net/ipv4/fib_semantics.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              IPv4 Forwarding Information Base: semantics.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  */
  11
  12 #include <linux/uaccess.h>
  13 #include <linux/bitops.h>
  14 #include <linux/types.h>
  15 #include <linux/kernel.h>
  16 #include <linux/jiffies.h>
  17 #include <linux/mm.h>
  18 #include <linux/string.h>
  19 #include <linux/socket.h>
  20 #include <linux/sockios.h>
  21 #include <linux/errno.h>
  22 #include <linux/in.h>
  23 #include <linux/inet.h>
  24 #include <linux/inetdevice.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/if_arp.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/netlink.h>
  32 #include <linux/hash.h>
  33
  34 #include <net/arp.h>
  35 #include <net/ip.h>
  36 #include <net/protocol.h>
  37 #include <net/route.h>
  38 #include <net/tcp.h>
  39 #include <net/sock.h>
  40 #include <net/ip_fib.h>
  41 #include <net/ip6_fib.h>
  42 #include <net/nexthop.h>
  43 #include <net/netlink.h>
  44 #include <net/rtnh.h>
  45 #include <net/lwtunnel.h>
  46 #include <net/fib_notifier.h>
  47 #include <net/addrconf.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 /* for_nexthops and change_nexthops only used when nexthop object
  62  * is not set in a fib_info. The logic within can reference fib_nh.
  63  */
  64 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  65
  66 #define for_nexthops(fi) {                                              \
  67         int nhsel; const struct fib_nh *nh;                             \
  68         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  69              nhsel < fib_info_num_path((fi));                           \
  70              nh++, nhsel++)
  71
  72 #define change_nexthops(fi) {                                           \
  73         int nhsel; struct fib_nh *nexthop_nh;                           \
  74         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  75              nhsel < fib_info_num_path((fi));                           \
  76              nexthop_nh++, nhsel++)
  77
  78 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  79
  80 /* Hope, that gcc will optimize it to get rid of dummy loop */
  81
  82 #define for_nexthops(fi) {                                              \
  83         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  84         for (nhsel = 0; nhsel < 1; nhsel++)
  85
  86 #define change_nexthops(fi) {                                           \
  87         int nhsel;                                                      \
  88         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  89         for (nhsel = 0; nhsel < 1; nhsel++)
  90
  91 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  92
  93 #define endfor_nexthops(fi) }
  94
  95
  96 const struct fib_prop fib_props[RTN_MAX + 1] = {
  97         [RTN_UNSPEC] = {
  98                 .error  = 0,
  99                 .scope  = RT_SCOPE_NOWHERE,
 100         },
 101         [RTN_UNICAST] = {
 102                 .error  = 0,
 103                 .scope  = RT_SCOPE_UNIVERSE,
 104         },
 105         [RTN_LOCAL] = {
 106                 .error  = 0,
 107                 .scope  = RT_SCOPE_HOST,
 108         },
 109         [RTN_BROADCAST] = {
 110                 .error  = 0,
 111                 .scope  = RT_SCOPE_LINK,
 112         },
 113         [RTN_ANYCAST] = {
 114                 .error  = 0,
 115                 .scope  = RT_SCOPE_LINK,
 116         },
 117         [RTN_MULTICAST] = {
 118                 .error  = 0,
 119                 .scope  = RT_SCOPE_UNIVERSE,
 120         },
 121         [RTN_BLACKHOLE] = {
 122                 .error  = -EINVAL,
 123                 .scope  = RT_SCOPE_UNIVERSE,
 124         },
 125         [RTN_UNREACHABLE] = {
 126                 .error  = -EHOSTUNREACH,
 127                 .scope  = RT_SCOPE_UNIVERSE,
 128         },
 129         [RTN_PROHIBIT] = {
 130                 .error  = -EACCES,
 131                 .scope  = RT_SCOPE_UNIVERSE,
 132         },
 133         [RTN_THROW] = {
 134                 .error  = -EAGAIN,
 135                 .scope  = RT_SCOPE_UNIVERSE,
 136         },
 137         [RTN_NAT] = {
 138                 .error  = -EINVAL,
 139                 .scope  = RT_SCOPE_NOWHERE,
 140         },
 141         [RTN_XRESOLVE] = {
 142                 .error  = -EINVAL,
 143                 .scope  = RT_SCOPE_NOWHERE,
 144         },
 145 };
 146
 147 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 148 {
 149         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 150
 151         if (!rt)
 152                 return;
 153
 154         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 155          * because we waited an RCU grace period before calling
 156          * free_fib_info_rcu()
 157          */
 158
 159         dst_dev_put(&rt->dst);
 160         dst_release_immediate(&rt->dst);
 161 }
 162
 163 static void free_nh_exceptions(struct fib_nh_common *nhc)
 164 {
 165         struct fnhe_hash_bucket *hash;
 166         int i;
 167
 168         hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
 169         if (!hash)
 170                 return;
 171         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 172                 struct fib_nh_exception *fnhe;
 173
 174                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 175                 while (fnhe) {
 176                         struct fib_nh_exception *next;
 177
 178                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 179
 180                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 181                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 182
 183                         kfree(fnhe);
 184
 185                         fnhe = next;
 186                 }
 187         }
 188         kfree(hash);
 189 }
 190
 191 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 192 {
 193         int cpu;
 194
 195         if (!rtp)
 196                 return;
 197
 198         for_each_possible_cpu(cpu) {
 199                 struct rtable *rt;
 200
 201                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 202                 if (rt) {
 203                         dst_dev_put(&rt->dst);
 204                         dst_release_immediate(&rt->dst);
 205                 }
 206         }
 207         free_percpu(rtp);
 208 }
 209
 210 void fib_nh_common_release(struct fib_nh_common *nhc)
 211 {
 212         if (nhc->nhc_dev)
 213                 dev_put(nhc->nhc_dev);
 214
 215         lwtstate_put(nhc->nhc_lwtstate);
 216         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 217         rt_fibinfo_free(&nhc->nhc_rth_input);
 218         free_nh_exceptions(nhc);
 219 }
 220 EXPORT_SYMBOL_GPL(fib_nh_common_release);
 221
 222 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
 223 {
 224 #ifdef CONFIG_IP_ROUTE_CLASSID
 225         if (fib_nh->nh_tclassid)
 226                 atomic_dec(&net->ipv4.fib_num_tclassid_users);
 227 #endif
 228         fib_nh_common_release(&fib_nh->nh_common);
 229 }
 230
 231 /* Release a nexthop info record */
 232 static void free_fib_info_rcu(struct rcu_head *head)
 233 {
 234         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 235
 236         if (fi->nh) {
 237                 nexthop_put(fi->nh);
 238         } else {
 239                 change_nexthops(fi) {
 240                         fib_nh_release(fi->fib_net, nexthop_nh);
 241                 } endfor_nexthops(fi);
 242         }
 243
 244         ip_fib_metrics_put(fi->fib_metrics);
 245
 246         kfree(fi);
 247 }
 248
 249 void free_fib_info(struct fib_info *fi)
 250 {
 251         if (fi->fib_dead == 0) {
 252                 pr_warn("Freeing alive fib_info %p\n", fi);
 253                 return;
 254         }
 255
 256         call_rcu(&fi->rcu, free_fib_info_rcu);
 257 }
 258 EXPORT_SYMBOL_GPL(free_fib_info);
 259
 260 void fib_release_info(struct fib_info *fi)
 261 {
 262         spin_lock_bh(&fib_info_lock);
 263         if (fi && --fi->fib_treeref == 0) {
 264                 hlist_del(&fi->fib_hash);
 265
 266                 /* Paired with READ_ONCE() in fib_create_info(). */
 267                 WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
 268
 269                 if (fi->fib_prefsrc)
 270                         hlist_del(&fi->fib_lhash);
 271                 if (fi->nh) {
 272                         list_del(&fi->nh_list);
 273                 } else {
 274                         change_nexthops(fi) {
 275                                 if (!nexthop_nh->fib_nh_dev)
 276                                         continue;
 277                                 hlist_del(&nexthop_nh->nh_hash);
 278                         } endfor_nexthops(fi)
 279                 }
 280                 fi->fib_dead = 1;
 281                 fib_info_put(fi);
 282         }
 283         spin_unlock_bh(&fib_info_lock);
 284 }
 285
 286 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
 287 {
 288         const struct fib_nh *onh;
 289
 290         if (fi->nh || ofi->nh)
 291                 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
 292
 293         if (ofi->fib_nhs == 0)
 294                 return 0;
 295
 296         for_nexthops(fi) {
 297                 onh = fib_info_nh(ofi, nhsel);
 298
 299                 if (nh->fib_nh_oif != onh->fib_nh_oif ||
 300                     nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
 301                     nh->fib_nh_scope != onh->fib_nh_scope ||
 302 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 303                     nh->fib_nh_weight != onh->fib_nh_weight ||
 304 #endif
 305 #ifdef CONFIG_IP_ROUTE_CLASSID
 306                     nh->nh_tclassid != onh->nh_tclassid ||
 307 #endif
 308                     lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
 309                     ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
 310                         return -1;
 311
 312                 if (nh->fib_nh_gw_family == AF_INET &&
 313                     nh->fib_nh_gw4 != onh->fib_nh_gw4)
 314                         return -1;
 315
 316                 if (nh->fib_nh_gw_family == AF_INET6 &&
 317                     ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
 318                         return -1;
 319         } endfor_nexthops(fi);
 320         return 0;
 321 }
 322
 323 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 324 {
 325         return hash_32(val, DEVINDEX_HASHBITS);
 326 }
 327
 328 static struct hlist_head *
 329 fib_info_devhash_bucket(const struct net_device *dev)
 330 {
 331         u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
 332
 333         return &fib_info_devhash[fib_devindex_hashfn(val)];
 334 }
 335
 336 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
 337                                       u32 prefsrc, u32 priority)
 338 {
 339         unsigned int val = init_val;
 340
 341         val ^= (protocol << 8) | scope;
 342         val ^= prefsrc;
 343         val ^= priority;
 344
 345         return val;
 346 }
 347
 348 static unsigned int fib_info_hashfn_result(unsigned int val)
 349 {
 350         unsigned int mask = (fib_info_hash_size - 1);
 351
 352         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 353 }
 354
 355 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
 356 {
 357         unsigned int val;
 358
 359         val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
 360                                 fi->fib_scope, (__force u32)fi->fib_prefsrc,
 361                                 fi->fib_priority);
 362
 363         if (fi->nh) {
 364                 val ^= fib_devindex_hashfn(fi->nh->id);
 365         } else {
 366                 for_nexthops(fi) {
 367                         val ^= fib_devindex_hashfn(nh->fib_nh_oif);
 368                 } endfor_nexthops(fi)
 369         }
 370
 371         return fib_info_hashfn_result(val);
 372 }
 373
 374 /* no metrics, only nexthop id */
 375 static struct fib_info *fib_find_info_nh(struct net *net,
 376                                          const struct fib_config *cfg)
 377 {
 378         struct hlist_head *head;
 379         struct fib_info *fi;
 380         unsigned int hash;
 381
 382         hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
 383                                  cfg->fc_protocol, cfg->fc_scope,
 384                                  (__force u32)cfg->fc_prefsrc,
 385                                  cfg->fc_priority);
 386         hash = fib_info_hashfn_result(hash);
 387         head = &fib_info_hash[hash];
 388
 389         hlist_for_each_entry(fi, head, fib_hash) {
 390                 if (!net_eq(fi->fib_net, net))
 391                         continue;
 392                 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
 393                         continue;
 394                 if (cfg->fc_protocol == fi->fib_protocol &&
 395                     cfg->fc_scope == fi->fib_scope &&
 396                     cfg->fc_prefsrc == fi->fib_prefsrc &&
 397                     cfg->fc_priority == fi->fib_priority &&
 398                     cfg->fc_type == fi->fib_type &&
 399                     cfg->fc_table == fi->fib_tb_id &&
 400                     !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
 401                         return fi;
 402         }
 403
 404         return NULL;
 405 }
 406
 407 static struct fib_info *fib_find_info(struct fib_info *nfi)
 408 {
 409         struct hlist_head *head;
 410         struct fib_info *fi;
 411         unsigned int hash;
 412
 413         hash = fib_info_hashfn(nfi);
 414         head = &fib_info_hash[hash];
 415
 416         hlist_for_each_entry(fi, head, fib_hash) {
 417                 if (!net_eq(fi->fib_net, nfi->fib_net))
 418                         continue;
 419                 if (fi->fib_nhs != nfi->fib_nhs)
 420                         continue;
 421                 if (nfi->fib_protocol == fi->fib_protocol &&
 422                     nfi->fib_scope == fi->fib_scope &&
 423                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 424                     nfi->fib_priority == fi->fib_priority &&
 425                     nfi->fib_type == fi->fib_type &&
 426                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 427                            sizeof(u32) * RTAX_MAX) == 0 &&
 428                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 429                     nh_comp(fi, nfi) == 0)
 430                         return fi;
 431         }
 432
 433         return NULL;
 434 }
 435
 436 /* Check, that the gateway is already configured.
 437  * Used only by redirect accept routine.
 438  */
 439 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 440 {
 441         struct hlist_head *head;
 442         struct fib_nh *nh;
 443
 444         spin_lock(&fib_info_lock);
 445
 446         head = fib_info_devhash_bucket(dev);
 447
 448         hlist_for_each_entry(nh, head, nh_hash) {
 449                 if (nh->fib_nh_dev == dev &&
 450                     nh->fib_nh_gw4 == gw &&
 451                     !(nh->fib_nh_flags & RTNH_F_DEAD)) {
 452                         spin_unlock(&fib_info_lock);
 453                         return 0;
 454                 }
 455         }
 456
 457         spin_unlock(&fib_info_lock);
 458
 459         return -1;
 460 }
 461
 462 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 463 {
 464         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 465                          + nla_total_size(4) /* RTA_TABLE */
 466                          + nla_total_size(4) /* RTA_DST */
 467                          + nla_total_size(4) /* RTA_PRIORITY */
 468                          + nla_total_size(4) /* RTA_PREFSRC */
 469                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 470         unsigned int nhs = fib_info_num_path(fi);
 471
 472         /* space for nested metrics */
 473         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 474
 475         if (fi->nh)
 476                 payload += nla_total_size(4); /* RTA_NH_ID */
 477
 478         if (nhs) {
 479                 size_t nh_encapsize = 0;
 480                 /* Also handles the special case nhs == 1 */
 481
 482                 /* each nexthop is packed in an attribute */
 483                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 484                 unsigned int i;
 485
 486                 /* may contain flow and gateway attribute */
 487                 nhsize += 2 * nla_total_size(4);
 488
 489                 /* grab encap info */
 490                 for (i = 0; i < fib_info_num_path(fi); i++) {
 491                         struct fib_nh_common *nhc = fib_info_nhc(fi, i);
 492
 493                         if (nhc->nhc_lwtstate) {
 494                                 /* RTA_ENCAP_TYPE */
 495                                 nh_encapsize += lwtunnel_get_encap_size(
 496                                                 nhc->nhc_lwtstate);
 497                                 /* RTA_ENCAP */
 498                                 nh_encapsize +=  nla_total_size(2);
 499                         }
 500                 }
 501
 502                 /* all nexthops are packed in a nested attribute */
 503                 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
 504
 505         }
 506
 507         return payload;
 508 }
 509
 510 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 511                int dst_len, u32 tb_id, const struct nl_info *info,
 512                unsigned int nlm_flags)
 513 {
 514         struct fib_rt_info fri;
 515         struct sk_buff *skb;
 516         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 517         int err = -ENOBUFS;
 518
 519         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 520         if (!skb)
 521                 goto errout;
 522
 523         fri.fi = fa->fa_info;
 524         fri.tb_id = tb_id;
 525         fri.dst = key;
 526         fri.dst_len = dst_len;
 527         fri.tos = fa->fa_tos;
 528         fri.type = fa->fa_type;
 529         fri.offload = fa->offload;
 530         fri.trap = fa->trap;
 531         err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
 532         if (err < 0) {
 533                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 534                 WARN_ON(err == -EMSGSIZE);
 535                 kfree_skb(skb);
 536                 goto errout;
 537         }
 538         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 539                     info->nlh, GFP_KERNEL);
 540         return;
 541 errout:
 542         if (err < 0)
 543                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 544 }
 545
 546 static int fib_detect_death(struct fib_info *fi, int order,
 547                             struct fib_info **last_resort, int *last_idx,
 548                             int dflt)
 549 {
 550         const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
 551         struct neighbour *n;
 552         int state = NUD_NONE;
 553
 554         if (likely(nhc->nhc_gw_family == AF_INET))
 555                 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
 556         else if (nhc->nhc_gw_family == AF_INET6)
 557                 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
 558                                  nhc->nhc_dev);
 559         else
 560                 n = NULL;
 561
 562         if (n) {
 563                 state = n->nud_state;
 564                 neigh_release(n);
 565         } else {
 566                 return 0;
 567         }
 568         if (state == NUD_REACHABLE)
 569                 return 0;
 570         if ((state & NUD_VALID) && order != dflt)
 571                 return 0;
 572         if ((state & NUD_VALID) ||
 573             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 574                 *last_resort = fi;
 575                 *last_idx = order;
 576         }
 577         return 1;
 578 }
 579
 580 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
 581                        struct nlattr *encap, u16 encap_type,
 582                        void *cfg, gfp_t gfp_flags,
 583                        struct netlink_ext_ack *extack)
 584 {
 585         int err;
 586
 587         nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
 588                                                     gfp_flags);
 589         if (!nhc->nhc_pcpu_rth_output)
 590                 return -ENOMEM;
 591
 592         if (encap) {
 593                 struct lwtunnel_state *lwtstate;
 594
 595                 if (encap_type == LWTUNNEL_ENCAP_NONE) {
 596                         NL_SET_ERR_MSG(extack, "LWT encap type not specified");
 597                         err = -EINVAL;
 598                         goto lwt_failure;
 599                 }
 600                 err = lwtunnel_build_state(net, encap_type, encap,
 601                                            nhc->nhc_family, cfg, &lwtstate,
 602                                            extack);
 603                 if (err)
 604                         goto lwt_failure;
 605
 606                 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
 607         }
 608
 609         return 0;
 610
 611 lwt_failure:
 612         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 613         nhc->nhc_pcpu_rth_output = NULL;
 614         return err;
 615 }
 616 EXPORT_SYMBOL_GPL(fib_nh_common_init);
 617
 618 int fib_nh_init(struct net *net, struct fib_nh *nh,
 619                 struct fib_config *cfg, int nh_weight,
 620                 struct netlink_ext_ack *extack)
 621 {
 622         int err;
 623
 624         nh->fib_nh_family = AF_INET;
 625
 626         err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
 627                                  cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
 628         if (err)
 629                 return err;
 630
 631         nh->fib_nh_oif = cfg->fc_oif;
 632         nh->fib_nh_gw_family = cfg->fc_gw_family;
 633         if (cfg->fc_gw_family == AF_INET)
 634                 nh->fib_nh_gw4 = cfg->fc_gw4;
 635         else if (cfg->fc_gw_family == AF_INET6)
 636                 nh->fib_nh_gw6 = cfg->fc_gw6;
 637
 638         nh->fib_nh_flags = cfg->fc_flags;
 639
 640 #ifdef CONFIG_IP_ROUTE_CLASSID
 641         nh->nh_tclassid = cfg->fc_flow;
 642         if (nh->nh_tclassid)
 643                 atomic_inc(&net->ipv4.fib_num_tclassid_users);
 644 #endif
 645 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 646         nh->fib_nh_weight = nh_weight;
 647 #endif
 648         return 0;
 649 }
 650
 651 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 652
 653 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
 654                               struct netlink_ext_ack *extack)
 655 {
 656         int nhs = 0;
 657
 658         while (rtnh_ok(rtnh, remaining)) {
 659                 nhs++;
 660                 rtnh = rtnh_next(rtnh, &remaining);
 661         }
 662
 663         /* leftover implies invalid nexthop configuration, discard it */
 664         if (remaining > 0) {
 665                 NL_SET_ERR_MSG(extack,
 666                                "Invalid nexthop configuration - extra data after nexthops");
 667                 nhs = 0;
 668         }
 669
 670         return nhs;
 671 }
 672
 673 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
 674                             struct netlink_ext_ack *extack)
 675 {
 676         if (nla_len(nla) < sizeof(*gw)) {
 677                 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
 678                 return -EINVAL;
 679         }
 680
 681         *gw = nla_get_in_addr(nla);
 682
 683         return 0;
 684 }
 685
 686 /* only called when fib_nh is integrated into fib_info */
 687 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 688                        int remaining, struct fib_config *cfg,
 689                        struct netlink_ext_ack *extack)
 690 {
 691         struct net *net = fi->fib_net;
 692         struct fib_config fib_cfg;
 693         struct fib_nh *nh;
 694         int ret;
 695
 696         change_nexthops(fi) {
 697                 int attrlen;
 698
 699                 memset(&fib_cfg, 0, sizeof(fib_cfg));
 700
 701                 if (!rtnh_ok(rtnh, remaining)) {
 702                         NL_SET_ERR_MSG(extack,
 703                                        "Invalid nexthop configuration - extra data after nexthop");
 704                         return -EINVAL;
 705                 }
 706
 707                 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
 708                         NL_SET_ERR_MSG(extack,
 709                                        "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
 710                         return -EINVAL;
 711                 }
 712
 713                 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 714                 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
 715
 716                 attrlen = rtnh_attrlen(rtnh);
 717                 if (attrlen > 0) {
 718                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 719
 720                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 721                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 722                         if (nla && nlav) {
 723                                 NL_SET_ERR_MSG(extack,
 724                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 725                                 return -EINVAL;
 726                         }
 727                         if (nla) {
 728                                 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
 729                                                        extack);
 730                                 if (ret)
 731                                         goto errout;
 732
 733                                 if (fib_cfg.fc_gw4)
 734                                         fib_cfg.fc_gw_family = AF_INET;
 735                         } else if (nlav) {
 736                                 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
 737                                 if (ret)
 738                                         goto errout;
 739                         }
 740
 741                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 742                         if (nla) {
 743                                 if (nla_len(nla) < sizeof(u32)) {
 744                                         NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
 745                                         return -EINVAL;
 746                                 }
 747                                 fib_cfg.fc_flow = nla_get_u32(nla);
 748                         }
 749
 750                         fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
 751                         /* RTA_ENCAP_TYPE length checked in
 752                          * lwtunnel_valid_encap_type_attr
 753                          */
 754                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
 755                         if (nla)
 756                                 fib_cfg.fc_encap_type = nla_get_u16(nla);
 757                 }
 758
 759                 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
 760                                   rtnh->rtnh_hops + 1, extack);
 761                 if (ret)
 762                         goto errout;
 763
 764                 rtnh = rtnh_next(rtnh, &remaining);
 765         } endfor_nexthops(fi);
 766
 767         ret = -EINVAL;
 768         nh = fib_info_nh(fi, 0);
 769         if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
 770                 NL_SET_ERR_MSG(extack,
 771                                "Nexthop device index does not match RTA_OIF");
 772                 goto errout;
 773         }
 774         if (cfg->fc_gw_family) {
 775                 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
 776                     (cfg->fc_gw_family == AF_INET &&
 777                      nh->fib_nh_gw4 != cfg->fc_gw4) ||
 778                     (cfg->fc_gw_family == AF_INET6 &&
 779                      ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
 780                         NL_SET_ERR_MSG(extack,
 781                                        "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
 782                         goto errout;
 783                 }
 784         }
 785 #ifdef CONFIG_IP_ROUTE_CLASSID
 786         if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
 787                 NL_SET_ERR_MSG(extack,
 788                                "Nexthop class id does not match RTA_FLOW");
 789                 goto errout;
 790         }
 791 #endif
 792         ret = 0;
 793 errout:
 794         return ret;
 795 }
 796
 797 /* only called when fib_nh is integrated into fib_info */
 798 static void fib_rebalance(struct fib_info *fi)
 799 {
 800         int total;
 801         int w;
 802
 803         if (fib_info_num_path(fi) < 2)
 804                 return;
 805
 806         total = 0;
 807         for_nexthops(fi) {
 808                 if (nh->fib_nh_flags & RTNH_F_DEAD)
 809                         continue;
 810
 811                 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
 812                     nh->fib_nh_flags & RTNH_F_LINKDOWN)
 813                         continue;
 814
 815                 total += nh->fib_nh_weight;
 816         } endfor_nexthops(fi);
 817
 818         w = 0;
 819         change_nexthops(fi) {
 820                 int upper_bound;
 821
 822                 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
 823                         upper_bound = -1;
 824                 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
 825                            nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
 826                         upper_bound = -1;
 827                 } else {
 828                         w += nexthop_nh->fib_nh_weight;
 829                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 830                                                             total) - 1;
 831                 }
 832
 833                 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
 834         } endfor_nexthops(fi);
 835 }
 836 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 837
 838 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 839                        int remaining, struct fib_config *cfg,
 840                        struct netlink_ext_ack *extack)
 841 {
 842         NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
 843
 844         return -EINVAL;
 845 }
 846
 847 #define fib_rebalance(fi) do { } while (0)
 848
 849 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 850
 851 static int fib_encap_match(struct net *net, u16 encap_type,
 852                            struct nlattr *encap,
 853                            const struct fib_nh *nh,
 854                            const struct fib_config *cfg,
 855                            struct netlink_ext_ack *extack)
 856 {
 857         struct lwtunnel_state *lwtstate;
 858         int ret, result = 0;
 859
 860         if (encap_type == LWTUNNEL_ENCAP_NONE)
 861                 return 0;
 862
 863         ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
 864                                    cfg, &lwtstate, extack);
 865         if (!ret) {
 866                 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
 867                 lwtstate_free(lwtstate);
 868         }
 869
 870         return result;
 871 }
 872
 873 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
 874                  struct netlink_ext_ack *extack)
 875 {
 876 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 877         struct rtnexthop *rtnh;
 878         int remaining;
 879 #endif
 880
 881         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 882                 return 1;
 883
 884         if (cfg->fc_nh_id) {
 885                 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
 886                         return 0;
 887                 return 1;
 888         }
 889
 890         if (cfg->fc_oif || cfg->fc_gw_family) {
 891                 struct fib_nh *nh;
 892
 893                 /* cannot match on nexthop object attributes */
 894                 if (fi->nh)
 895                         return 1;
 896
 897                 nh = fib_info_nh(fi, 0);
 898                 if (cfg->fc_encap) {
 899                         if (fib_encap_match(net, cfg->fc_encap_type,
 900                                             cfg->fc_encap, nh, cfg, extack))
 901                                 return 1;
 902                 }
 903 #ifdef CONFIG_IP_ROUTE_CLASSID
 904                 if (cfg->fc_flow &&
 905                     cfg->fc_flow != nh->nh_tclassid)
 906                         return 1;
 907 #endif
 908                 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
 909                     (cfg->fc_gw_family &&
 910                      cfg->fc_gw_family != nh->fib_nh_gw_family))
 911                         return 1;
 912
 913                 if (cfg->fc_gw_family == AF_INET &&
 914                     cfg->fc_gw4 != nh->fib_nh_gw4)
 915                         return 1;
 916
 917                 if (cfg->fc_gw_family == AF_INET6 &&
 918                     ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
 919                         return 1;
 920
 921                 return 0;
 922         }
 923
 924 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 925         if (!cfg->fc_mp)
 926                 return 0;
 927
 928         rtnh = cfg->fc_mp;
 929         remaining = cfg->fc_mp_len;
 930
 931         for_nexthops(fi) {
 932                 int attrlen;
 933
 934                 if (!rtnh_ok(rtnh, remaining))
 935                         return -EINVAL;
 936
 937                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
 938                         return 1;
 939
 940                 attrlen = rtnh_attrlen(rtnh);
 941                 if (attrlen > 0) {
 942                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 943                         int err;
 944
 945                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 946                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 947                         if (nla && nlav) {
 948                                 NL_SET_ERR_MSG(extack,
 949                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 950                                 return -EINVAL;
 951                         }
 952
 953                         if (nla) {
 954                                 __be32 gw;
 955
 956                                 err = fib_gw_from_attr(&gw, nla, extack);
 957                                 if (err)
 958                                         return err;
 959
 960                                 if (nh->fib_nh_gw_family != AF_INET ||
 961                                     gw != nh->fib_nh_gw4)
 962                                         return 1;
 963                         } else if (nlav) {
 964                                 struct fib_config cfg2;
 965
 966                                 err = fib_gw_from_via(&cfg2, nlav, extack);
 967                                 if (err)
 968                                         return err;
 969
 970                                 switch (nh->fib_nh_gw_family) {
 971                                 case AF_INET:
 972                                         if (cfg2.fc_gw_family != AF_INET ||
 973                                             cfg2.fc_gw4 != nh->fib_nh_gw4)
 974                                                 return 1;
 975                                         break;
 976                                 case AF_INET6:
 977                                         if (cfg2.fc_gw_family != AF_INET6 ||
 978                                             ipv6_addr_cmp(&cfg2.fc_gw6,
 979                                                           &nh->fib_nh_gw6))
 980                                                 return 1;
 981                                         break;
 982                                 }
 983                         }
 984
 985 #ifdef CONFIG_IP_ROUTE_CLASSID
 986                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 987                         if (nla) {
 988                                 if (nla_len(nla) < sizeof(u32)) {
 989                                         NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
 990                                         return -EINVAL;
 991                                 }
 992                                 if (nla_get_u32(nla) != nh->nh_tclassid)
 993                                         return 1;
 994                         }
 995 #endif
 996                 }
 997
 998                 rtnh = rtnh_next(rtnh, &remaining);
 999         } endfor_nexthops(fi);
1000 #endif
1001         return 0;
1002 }
1003
1004 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1005 {
1006         struct nlattr *nla;
1007         int remaining;
1008
1009         if (!cfg->fc_mx)
1010                 return true;
1011
1012         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1013                 int type = nla_type(nla);
1014                 u32 fi_val, val;
1015
1016                 if (!type)
1017                         continue;
1018                 if (type > RTAX_MAX)
1019                         return false;
1020
1021                 if (type == RTAX_CC_ALGO) {
1022                         char tmp[TCP_CA_NAME_MAX];
1023                         bool ecn_ca = false;
1024
1025                         nla_strlcpy(tmp, nla, sizeof(tmp));
1026                         val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1027                 } else {
1028                         if (nla_len(nla) != sizeof(u32))
1029                                 return false;
1030                         val = nla_get_u32(nla);
1031                 }
1032
1033                 fi_val = fi->fib_metrics->metrics[type - 1];
1034                 if (type == RTAX_FEATURES)
1035                         fi_val &= ~DST_FEATURE_ECN_CA;
1036
1037                 if (fi_val != val)
1038                         return false;
1039         }
1040
1041         return true;
1042 }
1043
1044 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1045                               u32 table, struct netlink_ext_ack *extack)
1046 {
1047         struct fib6_config cfg = {
1048                 .fc_table = table,
1049                 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1050                 .fc_ifindex = nh->fib_nh_oif,
1051                 .fc_gateway = nh->fib_nh_gw6,
1052         };
1053         struct fib6_nh fib6_nh = {};
1054         int err;
1055
1056         err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1057         if (!err) {
1058                 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1059                 dev_hold(nh->fib_nh_dev);
1060                 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1061                 nh->fib_nh_scope = RT_SCOPE_LINK;
1062
1063                 ipv6_stub->fib6_nh_release(&fib6_nh);
1064         }
1065
1066         return err;
1067 }
1068
1069 /*
1070  * Picture
1071  * -------
1072  *
1073  * Semantics of nexthop is very messy by historical reasons.
1074  * We have to take into account, that:
1075  * a) gateway can be actually local interface address,
1076  *    so that gatewayed route is direct.
1077  * b) gateway must be on-link address, possibly
1078  *    described not by an ifaddr, but also by a direct route.
1079  * c) If both gateway and interface are specified, they should not
1080  *    contradict.
1081  * d) If we use tunnel routes, gateway could be not on-link.
1082  *
1083  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1084  * results in pretty ugly and hairy code with obscure logic.
1085  *
1086  * I chose to generalized it instead, so that the size
1087  * of code does not increase practically, but it becomes
1088  * much more general.
1089  * Every prefix is assigned a "scope" value: "host" is local address,
1090  * "link" is direct route,
1091  * [ ... "site" ... "interior" ... ]
1092  * and "universe" is true gateway route with global meaning.
1093  *
1094  * Every prefix refers to a set of "nexthop"s (gw, oif),
1095  * where gw must have narrower scope. This recursion stops
1096  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1097  * which means that gw is forced to be on link.
1098  *
1099  * Code is still hairy, but now it is apparently logically
1100  * consistent and very flexible. F.e. as by-product it allows
1101  * to co-exists in peace independent exterior and interior
1102  * routing processes.
1103  *
1104  * Normally it looks as following.
1105  *
1106  * {universe prefix}  -> (gw, oif) [scope link]
1107  *                |
1108  *                |-> {link prefix} -> (gw, oif) [scope local]
1109  *                                      |
1110  *                                      |-> {local prefix} (terminal node)
1111  */
1112 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1113                               u8 scope, struct netlink_ext_ack *extack)
1114 {
1115         struct net_device *dev;
1116         struct fib_result res;
1117         int err = 0;
1118
1119         if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1120                 unsigned int addr_type;
1121
1122                 if (scope >= RT_SCOPE_LINK) {
1123                         NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1124                         return -EINVAL;
1125                 }
1126                 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1127                 if (!dev) {
1128                         NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1129                         return -ENODEV;
1130                 }
1131                 if (!(dev->flags & IFF_UP)) {
1132                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1133                         return -ENETDOWN;
1134                 }
1135                 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1136                 if (addr_type != RTN_UNICAST) {
1137                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1138                         return -EINVAL;
1139                 }
1140                 if (!netif_carrier_ok(dev))
1141                         nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1142                 nh->fib_nh_dev = dev;
1143                 dev_hold(dev);
1144                 nh->fib_nh_scope = RT_SCOPE_LINK;
1145                 return 0;
1146         }
1147         rcu_read_lock();
1148         {
1149                 struct fib_table *tbl = NULL;
1150                 struct flowi4 fl4 = {
1151                         .daddr = nh->fib_nh_gw4,
1152                         .flowi4_scope = scope + 1,
1153                         .flowi4_oif = nh->fib_nh_oif,
1154                         .flowi4_iif = LOOPBACK_IFINDEX,
1155                 };
1156
1157                 /* It is not necessary, but requires a bit of thinking */
1158                 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1159                         fl4.flowi4_scope = RT_SCOPE_LINK;
1160
1161                 if (table && table != RT_TABLE_MAIN)
1162                         tbl = fib_get_table(net, table);
1163
1164                 if (tbl)
1165                         err = fib_table_lookup(tbl, &fl4, &res,
1166                                                FIB_LOOKUP_IGNORE_LINKSTATE |
1167                                                FIB_LOOKUP_NOREF);
1168
1169                 /* on error or if no table given do full lookup. This
1170                  * is needed for example when nexthops are in the local
1171                  * table rather than the given table
1172                  */
1173                 if (!tbl || err) {
1174                         err = fib_lookup(net, &fl4, &res,
1175                                          FIB_LOOKUP_IGNORE_LINKSTATE);
1176                 }
1177
1178                 if (err) {
1179                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1180                         goto out;
1181                 }
1182         }
1183
1184         err = -EINVAL;
1185         if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1186                 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1187                 goto out;
1188         }
1189         nh->fib_nh_scope = res.scope;
1190         nh->fib_nh_oif = FIB_RES_OIF(res);
1191         nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1192         if (!dev) {
1193                 NL_SET_ERR_MSG(extack,
1194                                "No egress device for nexthop gateway");
1195                 goto out;
1196         }
1197         dev_hold(dev);
1198         if (!netif_carrier_ok(dev))
1199                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1200         err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1201 out:
1202         rcu_read_unlock();
1203         return err;
1204 }
1205
1206 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1207                               struct netlink_ext_ack *extack)
1208 {
1209         struct in_device *in_dev;
1210         int err;
1211
1212         if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1213                 NL_SET_ERR_MSG(extack,
1214                                "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1215                 return -EINVAL;
1216         }
1217
1218         rcu_read_lock();
1219
1220         err = -ENODEV;
1221         in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1222         if (!in_dev)
1223                 goto out;
1224         err = -ENETDOWN;
1225         if (!(in_dev->dev->flags & IFF_UP)) {
1226                 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1227                 goto out;
1228         }
1229
1230         nh->fib_nh_dev = in_dev->dev;
1231         dev_hold(nh->fib_nh_dev);
1232         nh->fib_nh_scope = RT_SCOPE_LINK;
1233         if (!netif_carrier_ok(nh->fib_nh_dev))
1234                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1235         err = 0;
1236 out:
1237         rcu_read_unlock();
1238         return err;
1239 }
1240
1241 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1242                  struct netlink_ext_ack *extack)
1243 {
1244         int err;
1245
1246         if (nh->fib_nh_gw_family == AF_INET)
1247                 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1248         else if (nh->fib_nh_gw_family == AF_INET6)
1249                 err = fib_check_nh_v6_gw(net, nh, table, extack);
1250         else
1251                 err = fib_check_nh_nongw(net, nh, extack);
1252
1253         return err;
1254 }
1255
1256 static inline unsigned int fib_laddr_hashfn(__be32 val)
1257 {
1258         unsigned int mask = (fib_info_hash_size - 1);
1259
1260         return ((__force u32)val ^
1261                 ((__force u32)val >> 7) ^
1262                 ((__force u32)val >> 14)) & mask;
1263 }
1264
1265 static struct hlist_head *fib_info_hash_alloc(int bytes)
1266 {
1267         if (bytes <= PAGE_SIZE)
1268                 return kzalloc(bytes, GFP_KERNEL);
1269         else
1270                 return (struct hlist_head *)
1271                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1272                                          get_order(bytes));
1273 }
1274
1275 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1276 {
1277         if (!hash)
1278                 return;
1279
1280         if (bytes <= PAGE_SIZE)
1281                 kfree(hash);
1282         else
1283                 free_pages((unsigned long) hash, get_order(bytes));
1284 }
1285
1286 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1287                                struct hlist_head *new_laddrhash,
1288                                unsigned int new_size)
1289 {
1290         struct hlist_head *old_info_hash, *old_laddrhash;
1291         unsigned int old_size = fib_info_hash_size;
1292         unsigned int i, bytes;
1293
1294         spin_lock_bh(&fib_info_lock);
1295         old_info_hash = fib_info_hash;
1296         old_laddrhash = fib_info_laddrhash;
1297         fib_info_hash_size = new_size;
1298
1299         for (i = 0; i < old_size; i++) {
1300                 struct hlist_head *head = &fib_info_hash[i];
1301                 struct hlist_node *n;
1302                 struct fib_info *fi;
1303
1304                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1305                         struct hlist_head *dest;
1306                         unsigned int new_hash;
1307
1308                         new_hash = fib_info_hashfn(fi);
1309                         dest = &new_info_hash[new_hash];
1310                         hlist_add_head(&fi->fib_hash, dest);
1311                 }
1312         }
1313         fib_info_hash = new_info_hash;
1314
1315         for (i = 0; i < old_size; i++) {
1316                 struct hlist_head *lhead = &fib_info_laddrhash[i];
1317                 struct hlist_node *n;
1318                 struct fib_info *fi;
1319
1320                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1321                         struct hlist_head *ldest;
1322                         unsigned int new_hash;
1323
1324                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1325                         ldest = &new_laddrhash[new_hash];
1326                         hlist_add_head(&fi->fib_lhash, ldest);
1327                 }
1328         }
1329         fib_info_laddrhash = new_laddrhash;
1330
1331         spin_unlock_bh(&fib_info_lock);
1332
1333         bytes = old_size * sizeof(struct hlist_head *);
1334         fib_info_hash_free(old_info_hash, bytes);
1335         fib_info_hash_free(old_laddrhash, bytes);
1336 }
1337
1338 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1339                                  unsigned char scope)
1340 {
1341         struct fib_nh *nh;
1342
1343         if (nhc->nhc_family != AF_INET)
1344                 return inet_select_addr(nhc->nhc_dev, 0, scope);
1345
1346         nh = container_of(nhc, struct fib_nh, nh_common);
1347         nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1348         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1349
1350         return nh->nh_saddr;
1351 }
1352
1353 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1354 {
1355         struct fib_nh_common *nhc = res->nhc;
1356
1357         if (res->fi->fib_prefsrc)
1358                 return res->fi->fib_prefsrc;
1359
1360         if (nhc->nhc_family == AF_INET) {
1361                 struct fib_nh *nh;
1362
1363                 nh = container_of(nhc, struct fib_nh, nh_common);
1364                 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1365                         return nh->nh_saddr;
1366         }
1367
1368         return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1369 }
1370
1371 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1372 {
1373         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1374             fib_prefsrc != cfg->fc_dst) {
1375                 u32 tb_id = cfg->fc_table;
1376                 int rc;
1377
1378                 if (tb_id == RT_TABLE_MAIN)
1379                         tb_id = RT_TABLE_LOCAL;
1380
1381                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1382                                           fib_prefsrc, tb_id);
1383
1384                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1385                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1386                                                   fib_prefsrc, RT_TABLE_LOCAL);
1387                 }
1388
1389                 if (rc != RTN_LOCAL)
1390                         return false;
1391         }
1392         return true;
1393 }
1394
1395 struct fib_info *fib_create_info(struct fib_config *cfg,
1396                                  struct netlink_ext_ack *extack)
1397 {
1398         int err;
1399         struct fib_info *fi = NULL;
1400         struct nexthop *nh = NULL;
1401         struct fib_info *ofi;
1402         int nhs = 1;
1403         struct net *net = cfg->fc_nlinfo.nl_net;
1404
1405         if (cfg->fc_type > RTN_MAX)
1406                 goto err_inval;
1407
1408         /* Fast check to catch the most weird cases */
1409         if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1410                 NL_SET_ERR_MSG(extack, "Invalid scope");
1411                 goto err_inval;
1412         }
1413
1414         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1415                 NL_SET_ERR_MSG(extack,
1416                                "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1417                 goto err_inval;
1418         }
1419
1420         if (cfg->fc_nh_id) {
1421                 if (!cfg->fc_mx) {
1422                         fi = fib_find_info_nh(net, cfg);
1423                         if (fi) {
1424                                 fi->fib_treeref++;
1425                                 return fi;
1426                         }
1427                 }
1428
1429                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1430                 if (!nh) {
1431                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1432                         goto err_inval;
1433                 }
1434                 nhs = 0;
1435         }
1436
1437 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1438         if (cfg->fc_mp) {
1439                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1440                 if (nhs == 0)
1441                         goto err_inval;
1442         }
1443 #endif
1444
1445         err = -ENOBUFS;
1446
1447         /* Paired with WRITE_ONCE() in fib_release_info() */
1448         if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1449                 unsigned int new_size = fib_info_hash_size << 1;
1450                 struct hlist_head *new_info_hash;
1451                 struct hlist_head *new_laddrhash;
1452                 unsigned int bytes;
1453
1454                 if (!new_size)
1455                         new_size = 16;
1456                 bytes = new_size * sizeof(struct hlist_head *);
1457                 new_info_hash = fib_info_hash_alloc(bytes);
1458                 new_laddrhash = fib_info_hash_alloc(bytes);
1459                 if (!new_info_hash || !new_laddrhash) {
1460                         fib_info_hash_free(new_info_hash, bytes);
1461                         fib_info_hash_free(new_laddrhash, bytes);
1462                 } else
1463                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1464
1465                 if (!fib_info_hash_size)
1466                         goto failure;
1467         }
1468
1469         fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1470         if (!fi)
1471                 goto failure;
1472         fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1473                                               cfg->fc_mx_len, extack);
1474         if (IS_ERR(fi->fib_metrics)) {
1475                 err = PTR_ERR(fi->fib_metrics);
1476                 kfree(fi);
1477                 return ERR_PTR(err);
1478         }
1479
1480         fi->fib_net = net;
1481         fi->fib_protocol = cfg->fc_protocol;
1482         fi->fib_scope = cfg->fc_scope;
1483         fi->fib_flags = cfg->fc_flags;
1484         fi->fib_priority = cfg->fc_priority;
1485         fi->fib_prefsrc = cfg->fc_prefsrc;
1486         fi->fib_type = cfg->fc_type;
1487         fi->fib_tb_id = cfg->fc_table;
1488
1489         fi->fib_nhs = nhs;
1490         if (nh) {
1491                 if (!nexthop_get(nh)) {
1492                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1493                         err = -EINVAL;
1494                 } else {
1495                         err = 0;
1496                         fi->nh = nh;
1497                 }
1498         } else {
1499                 change_nexthops(fi) {
1500                         nexthop_nh->nh_parent = fi;
1501                 } endfor_nexthops(fi)
1502
1503                 if (cfg->fc_mp)
1504                         err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1505                                           extack);
1506                 else
1507                         err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1508         }
1509
1510         if (err != 0)
1511                 goto failure;
1512
1513         if (fib_props[cfg->fc_type].error) {
1514                 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1515                         NL_SET_ERR_MSG(extack,
1516                                        "Gateway, device and multipath can not be specified for this route type");
1517                         goto err_inval;
1518                 }
1519                 goto link_it;
1520         } else {
1521                 switch (cfg->fc_type) {
1522                 case RTN_UNICAST:
1523                 case RTN_LOCAL:
1524                 case RTN_BROADCAST:
1525                 case RTN_ANYCAST:
1526                 case RTN_MULTICAST:
1527                         break;
1528                 default:
1529                         NL_SET_ERR_MSG(extack, "Invalid route type");
1530                         goto err_inval;
1531                 }
1532         }
1533
1534         if (cfg->fc_scope > RT_SCOPE_HOST) {
1535                 NL_SET_ERR_MSG(extack, "Invalid scope");
1536                 goto err_inval;
1537         }
1538
1539         if (fi->nh) {
1540                 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1541                 if (err)
1542                         goto failure;
1543         } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1544                 struct fib_nh *nh = fi->fib_nh;
1545
1546                 /* Local address is added. */
1547                 if (nhs != 1) {
1548                         NL_SET_ERR_MSG(extack,
1549                                        "Route with host scope can not have multiple nexthops");
1550                         goto err_inval;
1551                 }
1552                 if (nh->fib_nh_gw_family) {
1553                         NL_SET_ERR_MSG(extack,
1554                                        "Route with host scope can not have a gateway");
1555                         goto err_inval;
1556                 }
1557                 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1558                 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1559                 err = -ENODEV;
1560                 if (!nh->fib_nh_dev)
1561                         goto failure;
1562         } else {
1563                 int linkdown = 0;
1564
1565                 change_nexthops(fi) {
1566                         err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1567                                            cfg->fc_table, cfg->fc_scope,
1568                                            extack);
1569                         if (err != 0)
1570                                 goto failure;
1571                         if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1572                                 linkdown++;
1573                 } endfor_nexthops(fi)
1574                 if (linkdown == fi->fib_nhs)
1575                         fi->fib_flags |= RTNH_F_LINKDOWN;
1576         }
1577
1578         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1579                 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1580                 goto err_inval;
1581         }
1582
1583         if (!fi->nh) {
1584                 change_nexthops(fi) {
1585                         fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1586                                                   fi->fib_scope);
1587                         if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1588                                 fi->fib_nh_is_v6 = true;
1589                 } endfor_nexthops(fi)
1590
1591                 fib_rebalance(fi);
1592         }
1593
1594 link_it:
1595         ofi = fib_find_info(fi);
1596         if (ofi) {
1597                 fi->fib_dead = 1;
1598                 free_fib_info(fi);
1599                 ofi->fib_treeref++;
1600                 return ofi;
1601         }
1602
1603         fi->fib_treeref++;
1604         refcount_set(&fi->fib_clntref, 1);
1605         spin_lock_bh(&fib_info_lock);
1606         fib_info_cnt++;
1607         hlist_add_head(&fi->fib_hash,
1608                        &fib_info_hash[fib_info_hashfn(fi)]);
1609         if (fi->fib_prefsrc) {
1610                 struct hlist_head *head;
1611
1612                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1613                 hlist_add_head(&fi->fib_lhash, head);
1614         }
1615         if (fi->nh) {
1616                 list_add(&fi->nh_list, &nh->fi_list);
1617         } else {
1618                 change_nexthops(fi) {
1619                         struct hlist_head *head;
1620
1621                         if (!nexthop_nh->fib_nh_dev)
1622                                 continue;
1623                         head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1624                         hlist_add_head(&nexthop_nh->nh_hash, head);
1625                 } endfor_nexthops(fi)
1626         }
1627         spin_unlock_bh(&fib_info_lock);
1628         return fi;
1629
1630 err_inval:
1631         err = -EINVAL;
1632
1633 failure:
1634         if (fi) {
1635                 fi->fib_dead = 1;
1636                 free_fib_info(fi);
1637         }
1638
1639         return ERR_PTR(err);
1640 }
1641
1642 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1643                      u8 rt_family, unsigned char *flags, bool skip_oif)
1644 {
1645         if (nhc->nhc_flags & RTNH_F_DEAD)
1646                 *flags |= RTNH_F_DEAD;
1647
1648         if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1649                 *flags |= RTNH_F_LINKDOWN;
1650
1651                 rcu_read_lock();
1652                 switch (nhc->nhc_family) {
1653                 case AF_INET:
1654                         if (ip_ignore_linkdown(nhc->nhc_dev))
1655                                 *flags |= RTNH_F_DEAD;
1656                         break;
1657                 case AF_INET6:
1658                         if (ip6_ignore_linkdown(nhc->nhc_dev))
1659                                 *flags |= RTNH_F_DEAD;
1660                         break;
1661                 }
1662                 rcu_read_unlock();
1663         }
1664
1665         switch (nhc->nhc_gw_family) {
1666         case AF_INET:
1667                 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1668                         goto nla_put_failure;
1669                 break;
1670         case AF_INET6:
1671                 /* if gateway family does not match nexthop family
1672                  * gateway is encoded as RTA_VIA
1673                  */
1674                 if (rt_family != nhc->nhc_gw_family) {
1675                         int alen = sizeof(struct in6_addr);
1676                         struct nlattr *nla;
1677                         struct rtvia *via;
1678
1679                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
1680                         if (!nla)
1681                                 goto nla_put_failure;
1682
1683                         via = nla_data(nla);
1684                         via->rtvia_family = AF_INET6;
1685                         memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1686                 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1687                                             &nhc->nhc_gw.ipv6) < 0) {
1688                         goto nla_put_failure;
1689                 }
1690                 break;
1691         }
1692
1693         *flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1694         if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1695                 *flags |= RTNH_F_OFFLOAD;
1696
1697         if (!skip_oif && nhc->nhc_dev &&
1698             nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1699                 goto nla_put_failure;
1700
1701         if (nhc->nhc_lwtstate &&
1702             lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1703                                 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1704                 goto nla_put_failure;
1705
1706         return 0;
1707
1708 nla_put_failure:
1709         return -EMSGSIZE;
1710 }
1711 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1712
1713 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1714 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1715                     int nh_weight, u8 rt_family, u32 nh_tclassid)
1716 {
1717         const struct net_device *dev = nhc->nhc_dev;
1718         struct rtnexthop *rtnh;
1719         unsigned char flags = 0;
1720
1721         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1722         if (!rtnh)
1723                 goto nla_put_failure;
1724
1725         rtnh->rtnh_hops = nh_weight - 1;
1726         rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1727
1728         if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1729                 goto nla_put_failure;
1730
1731         rtnh->rtnh_flags = flags;
1732
1733         if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1734                 goto nla_put_failure;
1735
1736         /* length of rtnetlink header + attributes */
1737         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1738
1739         return 0;
1740
1741 nla_put_failure:
1742         return -EMSGSIZE;
1743 }
1744 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1745 #endif
1746
1747 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1748 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1749 {
1750         struct nlattr *mp;
1751
1752         mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1753         if (!mp)
1754                 goto nla_put_failure;
1755
1756         if (unlikely(fi->nh)) {
1757                 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1758                         goto nla_put_failure;
1759                 goto mp_end;
1760         }
1761
1762         for_nexthops(fi) {
1763                 u32 nh_tclassid = 0;
1764 #ifdef CONFIG_IP_ROUTE_CLASSID
1765                 nh_tclassid = nh->nh_tclassid;
1766 #endif
1767                 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1768                                     AF_INET, nh_tclassid) < 0)
1769                         goto nla_put_failure;
1770         } endfor_nexthops(fi);
1771
1772 mp_end:
1773         nla_nest_end(skb, mp);
1774
1775         return 0;
1776
1777 nla_put_failure:
1778         return -EMSGSIZE;
1779 }
1780 #else
1781 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1782 {
1783         return 0;
1784 }
1785 #endif
1786
1787 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1788                   struct fib_rt_info *fri, unsigned int flags)
1789 {
1790         unsigned int nhs = fib_info_num_path(fri->fi);
1791         struct fib_info *fi = fri->fi;
1792         u32 tb_id = fri->tb_id;
1793         struct nlmsghdr *nlh;
1794         struct rtmsg *rtm;
1795
1796         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1797         if (!nlh)
1798                 return -EMSGSIZE;
1799
1800         rtm = nlmsg_data(nlh);
1801         rtm->rtm_family = AF_INET;
1802         rtm->rtm_dst_len = fri->dst_len;
1803         rtm->rtm_src_len = 0;
1804         rtm->rtm_tos = fri->tos;
1805         if (tb_id < 256)
1806                 rtm->rtm_table = tb_id;
1807         else
1808                 rtm->rtm_table = RT_TABLE_COMPAT;
1809         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1810                 goto nla_put_failure;
1811         rtm->rtm_type = fri->type;
1812         rtm->rtm_flags = fi->fib_flags;
1813         rtm->rtm_scope = fi->fib_scope;
1814         rtm->rtm_protocol = fi->fib_protocol;
1815
1816         if (rtm->rtm_dst_len &&
1817             nla_put_in_addr(skb, RTA_DST, fri->dst))
1818                 goto nla_put_failure;
1819         if (fi->fib_priority &&
1820             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1821                 goto nla_put_failure;
1822         if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1823                 goto nla_put_failure;
1824
1825         if (fi->fib_prefsrc &&
1826             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1827                 goto nla_put_failure;
1828
1829         if (fi->nh) {
1830                 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1831                         goto nla_put_failure;
1832                 if (nexthop_is_blackhole(fi->nh))
1833                         rtm->rtm_type = RTN_BLACKHOLE;
1834                 if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1835                         goto offload;
1836         }
1837
1838         if (nhs == 1) {
1839                 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1840                 unsigned char flags = 0;
1841
1842                 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1843                         goto nla_put_failure;
1844
1845                 rtm->rtm_flags = flags;
1846 #ifdef CONFIG_IP_ROUTE_CLASSID
1847                 if (nhc->nhc_family == AF_INET) {
1848                         struct fib_nh *nh;
1849
1850                         nh = container_of(nhc, struct fib_nh, nh_common);
1851                         if (nh->nh_tclassid &&
1852                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1853                                 goto nla_put_failure;
1854                 }
1855 #endif
1856         } else {
1857                 if (fib_add_multipath(skb, fi) < 0)
1858                         goto nla_put_failure;
1859         }
1860
1861 offload:
1862         if (fri->offload)
1863                 rtm->rtm_flags |= RTM_F_OFFLOAD;
1864         if (fri->trap)
1865                 rtm->rtm_flags |= RTM_F_TRAP;
1866
1867         nlmsg_end(skb, nlh);
1868         return 0;
1869
1870 nla_put_failure:
1871         nlmsg_cancel(skb, nlh);
1872         return -EMSGSIZE;
1873 }
1874
1875 /*
1876  * Update FIB if:
1877  * - local address disappeared -> we must delete all the entries
1878  *   referring to it.
1879  * - device went down -> we must shutdown all nexthops going via it.
1880  */
1881 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1882 {
1883         int ret = 0;
1884         unsigned int hash = fib_laddr_hashfn(local);
1885         struct hlist_head *head = &fib_info_laddrhash[hash];
1886         int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1887         struct net *net = dev_net(dev);
1888         struct fib_info *fi;
1889
1890         if (!fib_info_laddrhash || local == 0)
1891                 return 0;
1892
1893         hlist_for_each_entry(fi, head, fib_lhash) {
1894                 if (!net_eq(fi->fib_net, net) ||
1895                     fi->fib_tb_id != tb_id)
1896                         continue;
1897                 if (fi->fib_prefsrc == local) {
1898                         fi->fib_flags |= RTNH_F_DEAD;
1899                         ret++;
1900                 }
1901         }
1902         return ret;
1903 }
1904
1905 static int call_fib_nh_notifiers(struct fib_nh *nh,
1906                                  enum fib_event_type event_type)
1907 {
1908         bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1909         struct fib_nh_notifier_info info = {
1910                 .fib_nh = nh,
1911         };
1912
1913         switch (event_type) {
1914         case FIB_EVENT_NH_ADD:
1915                 if (nh->fib_nh_flags & RTNH_F_DEAD)
1916                         break;
1917                 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1918                         break;
1919                 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1920                                            &info.info);
1921         case FIB_EVENT_NH_DEL:
1922                 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1923                     (nh->fib_nh_flags & RTNH_F_DEAD))
1924                         return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1925                                                    event_type, &info.info);
1926         default:
1927                 break;
1928         }
1929
1930         return NOTIFY_DONE;
1931 }
1932
1933 /* Update the PMTU of exceptions when:
1934  * - the new MTU of the first hop becomes smaller than the PMTU
1935  * - the old MTU was the same as the PMTU, and it limited discovery of
1936  *   larger MTUs on the path. With that limit raised, we can now
1937  *   discover larger MTUs
1938  * A special case is locked exceptions, for which the PMTU is smaller
1939  * than the minimal accepted PMTU:
1940  * - if the new MTU is greater than the PMTU, don't make any change
1941  * - otherwise, unlock and set PMTU
1942  */
1943 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1944 {
1945         struct fnhe_hash_bucket *bucket;
1946         int i;
1947
1948         bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1949         if (!bucket)
1950                 return;
1951
1952         for (i = 0; i < FNHE_HASH_SIZE; i++) {
1953                 struct fib_nh_exception *fnhe;
1954
1955                 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1956                      fnhe;
1957                      fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1958                         if (fnhe->fnhe_mtu_locked) {
1959                                 if (new <= fnhe->fnhe_pmtu) {
1960                                         fnhe->fnhe_pmtu = new;
1961                                         fnhe->fnhe_mtu_locked = false;
1962                                 }
1963                         } else if (new < fnhe->fnhe_pmtu ||
1964                                    orig == fnhe->fnhe_pmtu) {
1965                                 fnhe->fnhe_pmtu = new;
1966                         }
1967                 }
1968         }
1969 }
1970
1971 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1972 {
1973         struct hlist_head *head = fib_info_devhash_bucket(dev);
1974         struct fib_nh *nh;
1975
1976         hlist_for_each_entry(nh, head, nh_hash) {
1977                 if (nh->fib_nh_dev == dev)
1978                         fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1979         }
1980 }
1981
1982 /* Event              force Flags           Description
1983  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1984  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1985  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1986  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1987  *
1988  * only used when fib_nh is built into fib_info
1989  */
1990 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1991 {
1992         struct hlist_head *head = fib_info_devhash_bucket(dev);
1993         struct fib_info *prev_fi = NULL;
1994         int scope = RT_SCOPE_NOWHERE;
1995         struct fib_nh *nh;
1996         int ret = 0;
1997
1998         if (force)
1999                 scope = -1;
2000
2001         hlist_for_each_entry(nh, head, nh_hash) {
2002                 struct fib_info *fi = nh->nh_parent;
2003                 int dead;
2004
2005                 BUG_ON(!fi->fib_nhs);
2006                 if (nh->fib_nh_dev != dev || fi == prev_fi)
2007                         continue;
2008                 prev_fi = fi;
2009                 dead = 0;
2010                 change_nexthops(fi) {
2011                         if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2012                                 dead++;
2013                         else if (nexthop_nh->fib_nh_dev == dev &&
2014                                  nexthop_nh->fib_nh_scope != scope) {
2015                                 switch (event) {
2016                                 case NETDEV_DOWN:
2017                                 case NETDEV_UNREGISTER:
2018                                         nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2019                                         fallthrough;
2020                                 case NETDEV_CHANGE:
2021                                         nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2022                                         break;
2023                                 }
2024                                 call_fib_nh_notifiers(nexthop_nh,
2025                                                       FIB_EVENT_NH_DEL);
2026                                 dead++;
2027                         }
2028 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2029                         if (event == NETDEV_UNREGISTER &&
2030                             nexthop_nh->fib_nh_dev == dev) {
2031                                 dead = fi->fib_nhs;
2032                                 break;
2033                         }
2034 #endif
2035                 } endfor_nexthops(fi)
2036                 if (dead == fi->fib_nhs) {
2037                         switch (event) {
2038                         case NETDEV_DOWN:
2039                         case NETDEV_UNREGISTER:
2040                                 fi->fib_flags |= RTNH_F_DEAD;
2041                                 fallthrough;
2042                         case NETDEV_CHANGE:
2043                                 fi->fib_flags |= RTNH_F_LINKDOWN;
2044                                 break;
2045                         }
2046                         ret++;
2047                 }
2048
2049                 fib_rebalance(fi);
2050         }
2051
2052         return ret;
2053 }
2054
2055 /* Must be invoked inside of an RCU protected region.  */
2056 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2057 {
2058         struct fib_info *fi = NULL, *last_resort = NULL;
2059         struct hlist_head *fa_head = res->fa_head;
2060         struct fib_table *tb = res->table;
2061         u8 slen = 32 - res->prefixlen;
2062         int order = -1, last_idx = -1;
2063         struct fib_alias *fa, *fa1 = NULL;
2064         u32 last_prio = res->fi->fib_priority;
2065         u8 last_tos = 0;
2066
2067         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2068                 struct fib_info *next_fi = fa->fa_info;
2069                 struct fib_nh_common *nhc;
2070
2071                 if (fa->fa_slen != slen)
2072                         continue;
2073                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2074                         continue;
2075                 if (fa->tb_id != tb->tb_id)
2076                         continue;
2077                 if (next_fi->fib_priority > last_prio &&
2078                     fa->fa_tos == last_tos) {
2079                         if (last_tos)
2080                                 continue;
2081                         break;
2082                 }
2083                 if (next_fi->fib_flags & RTNH_F_DEAD)
2084                         continue;
2085                 last_tos = fa->fa_tos;
2086                 last_prio = next_fi->fib_priority;
2087
2088                 if (next_fi->fib_scope != res->scope ||
2089                     fa->fa_type != RTN_UNICAST)
2090                         continue;
2091
2092                 nhc = fib_info_nhc(next_fi, 0);
2093                 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2094                         continue;
2095
2096                 fib_alias_accessed(fa);
2097
2098                 if (!fi) {
2099                         if (next_fi != res->fi)
2100                                 break;
2101                         fa1 = fa;
2102                 } else if (!fib_detect_death(fi, order, &last_resort,
2103                                              &last_idx, fa1->fa_default)) {
2104                         fib_result_assign(res, fi);
2105                         fa1->fa_default = order;
2106                         goto out;
2107                 }
2108                 fi = next_fi;
2109                 order++;
2110         }
2111
2112         if (order <= 0 || !fi) {
2113                 if (fa1)
2114                         fa1->fa_default = -1;
2115                 goto out;
2116         }
2117
2118         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2119                               fa1->fa_default)) {
2120                 fib_result_assign(res, fi);
2121                 fa1->fa_default = order;
2122                 goto out;
2123         }
2124
2125         if (last_idx >= 0)
2126                 fib_result_assign(res, last_resort);
2127         fa1->fa_default = last_idx;
2128 out:
2129         return;
2130 }
2131
2132 /*
2133  * Dead device goes up. We wake up dead nexthops.
2134  * It takes sense only on multipath routes.
2135  *
2136  * only used when fib_nh is built into fib_info
2137  */
2138 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2139 {
2140         struct fib_info *prev_fi;
2141         struct hlist_head *head;
2142         struct fib_nh *nh;
2143         int ret;
2144
2145         if (!(dev->flags & IFF_UP))
2146                 return 0;
2147
2148         if (nh_flags & RTNH_F_DEAD) {
2149                 unsigned int flags = dev_get_flags(dev);
2150
2151                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2152                         nh_flags |= RTNH_F_LINKDOWN;
2153         }
2154
2155         prev_fi = NULL;
2156         head = fib_info_devhash_bucket(dev);
2157         ret = 0;
2158
2159         hlist_for_each_entry(nh, head, nh_hash) {
2160                 struct fib_info *fi = nh->nh_parent;
2161                 int alive;
2162
2163                 BUG_ON(!fi->fib_nhs);
2164                 if (nh->fib_nh_dev != dev || fi == prev_fi)
2165                         continue;
2166
2167                 prev_fi = fi;
2168                 alive = 0;
2169                 change_nexthops(fi) {
2170                         if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2171                                 alive++;
2172                                 continue;
2173                         }
2174                         if (!nexthop_nh->fib_nh_dev ||
2175                             !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2176                                 continue;
2177                         if (nexthop_nh->fib_nh_dev != dev ||
2178                             !__in_dev_get_rtnl(dev))
2179                                 continue;
2180                         alive++;
2181                         nexthop_nh->fib_nh_flags &= ~nh_flags;
2182                         call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2183                 } endfor_nexthops(fi)
2184
2185                 if (alive > 0) {
2186                         fi->fib_flags &= ~nh_flags;
2187                         ret++;
2188                 }
2189
2190                 fib_rebalance(fi);
2191         }
2192
2193         return ret;
2194 }
2195
2196 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2197 static bool fib_good_nh(const struct fib_nh *nh)
2198 {
2199         int state = NUD_REACHABLE;
2200
2201         if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2202                 struct neighbour *n;
2203
2204                 rcu_read_lock_bh();
2205
2206                 if (likely(nh->fib_nh_gw_family == AF_INET))
2207                         n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2208                                                    (__force u32)nh->fib_nh_gw4);
2209                 else if (nh->fib_nh_gw_family == AF_INET6)
2210                         n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2211                                                            &nh->fib_nh_gw6);
2212                 else
2213                         n = NULL;
2214                 if (n)
2215                         state = n->nud_state;
2216
2217                 rcu_read_unlock_bh();
2218         }
2219
2220         return !!(state & NUD_VALID);
2221 }
2222
2223 void fib_select_multipath(struct fib_result *res, int hash)
2224 {
2225         struct fib_info *fi = res->fi;
2226         struct net *net = fi->fib_net;
2227         bool first = false;
2228
2229         if (unlikely(res->fi->nh)) {
2230                 nexthop_path_fib_result(res, hash);
2231                 return;
2232         }
2233
2234         change_nexthops(fi) {
2235                 if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2236                         if (!fib_good_nh(nexthop_nh))
2237                                 continue;
2238                         if (!first) {
2239                                 res->nh_sel = nhsel;
2240                                 res->nhc = &nexthop_nh->nh_common;
2241                                 first = true;
2242                         }
2243                 }
2244
2245                 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2246                         continue;
2247
2248                 res->nh_sel = nhsel;
2249                 res->nhc = &nexthop_nh->nh_common;
2250                 return;
2251         } endfor_nexthops(fi);
2252 }
2253 #endif
2254
2255 void fib_select_path(struct net *net, struct fib_result *res,
2256                      struct flowi4 *fl4, const struct sk_buff *skb)
2257 {
2258         if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2259                 goto check_saddr;
2260
2261 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2262         if (fib_info_num_path(res->fi) > 1) {
2263                 int h = fib_multipath_hash(net, fl4, skb, NULL);
2264
2265                 fib_select_multipath(res, h);
2266         }
2267         else
2268 #endif
2269         if (!res->prefixlen &&
2270             res->table->tb_num_default > 1 &&
2271             res->type == RTN_UNICAST)
2272                 fib_select_default(fl4, res);
2273
2274 check_saddr:
2275         if (!fl4->saddr)
2276                 fl4->saddr = fib_result_prefsrc(net, res);
2277 }