oif) [scope local]
 777  *                                      |
 778  *                                      |-> {local prefix} (terminal node)
 779  */
 780 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
 781                         struct netlink_ext_ack *extack)
 782 {
 783         int err = 0;
 784         struct net *net;
 785         struct net_device *dev;
 786
 787         net = cfg->fc_nlinfo.nl_net;
 788         if (nh->nh_gw) {
 789                 struct fib_result res;
 790
 791                 if (nh->nh_flags & RTNH_F_ONLINK) {
 792                         unsigned int addr_type;
 793
 794                         if (cfg->fc_scope >= RT_SCOPE_LINK) {
 795                                 NL_SET_ERR_MSG(extack,
 796                                                "Nexthop has invalid scope");
 797                                 return -EINVAL;
 798                         }
 799                         dev = __dev_get_by_index(net, nh->nh_oif);
 800                         if (!dev)
 801                                 return -ENODEV;
 802                         if (!(dev->flags & IFF_UP)) {
 803                                 NL_SET_ERR_MSG(extack,
 804                                                "Nexthop device is not up");
 805                                 return -ENETDOWN;
 806                         }
 807                         addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
 808                         if (addr_type != RTN_UNICAST) {
 809                                 NL_SET_ERR_MSG(extack,
 810                                                "Nexthop has invalid gateway");
 811                                 return -EINVAL;
 812                         }
 813                         if (!netif_carrier_ok(dev))
 814                                 nh->nh_flags |= RTNH_F_LINKDOWN;
 815                         nh->nh_dev = dev;
 816                         dev_hold(dev);
 817                         nh->nh_scope = RT_SCOPE_LINK;
 818                         return 0;
 819                 }
 820                 rcu_read_lock();
 821                 {
 822                         struct fib_table *tbl = NULL;
 823                         struct flowi4 fl4 = {
 824                                 .daddr = nh->nh_gw,
 825                                 .flowi4_scope = cfg->fc_scope + 1,
 826                                 .flowi4_oif = nh->nh_oif,
 827                                 .flowi4_iif = LOOPBACK_IFINDEX,
 828                         };
 829
 830                         /* It is not necessary, but requires a bit of thinking */
 831                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 832                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 833
 834                         if (cfg->fc_table && cfg->fc_table != RT_TABLE_MAIN)
 835                                 tbl = fib_get_table(net, cfg->fc_table);
 836
 837                         if (tbl)
 838                                 err = fib_table_lookup(tbl, &fl4, &res,
 839                                                        FIB_LOOKUP_IGNORE_LINKSTATE |
 840                                                        FIB_LOOKUP_NOREF);
 841
 842                         /* on error or if no table given do full lookup. This
 843                          * is needed for example when nexthops are in the local
 844                          * table rather than the given table
 845                          */
 846                         if (!tbl || err) {
 847                                 err = fib_lookup(net, &fl4, &res,
 848                                                  FIB_LOOKUP_IGNORE_LINKSTATE);
 849                         }
 850
 851                         if (err) {
 852                                 NL_SET_ERR_MSG(extack,
 853                                                "Nexthop has invalid gateway");
 854                                 rcu_read_unlock();
 855                                 return err;
 856                         }
 857                 }
 858                 err = -EINVAL;
 859                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
 860                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
 861                         goto out;
 862                 }
 863                 nh->nh_scope = res.scope;
 864                 nh->nh_oif = FIB_RES_OIF(res);
 865                 nh->nh_dev = dev = FIB_RES_DEV(res);
 866                 if (!dev) {
 867                         NL_SET_ERR_MSG(extack,
 868                                        "No egress device for nexthop gateway");
 869                         goto out;
 870                 }
 871                 dev_hold(dev);
 872                 if (!netif_carrier_ok(dev))
 873                         nh->nh_flags |= RTNH_F_LINKDOWN;
 874                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 875         } else {
 876                 struct in_device *in_dev;
 877
 878                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
 879                         NL_SET_ERR_MSG(extack,
 880                                        "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
 881                         return -EINVAL;
 882                 }
 883                 rcu_read_lock();
 884                 err = -ENODEV;
 885                 in_dev = inetdev_by_index(net, nh->nh_oif);
 886                 if (!in_dev)
 887                         goto out;
 888                 err = -ENETDOWN;
 889                 if (!(in_dev->dev->flags & IFF_UP)) {
 890                         NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
 891                         goto out;
 892                 }
 893                 nh->nh_dev = in_dev->dev;
 894                 dev_hold(nh->nh_dev);
 895                 nh->nh_scope = RT_SCOPE_HOST;
 896                 if (!netif_carrier_ok(nh->nh_dev))
 897                         nh->nh_flags |= RTNH_F_LINKDOWN;
 898                 err = 0;
 899         }
 900 out:
 901         rcu_read_unlock();
 902         return err;
 903 }
 904
 905 static inline unsigned int fib_laddr_hashfn(__be32 val)
 906 {
 907         unsigned int mask = (fib_info_hash_size - 1);
 908
 909         return ((__force u32)val ^
 910                 ((__force u32)val >> 7) ^
 911                 ((__force u32)val >> 14)) & mask;
 912 }
 913
 914 static struct hlist_head *fib_info_hash_alloc(int bytes)
 915 {
 916         if (bytes <= PAGE_SIZE)
 917                 return kzalloc(bytes, GFP_KERNEL);
 918         else
 919                 return (struct hlist_head *)
 920                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 921                                          get_order(bytes));
 922 }
 923
 924 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 925 {
 926         if (!hash)
 927                 return;
 928
 929         if (bytes <= PAGE_SIZE)
 930                 kfree(hash);
 931         else
 932                 free_pages((unsigned long) hash, get_order(bytes));
 933 }
 934
 935 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 936                                struct hlist_head *new_laddrhash,
 937                                unsigned int new_size)
 938 {
 939         struct hlist_head *old_info_hash, *old_laddrhash;
 940         unsigned int old_size = fib_info_hash_size;
 941         unsigned int i, bytes;
 942
 943         spin_lock_bh(&fib_info_lock);
 944         old_info_hash = fib_info_hash;
 945         old_laddrhash = fib_info_laddrhash;
 946         fib_info_hash_size = new_size;
 947
 948         for (i = 0; i < old_size; i++) {
 949                 struct hlist_head *head = &fib_info_hash[i];
 950                 struct hlist_node *n;
 951                 struct fib_info *fi;
 952
 953                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 954                         struct hlist_head *dest;
 955                         unsigned int new_hash;
 956
 957                         new_hash = fib_info_hashfn(fi);
 958                         dest = &new_info_hash[new_hash];
 959                         hlist_add_head(&fi->fib_hash, dest);
 960                 }
 961         }
 962         fib_info_hash = new_info_hash;
 963
 964         for (i = 0; i < old_size; i++) {
 965                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 966                 struct hlist_node *n;
 967                 struct fib_info *fi;
 968
 969                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 970                         struct hlist_head *ldest;
 971                         unsigned int new_hash;
 972
 973                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 974                         ldest = &new_laddrhash[new_hash];
 975                         hlist_add_head(&fi->fib_lhash, ldest);
 976                 }
 977         }
 978         fib_info_laddrhash = new_laddrhash;
 979
 980         spin_unlock_bh(&fib_info_lock);
 981
 982         bytes = old_size * sizeof(struct hlist_head *);
 983         fib_info_hash_free(old_info_hash, bytes);
 984         fib_info_hash_free(old_laddrhash, bytes);
 985 }
 986
 987 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 988 {
 989         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 990                                         nh->nh_gw,
 991                                         nh->nh_parent->fib_scope);
 992         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 993
 994         return nh->nh_saddr;
 995 }
 996
 997 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 998 {
 999         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1000             fib_prefsrc != cfg->fc_dst) {
1001                 u32 tb_id = cfg->fc_table;
1002                 int rc;
1003
1004                 if (tb_id == RT_TABLE_MAIN)
1005                         tb_id = RT_TABLE_LOCAL;
1006
1007                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1008                                           fib_prefsrc, tb_id);
1009
1010                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1011                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1012                                                   fib_prefsrc, RT_TABLE_LOCAL);
1013                 }
1014
1015                 if (rc != RTN_LOCAL)
1016                         return false;
1017         }
1018         return true;
1019 }
1020
1021 static int
1022 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
1023 {
1024         return ip_metrics_convert(fi->fib_net, cfg->fc_mx, cfg->fc_mx_len,
1025                                   fi->fib_metrics->metrics);
1026 }
1027
1028 struct fib_info *fib_create_info(struct fib_config *cfg,
1029                                  struct netlink_ext_ack *extack)
1030 {
1031         int err;
1032         struct fib_info *fi = NULL;
1033         struct fib_info *ofi;
1034         int nhs = 1;
1035         struct net *net = cfg->fc_nlinfo.nl_net;
1036
1037         if (cfg->fc_type > RTN_MAX)
1038                 goto err_inval;
1039
1040         /* Fast check to catch the most weird cases */
1041         if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1042                 NL_SET_ERR_MSG(extack, "Invalid scope");
1043                 goto err_inval;
1044         }
1045
1046         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1047                 NL_SET_ERR_MSG(extack,
1048                                "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1049                 goto err_inval;
1050         }
1051
1052 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1053         if (cfg->fc_mp) {
1054                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1055                 if (nhs == 0)
1056                         goto err_inval;
1057         }
1058 #endif
1059
1060         err = -ENOBUFS;
1061         if (fib_info_cnt >= fib_info_hash_size) {
1062                 unsigned int new_size = fib_info_hash_size << 1;
1063                 struct hlist_head *new_info_hash;
1064                 struct hlist_head *new_laddrhash;
1065                 unsigned int bytes;
1066
1067                 if (!new_size)
1068                         new_size = 16;
1069                 bytes = new_size * sizeof(struct hlist_head *);
1070                 new_info_hash = fib_info_hash_alloc(bytes);
1071                 new_laddrhash = fib_info_hash_alloc(bytes);
1072                 if (!new_info_hash || !new_laddrhash) {
1073                         fib_info_hash_free(new_info_hash, bytes);
1074                         fib_info_hash_free(new_laddrhash, bytes);
1075                 } else
1076                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1077
1078                 if (!fib_info_hash_size)
1079                         goto failure;
1080         }
1081
1082         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1083         if (!fi)
1084                 goto failure;
1085         if (cfg->fc_mx) {
1086                 fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1087                 if (unlikely(!fi->fib_metrics)) {
1088                         kfree(fi);
1089                         return ERR_PTR(err);
1090                 }
1091                 refcount_set(&fi->fib_metrics->refcnt, 1);
1092         } else {
1093                 fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1094         }
1095         fib_info_cnt++;
1096         fi->fib_net = net;
1097         fi->fib_protocol = cfg->fc_protocol;
1098         fi->fib_scope = cfg->fc_scope;
1099         fi->fib_flags = cfg->fc_flags;
1100         fi->fib_priority = cfg->fc_priority;
1101         fi->fib_prefsrc = cfg->fc_prefsrc;
1102         fi->fib_type = cfg->fc_type;
1103         fi->fib_tb_id = cfg->fc_table;
1104
1105         fi->fib_nhs = nhs;
1106         change_nexthops(fi) {
1107                 nexthop_nh->nh_parent = fi;
1108                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1109                 if (!nexthop_nh->nh_pcpu_rth_output)
1110                         goto failure;
1111         } endfor_nexthops(fi)
1112
1113         err = fib_convert_metrics(fi, cfg);
1114         if (err)
1115                 goto failure;
1116
1117         if (cfg->fc_mp) {
1118 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1119                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1120                 if (err != 0)
1121                         goto failure;
1122                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1123                         NL_SET_ERR_MSG(extack,
1124                                        "Nexthop device index does not match RTA_OIF");
1125                         goto err_inval;
1126                 }
1127                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1128                         NL_SET_ERR_MSG(extack,
1129                                        "Nexthop gateway does not match RTA_GATEWAY");
1130                         goto err_inval;
1131                 }
1132 #ifdef CONFIG_IP_ROUTE_CLASSID
1133                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1134                         NL_SET_ERR_MSG(extack,
1135                                        "Nexthop class id does not match RTA_FLOW");
1136                         goto err_inval;
1137                 }
1138 #endif
1139 #else
1140                 NL_SET_ERR_MSG(extack,
1141                                "Multipath support not enabled in kernel");
1142                 goto err_inval;
1143 #endif
1144         } else {
1145                 struct fib_nh *nh = fi->fib_nh;
1146
1147                 if (cfg->fc_encap) {
1148                         struct lwtunnel_state *lwtstate;
1149
1150                         if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1151                                 NL_SET_ERR_MSG(extack,
1152                                                "LWT encap type not specified");
1153                                 goto err_inval;
1154                         }
1155                         err = lwtunnel_build_state(cfg->fc_encap_type,
1156                                                    cfg->fc_encap, AF_INET, cfg,
1157                                                    &lwtstate, extack);
1158                         if (err)
1159                                 goto failure;
1160
1161                         nh->nh_lwtstate = lwtstate_get(lwtstate);
1162                 }
1163                 nh->nh_oif = cfg->fc_oif;
1164                 nh->nh_gw = cfg->fc_gw;
1165                 nh->nh_flags = cfg->fc_flags;
1166 #ifdef CONFIG_IP_ROUTE_CLASSID
1167                 nh->nh_tclassid = cfg->fc_flow;
1168                 if (nh->nh_tclassid)
1169                         fi->fib_net->ipv4.fib_num_tclassid_users++;
1170 #endif
1171 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1172                 nh->nh_weight = 1;
1173 #endif
1174         }
1175
1176         if (fib_props[cfg->fc_type].error) {
1177                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1178                         NL_SET_ERR_MSG(extack,
1179                                        "Gateway, device and multipath can not be specified for this route type");
1180                         goto err_inval;
1181                 }
1182                 goto link_it;
1183         } else {
1184                 switch (cfg->fc_type) {
1185                 case RTN_UNICAST:
1186                 case RTN_LOCAL:
1187                 case RTN_BROADCAST:
1188                 case RTN_ANYCAST:
1189                 case RTN_MULTICAST:
1190                         break;
1191                 default:
1192                         NL_SET_ERR_MSG(extack, "Invalid route type");
1193                         goto err_inval;
1194                 }
1195         }
1196
1197         if (cfg->fc_scope > RT_SCOPE_HOST) {
1198                 NL_SET_ERR_MSG(extack, "Invalid scope");
1199                 goto err_inval;
1200         }
1201
1202         if (cfg->fc_scope == RT_SCOPE_HOST) {
1203                 struct fib_nh *nh = fi->fib_nh;
1204
1205                 /* Local address is added. */
1206                 if (nhs != 1) {
1207                         NL_SET_ERR_MSG(extack,
1208                                        "Route with host scope can not have multiple nexthops");
1209                         goto err_inval;
1210                 }
1211                 if (nh->nh_gw) {
1212                         NL_SET_ERR_MSG(extack,
1213                                        "Route with host scope can not have a gateway");
1214                         goto err_inval;
1215                 }
1216                 nh->nh_scope = RT_SCOPE_NOWHERE;
1217                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1218                 err = -ENODEV;
1219                 if (!nh->nh_dev)
1220                         goto failure;
1221         } else {
1222                 int linkdown = 0;
1223
1224                 change_nexthops(fi) {
1225                         err = fib_check_nh(cfg, nexthop_nh, extack);
1226                         if (err != 0)
1227                                 goto failure;
1228                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1229                                 linkdown++;
1230                 } endfor_nexthops(fi)
1231                 if (linkdown == fi->fib_nhs)
1232                         fi->fib_flags |= RTNH_F_LINKDOWN;
1233         }
1234
1235         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1236                 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1237                 goto err_inval;
1238         }
1239
1240         change_nexthops(fi) {
1241                 fib_info_update_nh_saddr(net, nexthop_nh);
1242         } endfor_nexthops(fi)
1243
1244         fib_rebalance(fi);
1245
1246 link_it:
1247         ofi = fib_find_info(fi);
1248         if (ofi) {
1249                 fi->fib_dead = 1;
1250                 free_fib_info(fi);
1251                 ofi->fib_treeref++;
1252                 return ofi;
1253         }
1254
1255         fi->fib_treeref++;
1256         refcount_set(&fi->fib_clntref, 1);
1257         spin_lock_bh(&fib_info_lock);
1258         hlist_add_head(&fi->fib_hash,
1259                        &fib_info_hash[fib_info_hashfn(fi)]);
1260         if (fi->fib_prefsrc) {
1261                 struct hlist_head *head;
1262
1263                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1264                 hlist_add_head(&fi->fib_lhash, head);
1265         }
1266         change_nexthops(fi) {
1267                 struct hlist_head *head;
1268                 unsigned int hash;
1269
1270                 if (!nexthop_nh->nh_dev)
1271                         continue;
1272                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1273                 head = &fib_info_devhash[hash];
1274                 hlist_add_head(&nexthop_nh->nh_hash, head);
1275         } endfor_nexthops(fi)
1276         spin_unlock_bh(&fib_info_lock);
1277         return fi;
1278
1279 err_inval:
1280         err = -EINVAL;
1281
1282 failure:
1283         if (fi) {
1284                 fi->fib_dead = 1;
1285                 free_fib_info(fi);
1286         }
1287
1288         return ERR_PTR(err);
1289 }
1290
1291 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1292                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1293                   struct fib_info *fi, unsigned int flags)
1294 {
1295         struct nlmsghdr *nlh;
1296         struct rtmsg *rtm;
1297
1298         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1299         if (!nlh)
1300                 return -EMSGSIZE;
1301
1302         rtm = nlmsg_data(nlh);
1303         rtm->rtm_family = AF_INET;
1304         rtm->rtm_dst_len = dst_len;
1305         rtm->rtm_src_len = 0;
1306         rtm->rtm_tos = tos;
1307         if (tb_id < 256)
1308                 rtm->rtm_table = tb_id;
1309         else
1310                 rtm->rtm_table = RT_TABLE_COMPAT;
1311         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1312                 goto nla_put_failure;
1313         rtm->rtm_type = type;
1314         rtm->rtm_flags = fi->fib_flags;
1315         rtm->rtm_scope = fi->fib_scope;
1316         rtm->rtm_protocol = fi->fib_protocol;
1317
1318         if (rtm->rtm_dst_len &&
1319             nla_put_in_addr(skb, RTA_DST, dst))
1320                 goto nla_put_failure;
1321         if (fi->fib_priority &&
1322             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1323                 goto nla_put_failure;
1324         if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1325                 goto nla_put_failure;
1326
1327         if (fi->fib_prefsrc &&
1328             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1329                 goto nla_put_failure;
1330         if (fi->fib_nhs == 1) {
1331                 if (fi->fib_nh->nh_gw &&
1332                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1333                         goto nla_put_failure;
1334                 if (fi->fib_nh->nh_oif &&
1335                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1336                         goto nla_put_failure;
1337                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1338                         struct in_device *in_dev;
1339
1340                         rcu_read_lock();
1341                         in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev);
1342                         if (in_dev &&
1343                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1344                                 rtm->rtm_flags |= RTNH_F_DEAD;
1345                         rcu_read_unlock();
1346                 }
1347                 if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1348                         rtm->rtm_flags |= RTNH_F_OFFLOAD;
1349 #ifdef CONFIG_IP_ROUTE_CLASSID
1350                 if (fi->fib_nh[0].nh_tclassid &&
1351                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1352                         goto nla_put_failure;
1353 #endif
1354                 if (fi->fib_nh->nh_lwtstate &&
1355                     lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1356                         goto nla_put_failure;
1357         }
1358 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1359         if (fi->fib_nhs > 1) {
1360                 struct rtnexthop *rtnh;
1361                 struct nlattr *mp;
1362
1363                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1364                 if (!mp)
1365                         goto nla_put_failure;
1366
1367                 for_nexthops(fi) {
1368                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1369                         if (!rtnh)
1370                                 goto nla_put_failure;
1371
1372                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1373                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1374                                 struct in_device *in_dev;
1375
1376                                 rcu_read_lock();
1377                                 in_dev = __in_dev_get_rcu(nh->nh_dev);
1378                                 if (in_dev &&
1379                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1380                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1381                                 rcu_read_unlock();
1382                         }
1383                         rtnh->rtnh_hops = nh->nh_weight - 1;
1384                         rtnh->rtnh_ifindex = nh->nh_oif;
1385
1386                         if (nh->nh_gw &&
1387                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1388                                 goto nla_put_failure;
1389 #ifdef CONFIG_IP_ROUTE_CLASSID
1390                         if (nh->nh_tclassid &&
1391                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1392                                 goto nla_put_failure;
1393 #endif
1394                         if (nh->nh_lwtstate &&
1395                             lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1396                                 goto nla_put_failure;
1397
1398                         /* length of rtnetlink header + attributes */
1399                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1400                 } endfor_nexthops(fi);
1401
1402                 nla_nest_end(skb, mp);
1403         }
1404 #endif
1405         nlmsg_end(skb, nlh);
1406         return 0;
1407
1408 nla_put_failure:
1409         nlmsg_cancel(skb, nlh);
1410         return -EMSGSIZE;
1411 }
1412
1413 /*
1414  * Update FIB if:
1415  * - local address disappeared -> we must delete all the entries
1416  *   referring to it.
1417  * - device went down -> we must shutdown all nexthops going via it.
1418  */
1419 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1420 {
1421         int ret = 0;
1422         unsigned int hash = fib_laddr_hashfn(local);
1423         struct hlist_head *head = &fib_info_laddrhash[hash];
1424         int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1425         struct net *net = dev_net(dev);
1426         struct fib_info *fi;
1427
1428         if (!fib_info_laddrhash || local == 0)
1429                 return 0;
1430
1431         hlist_for_each_entry(fi, head, fib_lhash) {
1432                 if (!net_eq(fi->fib_net, net) ||
1433                     fi->fib_tb_id != tb_id)
1434                         continue;
1435                 if (fi->fib_prefsrc == local) {
1436                         fi->fib_flags |= RTNH_F_DEAD;
1437                         ret++;
1438                 }
1439         }
1440         return ret;
1441 }
1442
1443 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1444                                  enum fib_event_type event_type)
1445 {
1446         struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1447         struct fib_nh_notifier_info info = {
1448                 .fib_nh = fib_nh,
1449         };
1450
1451         switch (event_type) {
1452         case FIB_EVENT_NH_ADD:
1453                 if (fib_nh->nh_flags & RTNH_F_DEAD)
1454                         break;
1455                 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1456                     fib_nh->nh_flags & RTNH_F_LINKDOWN)
1457                         break;
1458                 return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1459                                            &info.info);
1460         case FIB_EVENT_NH_DEL:
1461                 if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1462                      fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1463                     (fib_nh->nh_flags & RTNH_F_DEAD))
1464                         return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1465                                                    event_type, &info.info);
1466         default:
1467                 break;
1468         }
1469
1470         return NOTIFY_DONE;
1471 }
1472
1473 /* Update the PMTU of exceptions when:
1474  * - the new MTU of the first hop becomes smaller than the PMTU
1475  * - the old MTU was the same as the PMTU, and it limited discovery of
1476  *   larger MTUs on the path. With that limit raised, we can now
1477  *   discover larger MTUs
1478  * A special case is locked exceptions, for which the PMTU is smaller
1479  * than the minimal accepted PMTU:
1480  * - if the new MTU is greater than the PMTU, don't make any change
1481  * - otherwise, unlock and set PMTU
1482  */
1483 static void nh_update_mtu(struct fib_nh *nh, u32 new, u32 orig)
1484 {
1485         struct fnhe_hash_bucket *bucket;
1486         int i;
1487
1488         bucket = rcu_dereference_protected(nh->nh_exceptions, 1);
1489         if (!bucket)
1490                 return;
1491
1492         for (i = 0; i < FNHE_HASH_SIZE; i++) {
1493                 struct fib_nh_exception *fnhe;
1494
1495                 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1496                      fnhe;
1497                      fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1498                         if (fnhe->fnhe_mtu_locked) {
1499                                 if (new <= fnhe->fnhe_pmtu) {
1500                                         fnhe->fnhe_pmtu = new;
1501                                         fnhe->fnhe_mtu_locked = false;
1502                                 }
1503                         } else if (new < fnhe->fnhe_pmtu ||
1504                                    orig == fnhe->fnhe_pmtu) {
1505                                 fnhe->fnhe_pmtu = new;
1506                         }
1507                 }
1508         }
1509 }
1510
1511 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1512 {
1513         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1514         struct hlist_head *head = &fib_info_devhash[hash];
1515         struct fib_nh *nh;
1516
1517         hlist_for_each_entry(nh, head, nh_hash) {
1518                 if (nh->nh_dev == dev)
1519                         nh_update_mtu(nh, dev->mtu, orig_mtu);
1520         }
1521 }
1522
1523 /* Event              force Flags           Description
1524  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1525  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1526  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1527  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1528  */
1529 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1530 {
1531         int ret = 0;
1532         int scope = RT_SCOPE_NOWHERE;
1533         struct fib_info *prev_fi = NULL;
1534         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1535         struct hlist_head *head = &fib_info_devhash[hash];
1536         struct fib_nh *nh;
1537
1538         if (force)
1539                 scope = -1;
1540
1541         hlist_for_each_entry(nh, head, nh_hash) {
1542                 struct fib_info *fi = nh->nh_parent;
1543                 int dead;
1544
1545                 BUG_ON(!fi->fib_nhs);
1546                 if (nh->nh_dev != dev || fi == prev_fi)
1547                         continue;
1548                 prev_fi = fi;
1549                 dead = 0;
1550                 change_nexthops(fi) {
1551                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1552                                 dead++;
1553                         else if (nexthop_nh->nh_dev == dev &&
1554                                  nexthop_nh->nh_scope != scope) {
1555                                 switch (event) {
1556                                 case NETDEV_DOWN:
1557                                 case NETDEV_UNREGISTER:
1558                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1559                                         /* fall through */
1560                                 case NETDEV_CHANGE:
1561                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1562                                         break;
1563                                 }
1564                                 call_fib_nh_notifiers(nexthop_nh,
1565                                                       FIB_EVENT_NH_DEL);
1566                                 dead++;
1567                         }
1568 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1569                         if (event == NETDEV_UNREGISTER &&
1570                             nexthop_nh->nh_dev == dev) {
1571                                 dead = fi->fib_nhs;
1572                                 break;
1573                         }
1574 #endif
1575                 } endfor_nexthops(fi)
1576                 if (dead == fi->fib_nhs) {
1577                         switch (event) {
1578                         case NETDEV_DOWN:
1579                         case NETDEV_UNREGISTER:
1580                                 fi->fib_flags |= RTNH_F_DEAD;
1581                                 /* fall through */
1582                         case NETDEV_CHANGE:
1583                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1584                                 break;
1585                         }
1586                         ret++;
1587                 }
1588
1589                 fib_rebalance(fi);
1590         }
1591
1592         return ret;
1593 }
1594
1595 /* Must be invoked inside of an RCU protected region.  */
1596 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1597 {
1598         struct fib_info *fi = NULL, *last_resort = NULL;
1599         struct hlist_head *fa_head = res->fa_head;
1600         struct fib_table *tb = res->table;
1601         u8 slen = 32 - res->prefixlen;
1602         int order = -1, last_idx = -1;
1603         struct fib_alias *fa, *fa1 = NULL;
1604         u32 last_prio = res->fi->fib_priority;
1605         u8 last_tos = 0;
1606
1607         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1608                 struct fib_info *next_fi = fa->fa_info;
1609
1610                 if (fa->fa_slen != slen)
1611                         continue;
1612                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1613                         continue;
1614                 if (fa->tb_id != tb->tb_id)
1615                         continue;
1616                 if (next_fi->fib_priority > last_prio &&
1617                     fa->fa_tos == last_tos) {
1618                         if (last_tos)
1619                                 continue;
1620                         break;
1621                 }
1622                 if (next_fi->fib_flags & RTNH_F_DEAD)
1623                         continue;
1624                 last_tos = fa->fa_tos;
1625                 last_prio = next_fi->fib_priority;
1626
1627                 if (next_fi->fib_scope != res->scope ||
1628                     fa->fa_type != RTN_UNICAST)
1629                         continue;
1630                 if (!next_fi->fib_nh[0].nh_gw ||
1631                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1632                         continue;
1633
1634                 fib_alias_accessed(fa);
1635
1636                 if (!fi) {
1637                         if (next_fi != res->fi)
1638                                 break;
1639                         fa1 = fa;
1640                 } else if (!fib_detect_death(fi, order, &last_resort,
1641                                              &last_idx, fa1->fa_default)) {
1642                         fib_result_assign(res, fi);
1643                         fa1->fa_default = order;
1644                         goto out;
1645                 }
1646                 fi = next_fi;
1647                 order++;
1648         }
1649
1650         if (order <= 0 || !fi) {
1651                 if (fa1)
1652                         fa1->fa_default = -1;
1653                 goto out;
1654         }
1655
1656         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1657                               fa1->fa_default)) {
1658                 fib_result_assign(res, fi);
1659                 fa1->fa_default = order;
1660                 goto out;
1661         }
1662
1663         if (last_idx >= 0)
1664                 fib_result_assign(res, last_resort);
1665         fa1->fa_default = last_idx;
1666 out:
1667         return;
1668 }
1669
1670 /*
1671  * Dead device goes up. We wake up dead nexthops.
1672  * It takes sense only on multipath routes.
1673  */
1674 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1675 {
1676         struct fib_info *prev_fi;
1677         unsigned int hash;
1678         struct hlist_head *head;
1679         struct fib_nh *nh;
1680         int ret;
1681
1682         if (!(dev->flags & IFF_UP))
1683                 return 0;
1684
1685         if (nh_flags & RTNH_F_DEAD) {
1686                 unsigned int flags = dev_get_flags(dev);
1687
1688                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1689                         nh_flags |= RTNH_F_LINKDOWN;
1690         }
1691
1692         prev_fi = NULL;
1693         hash = fib_devindex_hashfn(dev->ifindex);
1694         head = &fib_info_devhash[hash];
1695         ret = 0;
1696
1697         hlist_for_each_entry(nh, head, nh_hash) {
1698                 struct fib_info *fi = nh->nh_parent;
1699                 int alive;
1700
1701                 BUG_ON(!fi->fib_nhs);
1702                 if (nh->nh_dev != dev || fi == prev_fi)
1703                         continue;
1704
1705                 prev_fi = fi;
1706                 alive = 0;
1707                 change_nexthops(fi) {
1708                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1709                                 alive++;
1710                                 continue;
1711                         }
1712                         if (!nexthop_nh->nh_dev ||
1713                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1714                                 continue;
1715                         if (nexthop_nh->nh_dev != dev ||
1716                             !__in_dev_get_rtnl(dev))
1717                                 continue;
1718                         alive++;
1719                         nexthop_nh->nh_flags &= ~nh_flags;
1720                         call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1721                 } endfor_nexthops(fi)
1722
1723                 if (alive > 0) {
1724                         fi->fib_flags &= ~nh_flags;
1725                         ret++;
1726                 }
1727
1728                 fib_rebalance(fi);
1729         }
1730
1731         return ret;
1732 }
1733
1734 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1735 static bool fib_good_nh(const struct fib_nh *nh)
1736 {
1737         int state = NUD_REACHABLE;
1738
1739         if (nh->nh_scope == RT_SCOPE_LINK) {
1740                 struct neighbour *n;
1741
1742                 rcu_read_lock_bh();
1743
1744                 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1745                                               (__force u32)nh->nh_gw);
1746                 if (n)
1747                         state = n->nud_state;
1748
1749                 rcu_read_unlock_bh();
1750         }
1751
1752         return !!(state & NUD_VALID);
1753 }
1754
1755 void fib_select_multipath(struct fib_result *res, int hash)
1756 {
1757         struct fib_info *fi = res->fi;
1758         struct net *net = fi->fib_net;
1759         bool first = false;
1760
1761         for_nexthops(fi) {
1762                 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
1763                         if (!fib_good_nh(nh))
1764                                 continue;
1765                         if (!first) {
1766                                 res->nh_sel = nhsel;
1767                                 first = true;
1768                         }
1769                 }
1770
1771                 if (hash > atomic_read(&nh->nh_upper_bound))
1772                         continue;
1773
1774                 res->nh_sel = nhsel;
1775                 return;
1776         } endfor_nexthops(fi);
1777 }
1778 #endif
1779
1780 void fib_select_path(struct net *net, struct fib_result *res,
1781                      struct flowi4 *fl4, const struct sk_buff *skb)
1782 {
1783         if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
1784                 goto check_saddr;
1785
1786 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1787         if (res->fi->fib_nhs > 1) {
1788                 int h = fib_multipath_hash(net, fl4, skb, NULL);
1789
1790                 fib_select_multipath(res, h);
1791         }
1792         else
1793 #endif
1794         if (!res->prefixlen &&
1795             res->table->tb_num_default > 1 &&
1796             res->type == RTN_UNICAST)
1797                 fib_select_default(fl4, res);
1798
1799 check_saddr:
1800         if (!fl4->saddr)
1801                 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1802 }