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