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 #include <linux/nospec.h>
  34
  35 #include <net/arp.h>
  36 #include <net/ip.h>
  37 #include <net/protocol.h>
  38 #include <net/route.h>
  39 #include <net/tcp.h>
  40 #include <net/sock.h>
  41 #include <net/ip_fib.h>
  42 #include <net/ip6_fib.h>
  43 #include <net/nexthop.h>
  44 #include <net/netlink.h>
  45 #include <net/rtnh.h>
  46 #include <net/lwtunnel.h>
  47 #include <net/fib_notifier.h>
  48 #include <net/addrconf.h>
  49
  50 #include "fib_lookup.h"
  51
  52 static DEFINE_SPINLOCK(fib_info_lock);
  53 static struct hlist_head *fib_info_hash;
  54 static struct hlist_head *fib_info_laddrhash;
  55 static unsigned int fib_info_hash_size;
  56 static unsigned int fib_info_cnt;
  57
  58 #define DEVINDEX_HASHBITS 8
  59 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  60 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  61
  62 /* for_nexthops and change_nexthops only used when nexthop object
  63  * is not set in a fib_info. The logic within can reference fib_nh.
  64  */
  65 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  66
  67 #define for_nexthops(fi) {                                              \
  68         int nhsel; const struct fib_nh *nh;                             \
  69         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  70              nhsel < fib_info_num_path((fi));                           \
  71              nh++, nhsel++)
  72
  73 #define change_nexthops(fi) {                                           \
  74         int nhsel; struct fib_nh *nexthop_nh;                           \
  75         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  76              nhsel < fib_info_num_path((fi));                           \
  77              nexthop_nh++, nhsel++)
  78
  79 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  80
  81 /* Hope, that gcc will optimize it to get rid of dummy loop */
  82
  83 #define for_nexthops(fi) {                                              \
  84         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  85         for (nhsel = 0; nhsel < 1; nhsel++)
  86
  87 #define change_nexthops(fi) {                                           \
  88         int nhsel;                                                      \
  89         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  90         for (nhsel = 0; nhsel < 1; nhsel++)
  91
  92 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  93
  94 #define endfor_nexthops(fi) }
  95
  96
  97 const struct fib_prop fib_props[RTN_MAX + 1] = {
  98         [RTN_UNSPEC] = {
  99                 .error  = 0,
 100                 .scope  = RT_SCOPE_NOWHERE,
 101         },
 102         [RTN_UNICAST] = {
 103                 .error  = 0,
 104                 .scope  = RT_SCOPE_UNIVERSE,
 105         },
 106         [RTN_LOCAL] = {
 107                 .error  = 0,
 108                 .scope  = RT_SCOPE_HOST,
 109         },
 110         [RTN_BROADCAST] = {
 111                 .error  = 0,
 112                 .scope  = RT_SCOPE_LINK,
 113         },
 114         [RTN_ANYCAST] = {
 115                 .error  = 0,
 116                 .scope  = RT_SCOPE_LINK,
 117         },
 118         [RTN_MULTICAST] = {
 119                 .error  = 0,
 120                 .scope  = RT_SCOPE_UNIVERSE,
 121         },
 122         [RTN_BLACKHOLE] = {
 123                 .error  = -EINVAL,
 124                 .scope  = RT_SCOPE_UNIVERSE,
 125         },
 126         [RTN_UNREACHABLE] = {
 127                 .error  = -EHOSTUNREACH,
 128                 .scope  = RT_SCOPE_UNIVERSE,
 129         },
 130         [RTN_PROHIBIT] = {
 131                 .error  = -EACCES,
 132                 .scope  = RT_SCOPE_UNIVERSE,
 133         },
 134         [RTN_THROW] = {
 135                 .error  = -EAGAIN,
 136                 .scope  = RT_SCOPE_UNIVERSE,
 137         },
 138         [RTN_NAT] = {
 139                 .error  = -EINVAL,
 140                 .scope  = RT_SCOPE_NOWHERE,
 141         },
 142         [RTN_XRESOLVE] = {
 143                 .error  = -EINVAL,
 144                 .scope  = RT_SCOPE_NOWHERE,
 145         },
 146 };
 147
 148 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 149 {
 150         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 151
 152         if (!rt)
 153                 return;
 154
 155         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 156          * because we waited an RCU grace period before calling
 157          * free_fib_info_rcu()
 158          */
 159
 160         dst_dev_put(&rt->dst);
 161         dst_release_immediate(&rt->dst);
 162 }
 163
 164 static void free_nh_exceptions(struct fib_nh_common *nhc)
 165 {
 166         struct fnhe_hash_bucket *hash;
 167         int i;
 168
 169         hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
 170         if (!hash)
 171                 return;
 172         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 173                 struct fib_nh_exception *fnhe;
 174
 175                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 176                 while (fnhe) {
 177                         struct fib_nh_exception *next;
 178
 179                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 180
 181                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 182                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 183
 184                         kfree(fnhe);
 185
 186                         fnhe = next;
 187                 }
 188         }
 189         kfree(hash);
 190 }
 191
 192 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 193 {
 194         int cpu;
 195
 196         if (!rtp)
 197                 return;
 198
 199         for_each_possible_cpu(cpu) {
 200                 struct rtable *rt;
 201
 202                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 203                 if (rt) {
 204                         dst_dev_put(&rt->dst);
 205                         dst_release_immediate(&rt->dst);
 206                 }
 207         }
 208         free_percpu(rtp);
 209 }
 210
 211 void fib_nh_common_release(struct fib_nh_common *nhc)
 212 {
 213         dev_put(nhc->nhc_dev);
 214         lwtstate_put(nhc->nhc_lwtstate);
 215         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 216         rt_fibinfo_free(&nhc->nhc_rth_input);
 217         free_nh_exceptions(nhc);
 218 }
 219 EXPORT_SYMBOL_GPL(fib_nh_common_release);
 220
 221 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
 222 {
 223 #ifdef CONFIG_IP_ROUTE_CLASSID
 224         if (fib_nh->nh_tclassid)
 225                 atomic_dec(&net->ipv4.fib_num_tclassid_users);
 226 #endif
 227         fib_nh_common_release(&fib_nh->nh_common);
 228 }
 229
 230 /* Release a nexthop info record */
 231 static void free_fib_info_rcu(struct rcu_head *head)
 232 {
 233         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 234
 235         if (fi->nh) {
 236                 nexthop_put(fi->nh);
 237         } else {
 238                 change_nexthops(fi) {
 239                         fib_nh_release(fi->fib_net, nexthop_nh);
 240                 } endfor_nexthops(fi);
 241         }
 242
 243         ip_fib_metrics_put(fi->fib_metrics);
 244
 245         kfree(fi);
 246 }
 247
 248 void free_fib_info(struct fib_info *fi)
 249 {
 250         if (fi->fib_dead == 0) {
 251                 pr_warn("Freeing alive fib_info %p\n", fi);
 252                 return;
 253         }
 254
 255         call_rcu(&fi->rcu, free_fib_info_rcu);
 256 }
 257 EXPORT_SYMBOL_GPL(free_fib_info);
 258
 259 void fib_release_info(struct fib_info *fi)
 260 {
 261         spin_lock_bh(&fib_info_lock);
 262         if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
 263                 hlist_del(&fi->fib_hash);
 264
 265                 /* Paired with READ_ONCE() in fib_create_info(). */
 266                 WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
 267
 268                 if (fi->fib_prefsrc)
 269                         hlist_del(&fi->fib_lhash);
 270                 if (fi->nh) {
 271                         list_del(&fi->nh_list);
 272                 } else {
 273                         change_nexthops(fi) {
 274                                 if (!nexthop_nh->fib_nh_dev)
 275                                         continue;
 276                                 hlist_del(&nexthop_nh->nh_hash);
 277                         } endfor_nexthops(fi)
 278                 }
 279                 /* Paired with READ_ONCE() from fib_table_lookup() */
 280                 WRITE_ONCE(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                     nfi->fib_tb_id == fi->fib_tb_id &&
 427                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 428                            sizeof(u32) * RTAX_MAX) == 0 &&
 429                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 430                     nh_comp(fi, nfi) == 0)
 431                         return fi;
 432         }
 433
 434         return NULL;
 435 }
 436
 437 /* Check, that the gateway is already configured.
 438  * Used only by redirect accept routine.
 439  */
 440 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 441 {
 442         struct hlist_head *head;
 443         struct fib_nh *nh;
 444
 445         spin_lock(&fib_info_lock);
 446
 447         head = fib_info_devhash_bucket(dev);
 448
 449         hlist_for_each_entry(nh, head, nh_hash) {
 450                 if (nh->fib_nh_dev == dev &&
 451                     nh->fib_nh_gw4 == gw &&
 452                     !(nh->fib_nh_flags & RTNH_F_DEAD)) {
 453                         spin_unlock(&fib_info_lock);
 454                         return 0;
 455                 }
 456         }
 457
 458         spin_unlock(&fib_info_lock);
 459
 460         return -1;
 461 }
 462
 463 size_t fib_nlmsg_size(struct fib_info *fi)
 464 {
 465         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 466                          + nla_total_size(4) /* RTA_TABLE */
 467                          + nla_total_size(4) /* RTA_DST */
 468                          + nla_total_size(4) /* RTA_PRIORITY */
 469                          + nla_total_size(4) /* RTA_PREFSRC */
 470                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 471         unsigned int nhs = fib_info_num_path(fi);
 472
 473         /* space for nested metrics */
 474         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 475
 476         if (fi->nh)
 477                 payload += nla_total_size(4); /* RTA_NH_ID */
 478
 479         if (nhs) {
 480                 size_t nh_encapsize = 0;
 481                 /* Also handles the special case nhs == 1 */
 482
 483                 /* each nexthop is packed in an attribute */
 484                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 485                 unsigned int i;
 486
 487                 /* may contain flow and gateway attribute */
 488                 nhsize += 2 * nla_total_size(4);
 489
 490                 /* grab encap info */
 491                 for (i = 0; i < fib_info_num_path(fi); i++) {
 492                         struct fib_nh_common *nhc = fib_info_nhc(fi, i);
 493
 494                         if (nhc->nhc_lwtstate) {
 495                                 /* RTA_ENCAP_TYPE */
 496                                 nh_encapsize += lwtunnel_get_encap_size(
 497                                                 nhc->nhc_lwtstate);
 498                                 /* RTA_ENCAP */
 499                                 nh_encapsize +=  nla_total_size(2);
 500                         }
 501                 }
 502
 503                 /* all nexthops are packed in a nested attribute */
 504                 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
 505
 506         }
 507
 508         return payload;
 509 }
 510
 511 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 512                int dst_len, u32 tb_id, const struct nl_info *info,
 513                unsigned int nlm_flags)
 514 {
 515         struct fib_rt_info fri;
 516         struct sk_buff *skb;
 517         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 518         int err = -ENOBUFS;
 519
 520         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 521         if (!skb)
 522                 goto errout;
 523
 524         fri.fi = fa->fa_info;
 525         fri.tb_id = tb_id;
 526         fri.dst = key;
 527         fri.dst_len = dst_len;
 528         fri.tos = fa->fa_tos;
 529         fri.type = fa->fa_type;
 530         fri.offload = READ_ONCE(fa->offload);
 531         fri.trap = READ_ONCE(fa->trap);
 532         fri.offload_failed = READ_ONCE(fa->offload_failed);
 533         err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
 534         if (err < 0) {
 535                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 536                 WARN_ON(err == -EMSGSIZE);
 537                 kfree_skb(skb);
 538                 goto errout;
 539         }
 540         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 541                     info->nlh, GFP_KERNEL);
 542         return;
 543 errout:
 544         if (err < 0)
 545                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 546 }
 547
 548 static int fib_detect_death(struct fib_info *fi, int order,
 549                             struct fib_info **last_resort, int *last_idx,
 550                             int dflt)
 551 {
 552         const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
 553         struct neighbour *n;
 554         int state = NUD_NONE;
 555
 556         if (likely(nhc->nhc_gw_family == AF_INET))
 557                 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
 558         else if (nhc->nhc_gw_family == AF_INET6)
 559                 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
 560                                  nhc->nhc_dev);
 561         else
 562                 n = NULL;
 563
 564         if (n) {
 565                 state = n->nud_state;
 566                 neigh_release(n);
 567         } else {
 568                 return 0;
 569         }
 570         if (state == NUD_REACHABLE)
 571                 return 0;
 572         if ((state & NUD_VALID) && order != dflt)
 573                 return 0;
 574         if ((state & NUD_VALID) ||
 575             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 576                 *last_resort = fi;
 577                 *last_idx = order;
 578         }
 579         return 1;
 580 }
 581
 582 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
 583                        struct nlattr *encap, u16 encap_type,
 584                        void *cfg, gfp_t gfp_flags,
 585                        struct netlink_ext_ack *extack)
 586 {
 587         int err;
 588
 589         nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
 590                                                     gfp_flags);
 591         if (!nhc->nhc_pcpu_rth_output)
 592                 return -ENOMEM;
 593
 594         if (encap) {
 595                 struct lwtunnel_state *lwtstate;
 596
 597                 if (encap_type == LWTUNNEL_ENCAP_NONE) {
 598                         NL_SET_ERR_MSG(extack, "LWT encap type not specified");
 599                         err = -EINVAL;
 600                         goto lwt_failure;
 601                 }
 602                 err = lwtunnel_build_state(net, encap_type, encap,
 603                                            nhc->nhc_family, cfg, &lwtstate,
 604                                            extack);
 605                 if (err)
 606                         goto lwt_failure;
 607
 608                 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
 609         }
 610
 611         return 0;
 612
 613 lwt_failure:
 614         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 615         nhc->nhc_pcpu_rth_output = NULL;
 616         return err;
 617 }
 618 EXPORT_SYMBOL_GPL(fib_nh_common_init);
 619
 620 int fib_nh_init(struct net *net, struct fib_nh *nh,
 621                 struct fib_config *cfg, int nh_weight,
 622                 struct netlink_ext_ack *extack)
 623 {
 624         int err;
 625
 626         nh->fib_nh_family = AF_INET;
 627
 628         err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
 629                                  cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
 630         if (err)
 631                 return err;
 632
 633         nh->fib_nh_oif = cfg->fc_oif;
 634         nh->fib_nh_gw_family = cfg->fc_gw_family;
 635         if (cfg->fc_gw_family == AF_INET)
 636                 nh->fib_nh_gw4 = cfg->fc_gw4;
 637         else if (cfg->fc_gw_family == AF_INET6)
 638                 nh->fib_nh_gw6 = cfg->fc_gw6;
 639
 640         nh->fib_nh_flags = cfg->fc_flags;
 641
 642 #ifdef CONFIG_IP_ROUTE_CLASSID
 643         nh->nh_tclassid = cfg->fc_flow;
 644         if (nh->nh_tclassid)
 645                 atomic_inc(&net->ipv4.fib_num_tclassid_users);
 646 #endif
 647 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 648         nh->fib_nh_weight = nh_weight;
 649 #endif
 650         return 0;
 651 }
 652
 653 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 654
 655 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
 656                               struct netlink_ext_ack *extack)
 657 {
 658         int nhs = 0;
 659
 660         while (rtnh_ok(rtnh, remaining)) {
 661                 nhs++;
 662                 rtnh = rtnh_next(rtnh, &remaining);
 663         }
 664
 665         /* leftover implies invalid nexthop configuration, discard it */
 666         if (remaining > 0) {
 667                 NL_SET_ERR_MSG(extack,
 668                                "Invalid nexthop configuration - extra data after nexthops");
 669                 nhs = 0;
 670         }
 671
 672         return nhs;
 673 }
 674
 675 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
 676                             struct netlink_ext_ack *extack)
 677 {
 678         if (nla_len(nla) < sizeof(*gw)) {
 679                 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
 680                 return -EINVAL;
 681         }
 682
 683         *gw = nla_get_in_addr(nla);
 684
 685         return 0;
 686 }
 687
 688 /* only called when fib_nh is integrated into fib_info */
 689 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 690                        int remaining, struct fib_config *cfg,
 691                        struct netlink_ext_ack *extack)
 692 {
 693         struct net *net = fi->fib_net;
 694         struct fib_config fib_cfg;
 695         struct fib_nh *nh;
 696         int ret;
 697
 698         change_nexthops(fi) {
 699                 int attrlen;
 700
 701                 memset(&fib_cfg, 0, sizeof(fib_cfg));
 702
 703                 if (!rtnh_ok(rtnh, remaining)) {
 704                         NL_SET_ERR_MSG(extack,
 705                                        "Invalid nexthop configuration - extra data after nexthop");
 706                         return -EINVAL;
 707                 }
 708
 709                 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
 710                         NL_SET_ERR_MSG(extack,
 711                                        "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
 712                         return -EINVAL;
 713                 }
 714
 715                 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 716                 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
 717
 718                 attrlen = rtnh_attrlen(rtnh);
 719                 if (attrlen > 0) {
 720                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 721
 722                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 723                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 724                         if (nla && nlav) {
 725                                 NL_SET_ERR_MSG(extack,
 726                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 727                                 return -EINVAL;
 728                         }
 729                         if (nla) {
 730                                 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
 731                                                        extack);
 732                                 if (ret)
 733                                         goto errout;
 734
 735                                 if (fib_cfg.fc_gw4)
 736                                         fib_cfg.fc_gw_family = AF_INET;
 737                         } else if (nlav) {
 738                                 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
 739                                 if (ret)
 740                                         goto errout;
 741                         }
 742
 743                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 744                         if (nla) {
 745                                 if (nla_len(nla) < sizeof(u32)) {
 746                                         NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
 747                                         return -EINVAL;
 748                                 }
 749                                 fib_cfg.fc_flow = nla_get_u32(nla);
 750                         }
 751
 752                         fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
 753                         /* RTA_ENCAP_TYPE length checked in
 754                          * lwtunnel_valid_encap_type_attr
 755                          */
 756                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
 757                         if (nla)
 758                                 fib_cfg.fc_encap_type = nla_get_u16(nla);
 759                 }
 760
 761                 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
 762                                   rtnh->rtnh_hops + 1, extack);
 763                 if (ret)
 764                         goto errout;
 765
 766                 rtnh = rtnh_next(rtnh, &remaining);
 767         } endfor_nexthops(fi);
 768
 769         ret = -EINVAL;
 770         nh = fib_info_nh(fi, 0);
 771         if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
 772                 NL_SET_ERR_MSG(extack,
 773                                "Nexthop device index does not match RTA_OIF");
 774                 goto errout;
 775         }
 776         if (cfg->fc_gw_family) {
 777                 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
 778                     (cfg->fc_gw_family == AF_INET &&
 779                      nh->fib_nh_gw4 != cfg->fc_gw4) ||
 780                     (cfg->fc_gw_family == AF_INET6 &&
 781                      ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
 782                         NL_SET_ERR_MSG(extack,
 783                                        "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
 784                         goto errout;
 785                 }
 786         }
 787 #ifdef CONFIG_IP_ROUTE_CLASSID
 788         if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
 789                 NL_SET_ERR_MSG(extack,
 790                                "Nexthop class id does not match RTA_FLOW");
 791                 goto errout;
 792         }
 793 #endif
 794         ret = 0;
 795 errout:
 796         return ret;
 797 }
 798
 799 /* only called when fib_nh is integrated into fib_info */
 800 static void fib_rebalance(struct fib_info *fi)
 801 {
 802         int total;
 803         int w;
 804
 805         if (fib_info_num_path(fi) < 2)
 806                 return;
 807
 808         total = 0;
 809         for_nexthops(fi) {
 810                 if (nh->fib_nh_flags & RTNH_F_DEAD)
 811                         continue;
 812
 813                 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
 814                     nh->fib_nh_flags & RTNH_F_LINKDOWN)
 815                         continue;
 816
 817                 total += nh->fib_nh_weight;
 818         } endfor_nexthops(fi);
 819
 820         w = 0;
 821         change_nexthops(fi) {
 822                 int upper_bound;
 823
 824                 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
 825                         upper_bound = -1;
 826                 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
 827                            nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
 828                         upper_bound = -1;
 829                 } else {
 830                         w += nexthop_nh->fib_nh_weight;
 831                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 832                                                             total) - 1;
 833                 }
 834
 835                 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
 836         } endfor_nexthops(fi);
 837 }
 838 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 839
 840 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 841                        int remaining, struct fib_config *cfg,
 842                        struct netlink_ext_ack *extack)
 843 {
 844         NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
 845
 846         return -EINVAL;
 847 }
 848
 849 #define fib_rebalance(fi) do { } while (0)
 850
 851 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 852
 853 static int fib_encap_match(struct net *net, u16 encap_type,
 854                            struct nlattr *encap,
 855                            const struct fib_nh *nh,
 856                            const struct fib_config *cfg,
 857                            struct netlink_ext_ack *extack)
 858 {
 859         struct lwtunnel_state *lwtstate;
 860         int ret, result = 0;
 861
 862         if (encap_type == LWTUNNEL_ENCAP_NONE)
 863                 return 0;
 864
 865         ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
 866                                    cfg, &lwtstate, extack);
 867         if (!ret) {
 868                 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
 869                 lwtstate_free(lwtstate);
 870         }
 871
 872         return result;
 873 }
 874
 875 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
 876                  struct netlink_ext_ack *extack)
 877 {
 878 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 879         struct rtnexthop *rtnh;
 880         int remaining;
 881 #endif
 882
 883         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 884                 return 1;
 885
 886         if (cfg->fc_nh_id) {
 887                 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
 888                         return 0;
 889                 return 1;
 890         }
 891
 892         if (fi->nh) {
 893                 if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
 894                         return 1;
 895                 return 0;
 896         }
 897
 898         if (cfg->fc_oif || cfg->fc_gw_family) {
 899                 struct fib_nh *nh;
 900
 901                 nh = fib_info_nh(fi, 0);
 902                 if (cfg->fc_encap) {
 903                         if (fib_encap_match(net, cfg->fc_encap_type,
 904                                             cfg->fc_encap, nh, cfg, extack))
 905                                 return 1;
 906                 }
 907 #ifdef CONFIG_IP_ROUTE_CLASSID
 908                 if (cfg->fc_flow &&
 909                     cfg->fc_flow != nh->nh_tclassid)
 910                         return 1;
 911 #endif
 912                 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
 913                     (cfg->fc_gw_family &&
 914                      cfg->fc_gw_family != nh->fib_nh_gw_family))
 915                         return 1;
 916
 917                 if (cfg->fc_gw_family == AF_INET &&
 918                     cfg->fc_gw4 != nh->fib_nh_gw4)
 919                         return 1;
 920
 921                 if (cfg->fc_gw_family == AF_INET6 &&
 922                     ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
 923                         return 1;
 924
 925                 return 0;
 926         }
 927
 928 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 929         if (!cfg->fc_mp)
 930                 return 0;
 931
 932         rtnh = cfg->fc_mp;
 933         remaining = cfg->fc_mp_len;
 934
 935         for_nexthops(fi) {
 936                 int attrlen;
 937
 938                 if (!rtnh_ok(rtnh, remaining))
 939                         return -EINVAL;
 940
 941                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
 942                         return 1;
 943
 944                 attrlen = rtnh_attrlen(rtnh);
 945                 if (attrlen > 0) {
 946                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 947                         int err;
 948
 949                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 950                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 951                         if (nla && nlav) {
 952                                 NL_SET_ERR_MSG(extack,
 953                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 954                                 return -EINVAL;
 955                         }
 956
 957                         if (nla) {
 958                                 __be32 gw;
 959
 960                                 err = fib_gw_from_attr(&gw, nla, extack);
 961                                 if (err)
 962                                         return err;
 963
 964                                 if (nh->fib_nh_gw_family != AF_INET ||
 965                                     gw != nh->fib_nh_gw4)
 966                                         return 1;
 967                         } else if (nlav) {
 968                                 struct fib_config cfg2;
 969
 970                                 err = fib_gw_from_via(&cfg2, nlav, extack);
 971                                 if (err)
 972                                         return err;
 973
 974                                 switch (nh->fib_nh_gw_family) {
 975                                 case AF_INET:
 976                                         if (cfg2.fc_gw_family != AF_INET ||
 977                                             cfg2.fc_gw4 != nh->fib_nh_gw4)
 978                                                 return 1;
 979                                         break;
 980                                 case AF_INET6:
 981                                         if (cfg2.fc_gw_family != AF_INET6 ||
 982                                             ipv6_addr_cmp(&cfg2.fc_gw6,
 983                                                           &nh->fib_nh_gw6))
 984                                                 return 1;
 985                                         break;
 986                                 }
 987                         }
 988
 989 #ifdef CONFIG_IP_ROUTE_CLASSID
 990                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 991                         if (nla) {
 992                                 if (nla_len(nla) < sizeof(u32)) {
 993                                         NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
 994                                         return -EINVAL;
 995                                 }
 996                                 if (nla_get_u32(nla) != nh->nh_tclassid)
 997                                         return 1;
 998                         }
 999 #endif
1000                 }
1001
1002                 rtnh = rtnh_next(rtnh, &remaining);
1003         } endfor_nexthops(fi);
1004 #endif
1005         return 0;
1006 }
1007
1008 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1009 {
1010         struct nlattr *nla;
1011         int remaining;
1012
1013         if (!cfg->fc_mx)
1014                 return true;
1015
1016         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1017                 int type = nla_type(nla);
1018                 u32 fi_val, val;
1019
1020                 if (!type)
1021                         continue;
1022                 if (type > RTAX_MAX)
1023                         return false;
1024
1025                 type = array_index_nospec(type, RTAX_MAX + 1);
1026                 if (type == RTAX_CC_ALGO) {
1027                         char tmp[TCP_CA_NAME_MAX];
1028                         bool ecn_ca = false;
1029
1030                         nla_strscpy(tmp, nla, sizeof(tmp));
1031                         val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1032                 } else {
1033                         if (nla_len(nla) != sizeof(u32))
1034                                 return false;
1035                         val = nla_get_u32(nla);
1036                 }
1037
1038                 fi_val = fi->fib_metrics->metrics[type - 1];
1039                 if (type == RTAX_FEATURES)
1040                         fi_val &= ~DST_FEATURE_ECN_CA;
1041
1042                 if (fi_val != val)
1043                         return false;
1044         }
1045
1046         return true;
1047 }
1048
1049 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1050                               u32 table, struct netlink_ext_ack *extack)
1051 {
1052         struct fib6_config cfg = {
1053                 .fc_table = table,
1054                 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1055                 .fc_ifindex = nh->fib_nh_oif,
1056                 .fc_gateway = nh->fib_nh_gw6,
1057         };
1058         struct fib6_nh fib6_nh = {};
1059         int err;
1060
1061         err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1062         if (!err) {
1063                 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1064                 dev_hold(nh->fib_nh_dev);
1065                 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1066                 nh->fib_nh_scope = RT_SCOPE_LINK;
1067
1068                 ipv6_stub->fib6_nh_release(&fib6_nh);
1069         }
1070
1071         return err;
1072 }
1073
1074 /*
1075  * Picture
1076  * -------
1077  *
1078  * Semantics of nexthop is very messy by historical reasons.
1079  * We have to take into account, that:
1080  * a) gateway can be actually local interface address,
1081  *    so that gatewayed route is direct.
1082  * b) gateway must be on-link address, possibly
1083  *    described not by an ifaddr, but also by a direct route.
1084  * c) If both gateway and interface are specified, they should not
1085  *    contradict.
1086  * d) If we use tunnel routes, gateway could be not on-link.
1087  *
1088  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1089  * results in pretty ugly and hairy code with obscure logic.
1090  *
1091  * I chose to generalized it instead, so that the size
1092  * of code does not increase practically, but it becomes
1093  * much more general.
1094  * Every prefix is assigned a "scope" value: "host" is local address,
1095  * "link" is direct route,
1096  * [ ... "site" ... "interior" ... ]
1097  * and "universe" is true gateway route with global meaning.
1098  *
1099  * Every prefix refers to a set of "nexthop"s (gw, oif),
1100  * where gw must have narrower scope. This recursion stops
1101  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1102  * which means that gw is forced to be on link.
1103  *
1104  * Code is still hairy, but now it is apparently logically
1105  * consistent and very flexible. F.e. as by-product it allows
1106  * to co-exists in peace independent exterior and interior
1107  * routing processes.
1108  *
1109  * Normally it looks as following.
1110  *
1111  * {universe prefix}  -> (gw, oif) [scope link]
1112  *                |
1113  *                |-> {link prefix} -> (gw, oif) [scope local]
1114  *                                      |
1115  *                                      |-> {local prefix} (terminal node)
1116  */
1117 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1118                               u8 scope, struct netlink_ext_ack *extack)
1119 {
1120         struct net_device *dev;
1121         struct fib_result res;
1122         int err = 0;
1123
1124         if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1125                 unsigned int addr_type;
1126
1127                 if (scope >= RT_SCOPE_LINK) {
1128                         NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1129                         return -EINVAL;
1130                 }
1131                 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1132                 if (!dev) {
1133                         NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1134                         return -ENODEV;
1135                 }
1136                 if (!(dev->flags & IFF_UP)) {
1137                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1138                         return -ENETDOWN;
1139                 }
1140                 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1141                 if (addr_type != RTN_UNICAST) {
1142                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1143                         return -EINVAL;
1144                 }
1145                 if (!netif_carrier_ok(dev))
1146                         nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1147                 nh->fib_nh_dev = dev;
1148                 dev_hold(dev);
1149                 nh->fib_nh_scope = RT_SCOPE_LINK;
1150                 return 0;
1151         }
1152         rcu_read_lock();
1153         {
1154                 struct fib_table *tbl = NULL;
1155                 struct flowi4 fl4 = {
1156                         .daddr = nh->fib_nh_gw4,
1157                         .flowi4_scope = scope + 1,
1158                         .flowi4_oif = nh->fib_nh_oif,
1159                         .flowi4_iif = LOOPBACK_IFINDEX,
1160                 };
1161
1162                 /* It is not necessary, but requires a bit of thinking */
1163                 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1164                         fl4.flowi4_scope = RT_SCOPE_LINK;
1165
1166                 if (table && table != RT_TABLE_MAIN)
1167                         tbl = fib_get_table(net, table);
1168
1169                 if (tbl)
1170                         err = fib_table_lookup(tbl, &fl4, &res,
1171                                                FIB_LOOKUP_IGNORE_LINKSTATE |
1172                                                FIB_LOOKUP_NOREF);
1173
1174                 /* on error or if no table given do full lookup. This
1175                  * is needed for example when nexthops are in the local
1176                  * table rather than the given table
1177                  */
1178                 if (!tbl || err) {
1179                         err = fib_lookup(net, &fl4, &res,
1180                                          FIB_LOOKUP_IGNORE_LINKSTATE);
1181                 }
1182
1183                 if (err) {
1184                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1185                         goto out;
1186                 }
1187         }
1188
1189         err = -EINVAL;
1190         if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1191                 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1192                 goto out;
1193         }
1194         nh->fib_nh_scope = res.scope;
1195         nh->fib_nh_oif = FIB_RES_OIF(res);
1196         nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1197         if (!dev) {
1198                 NL_SET_ERR_MSG(extack,
1199                                "No egress device for nexthop gateway");
1200                 goto out;
1201         }
1202         dev_hold(dev);
1203         if (!netif_carrier_ok(dev))
1204                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1205         err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1206 out:
1207         rcu_read_unlock();
1208         return err;
1209 }
1210
1211 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1212                               struct netlink_ext_ack *extack)
1213 {
1214         struct in_device *in_dev;
1215         int err;
1216
1217         if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1218                 NL_SET_ERR_MSG(extack,
1219                                "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1220                 return -EINVAL;
1221         }
1222
1223         rcu_read_lock();
1224
1225         err = -ENODEV;
1226         in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1227         if (!in_dev)
1228                 goto out;
1229         err = -ENETDOWN;
1230         if (!(in_dev->dev->flags & IFF_UP)) {
1231                 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1232                 goto out;
1233         }
1234
1235         nh->fib_nh_dev = in_dev->dev;
1236         dev_hold(nh->fib_nh_dev);
1237         nh->fib_nh_scope = RT_SCOPE_LINK;
1238         if (!netif_carrier_ok(nh->fib_nh_dev))
1239                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1240         err = 0;
1241 out:
1242         rcu_read_unlock();
1243         return err;
1244 }
1245
1246 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1247                  struct netlink_ext_ack *extack)
1248 {
1249         int err;
1250
1251         if (nh->fib_nh_gw_family == AF_INET)
1252                 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1253         else if (nh->fib_nh_gw_family == AF_INET6)
1254                 err = fib_check_nh_v6_gw(net, nh, table, extack);
1255         else
1256                 err = fib_check_nh_nongw(net, nh, extack);
1257
1258         return err;
1259 }
1260
1261 static inline unsigned int fib_laddr_hashfn(__be32 val)
1262 {
1263         unsigned int mask = (fib_info_hash_size - 1);
1264
1265         return ((__force u32)val ^
1266                 ((__force u32)val >> 7) ^
1267                 ((__force u32)val >> 14)) & mask;
1268 }
1269
1270 static struct hlist_head *fib_info_hash_alloc(int bytes)
1271 {
1272         if (bytes <= PAGE_SIZE)
1273                 return kzalloc(bytes, GFP_KERNEL);
1274         else
1275                 return (struct hlist_head *)
1276                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1277                                          get_order(bytes));
1278 }
1279
1280 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1281 {
1282         if (!hash)
1283                 return;
1284
1285         if (bytes <= PAGE_SIZE)
1286                 kfree(hash);
1287         else
1288                 free_pages((unsigned long) hash, get_order(bytes));
1289 }
1290
1291 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1292                                struct hlist_head *new_laddrhash,
1293                                unsigned int new_size)
1294 {
1295         struct hlist_head *old_info_hash, *old_laddrhash;
1296         unsigned int old_size = fib_info_hash_size;
1297         unsigned int i, bytes;
1298
1299         spin_lock_bh(&fib_info_lock);
1300         old_info_hash = fib_info_hash;
1301         old_laddrhash = fib_info_laddrhash;
1302         fib_info_hash_size = new_size;
1303
1304         for (i = 0; i < old_size; i++) {
1305                 struct hlist_head *head = &fib_info_hash[i];
1306                 struct hlist_node *n;
1307                 struct fib_info *fi;
1308
1309                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1310                         struct hlist_head *dest;
1311                         unsigned int new_hash;
1312
1313                         new_hash = fib_info_hashfn(fi);
1314                         dest = &new_info_hash[new_hash];
1315                         hlist_add_head(&fi->fib_hash, dest);
1316                 }
1317         }
1318         fib_info_hash = new_info_hash;
1319
1320         for (i = 0; i < old_size; i++) {
1321                 struct hlist_head *lhead = &fib_info_laddrhash[i];
1322                 struct hlist_node *n;
1323                 struct fib_info *fi;
1324
1325                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1326                         struct hlist_head *ldest;
1327                         unsigned int new_hash;
1328
1329                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1330                         ldest = &new_laddrhash[new_hash];
1331                         hlist_add_head(&fi->fib_lhash, ldest);
1332                 }
1333         }
1334         fib_info_laddrhash = new_laddrhash;
1335
1336         spin_unlock_bh(&fib_info_lock);
1337
1338         bytes = old_size * sizeof(struct hlist_head *);
1339         fib_info_hash_free(old_info_hash, bytes);
1340         fib_info_hash_free(old_laddrhash, bytes);
1341 }
1342
1343 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1344                                  unsigned char scope)
1345 {
1346         struct fib_nh *nh;
1347         __be32 saddr;
1348
1349         if (nhc->nhc_family != AF_INET)
1350                 return inet_select_addr(nhc->nhc_dev, 0, scope);
1351
1352         nh = container_of(nhc, struct fib_nh, nh_common);
1353         saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1354
1355         WRITE_ONCE(nh->nh_saddr, saddr);
1356         WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid));
1357
1358         return saddr;
1359 }
1360
1361 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1362 {
1363         struct fib_nh_common *nhc = res->nhc;
1364
1365         if (res->fi->fib_prefsrc)
1366                 return res->fi->fib_prefsrc;
1367
1368         if (nhc->nhc_family == AF_INET) {
1369                 struct fib_nh *nh;
1370
1371                 nh = container_of(nhc, struct fib_nh, nh_common);
1372                 if (READ_ONCE(nh->nh_saddr_genid) ==
1373                     atomic_read(&net->ipv4.dev_addr_genid))
1374                         return READ_ONCE(nh->nh_saddr);
1375         }
1376
1377         return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1378 }
1379
1380 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1381 {
1382         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1383             fib_prefsrc != cfg->fc_dst) {
1384                 u32 tb_id = cfg->fc_table;
1385                 int rc;
1386
1387                 if (tb_id == RT_TABLE_MAIN)
1388                         tb_id = RT_TABLE_LOCAL;
1389
1390                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1391                                           fib_prefsrc, tb_id);
1392
1393                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1394                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1395                                                   fib_prefsrc, RT_TABLE_LOCAL);
1396                 }
1397
1398                 if (rc != RTN_LOCAL)
1399                         return false;
1400         }
1401         return true;
1402 }
1403
1404 struct fib_info *fib_create_info(struct fib_config *cfg,
1405                                  struct netlink_ext_ack *extack)
1406 {
1407         int err;
1408         struct fib_info *fi = NULL;
1409         struct nexthop *nh = NULL;
1410         struct fib_info *ofi;
1411         int nhs = 1;
1412         struct net *net = cfg->fc_nlinfo.nl_net;
1413
1414         if (cfg->fc_type > RTN_MAX)
1415                 goto err_inval;
1416
1417         /* Fast check to catch the most weird cases */
1418         if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1419                 NL_SET_ERR_MSG(extack, "Invalid scope");
1420                 goto err_inval;
1421         }
1422
1423         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1424                 NL_SET_ERR_MSG(extack,
1425                                "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1426                 goto err_inval;
1427         }
1428
1429         if (cfg->fc_nh_id) {
1430                 if (!cfg->fc_mx) {
1431                         fi = fib_find_info_nh(net, cfg);
1432                         if (fi) {
1433                                 refcount_inc(&fi->fib_treeref);
1434                                 return fi;
1435                         }
1436                 }
1437
1438                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1439                 if (!nh) {
1440                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1441                         goto err_inval;
1442                 }
1443                 nhs = 0;
1444         }
1445
1446 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1447         if (cfg->fc_mp) {
1448                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1449                 if (nhs == 0)
1450                         goto err_inval;
1451         }
1452 #endif
1453
1454         err = -ENOBUFS;
1455
1456         /* Paired with WRITE_ONCE() in fib_release_info() */
1457         if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1458                 unsigned int new_size = fib_info_hash_size << 1;
1459                 struct hlist_head *new_info_hash;
1460                 struct hlist_head *new_laddrhash;
1461                 unsigned int bytes;
1462
1463                 if (!new_size)
1464                         new_size = 16;
1465                 bytes = new_size * sizeof(struct hlist_head *);
1466                 new_info_hash = fib_info_hash_alloc(bytes);
1467                 new_laddrhash = fib_info_hash_alloc(bytes);
1468                 if (!new_info_hash || !new_laddrhash) {
1469                         fib_info_hash_free(new_info_hash, bytes);
1470                         fib_info_hash_free(new_laddrhash, bytes);
1471                 } else
1472                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1473
1474                 if (!fib_info_hash_size)
1475                         goto failure;
1476         }
1477
1478         fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1479         if (!fi)
1480                 goto failure;
1481         fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1482                                               cfg->fc_mx_len, extack);
1483         if (IS_ERR(fi->fib_metrics)) {
1484                 err = PTR_ERR(fi->fib_metrics);
1485                 kfree(fi);
1486                 return ERR_PTR(err);
1487         }
1488
1489         fi->fib_net = net;
1490         fi->fib_protocol = cfg->fc_protocol;
1491         fi->fib_scope = cfg->fc_scope;
1492         fi->fib_flags = cfg->fc_flags;
1493         fi->fib_priority = cfg->fc_priority;
1494         fi->fib_prefsrc = cfg->fc_prefsrc;
1495         fi->fib_type = cfg->fc_type;
1496         fi->fib_tb_id = cfg->fc_table;
1497
1498         fi->fib_nhs = nhs;
1499         if (nh) {
1500                 if (!nexthop_get(nh)) {
1501                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1502                         err = -EINVAL;
1503                 } else {
1504                         err = 0;
1505                         fi->nh = nh;
1506                 }
1507         } else {
1508                 change_nexthops(fi) {
1509                         nexthop_nh->nh_parent = fi;
1510                 } endfor_nexthops(fi)
1511
1512                 if (cfg->fc_mp)
1513                         err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1514                                           extack);
1515                 else
1516                         err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1517         }
1518
1519         if (err != 0)
1520                 goto failure;
1521
1522         if (fib_props[cfg->fc_type].error) {
1523                 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1524                         NL_SET_ERR_MSG(extack,
1525                                        "Gateway, device and multipath can not be specified for this route type");
1526                         goto err_inval;
1527                 }
1528                 goto link_it;
1529         } else {
1530                 switch (cfg->fc_type) {
1531                 case RTN_UNICAST:
1532                 case RTN_LOCAL:
1533                 case RTN_BROADCAST:
1534                 case RTN_ANYCAST:
1535                 case RTN_MULTICAST:
1536                         break;
1537                 default:
1538                         NL_SET_ERR_MSG(extack, "Invalid route type");
1539                         goto err_inval;
1540                 }
1541         }
1542
1543         if (cfg->fc_scope > RT_SCOPE_HOST) {
1544                 NL_SET_ERR_MSG(extack, "Invalid scope");
1545                 goto err_inval;
1546         }
1547
1548         if (fi->nh) {
1549                 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1550                 if (err)
1551                         goto failure;
1552         } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1553                 struct fib_nh *nh = fi->fib_nh;
1554
1555                 /* Local address is added. */
1556                 if (nhs != 1) {
1557                         NL_SET_ERR_MSG(extack,
1558                                        "Route with host scope can not have multiple nexthops");
1559                         goto err_inval;
1560                 }
1561                 if (nh->fib_nh_gw_family) {
1562                         NL_SET_ERR_MSG(extack,
1563                                        "Route with host scope can not have a gateway");
1564                         goto err_inval;
1565                 }
1566                 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1567                 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1568                 err = -ENODEV;
1569                 if (!nh->fib_nh_dev)
1570                         goto failure;
1571         } else {
1572                 int linkdown = 0;
1573
1574                 change_nexthops(fi) {
1575                         err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1576                                            cfg->fc_table, cfg->fc_scope,
1577                                            extack);
1578                         if (err != 0)
1579                                 goto failure;
1580                         if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1581                                 linkdown++;
1582                 } endfor_nexthops(fi)
1583                 if (linkdown == fi->fib_nhs)
1584                         fi->fib_flags |= RTNH_F_LINKDOWN;
1585         }
1586
1587         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1588                 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1589                 goto err_inval;
1590         }
1591
1592         if (!fi->nh) {
1593                 change_nexthops(fi) {
1594                         fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1595                                                   fi->fib_scope);
1596                         if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1597                                 fi->fib_nh_is_v6 = true;
1598                 } endfor_nexthops(fi)
1599
1600                 fib_rebalance(fi);
1601         }
1602
1603 link_it:
1604         ofi = fib_find_info(fi);
1605         if (ofi) {
1606                 /* fib_table_lookup() should not see @fi yet. */
1607                 fi->fib_dead = 1;
1608                 free_fib_info(fi);
1609                 refcount_inc(&ofi->fib_treeref);
1610                 return ofi;
1611         }
1612
1613         refcount_set(&fi->fib_treeref, 1);
1614         refcount_set(&fi->fib_clntref, 1);
1615         spin_lock_bh(&fib_info_lock);
1616         fib_info_cnt++;
1617         hlist_add_head(&fi->fib_hash,
1618                        &fib_info_hash[fib_info_hashfn(fi)]);
1619         if (fi->fib_prefsrc) {
1620                 struct hlist_head *head;
1621
1622                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1623                 hlist_add_head(&fi->fib_lhash, head);
1624         }
1625         if (fi->nh) {
1626                 list_add(&fi->nh_list, &nh->fi_list);
1627         } else {
1628                 change_nexthops(fi) {
1629                         struct hlist_head *head;
1630
1631                         if (!nexthop_nh->fib_nh_dev)
1632                                 continue;
1633                         head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1634                         hlist_add_head(&nexthop_nh->nh_hash, head);
1635                 } endfor_nexthops(fi)
1636         }
1637         spin_unlock_bh(&fib_info_lock);
1638         return fi;
1639
1640 err_inval:
1641         err = -EINVAL;
1642
1643 failure:
1644         if (fi) {
1645                 /* fib_table_lookup() should not see @fi yet. */
1646                 fi->fib_dead = 1;
1647                 free_fib_info(fi);
1648         }
1649
1650         return ERR_PTR(err);
1651 }
1652
1653 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1654                      u8 rt_family, unsigned char *flags, bool skip_oif)
1655 {
1656         if (nhc->nhc_flags & RTNH_F_DEAD)
1657                 *flags |= RTNH_F_DEAD;
1658
1659         if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1660                 *flags |= RTNH_F_LINKDOWN;
1661
1662                 rcu_read_lock();
1663                 switch (nhc->nhc_family) {
1664                 case AF_INET:
1665                         if (ip_ignore_linkdown(nhc->nhc_dev))
1666                                 *flags |= RTNH_F_DEAD;
1667                         break;
1668                 case AF_INET6:
1669                         if (ip6_ignore_linkdown(nhc->nhc_dev))
1670                                 *flags |= RTNH_F_DEAD;
1671                         break;
1672                 }
1673                 rcu_read_unlock();
1674         }
1675
1676         switch (nhc->nhc_gw_family) {
1677         case AF_INET:
1678                 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1679                         goto nla_put_failure;
1680                 break;
1681         case AF_INET6:
1682                 /* if gateway family does not match nexthop family
1683                  * gateway is encoded as RTA_VIA
1684                  */
1685                 if (rt_family != nhc->nhc_gw_family) {
1686                         int alen = sizeof(struct in6_addr);
1687                         struct nlattr *nla;
1688                         struct rtvia *via;
1689
1690                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
1691                         if (!nla)
1692                                 goto nla_put_failure;
1693
1694                         via = nla_data(nla);
1695                         via->rtvia_family = AF_INET6;
1696                         memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1697                 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1698                                             &nhc->nhc_gw.ipv6) < 0) {
1699                         goto nla_put_failure;
1700                 }
1701                 break;
1702         }
1703
1704         *flags |= (nhc->nhc_flags &
1705                    (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1706
1707         if (!skip_oif && nhc->nhc_dev &&
1708             nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1709                 goto nla_put_failure;
1710
1711         if (nhc->nhc_lwtstate &&
1712             lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1713                                 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1714                 goto nla_put_failure;
1715
1716         return 0;
1717
1718 nla_put_failure:
1719         return -EMSGSIZE;
1720 }
1721 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1722
1723 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1724 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1725                     int nh_weight, u8 rt_family, u32 nh_tclassid)
1726 {
1727         const struct net_device *dev = nhc->nhc_dev;
1728         struct rtnexthop *rtnh;
1729         unsigned char flags = 0;
1730
1731         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1732         if (!rtnh)
1733                 goto nla_put_failure;
1734
1735         rtnh->rtnh_hops = nh_weight - 1;
1736         rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1737
1738         if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1739                 goto nla_put_failure;
1740
1741         rtnh->rtnh_flags = flags;
1742
1743         if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1744                 goto nla_put_failure;
1745
1746         /* length of rtnetlink header + attributes */
1747         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1748
1749         return 0;
1750
1751 nla_put_failure:
1752         return -EMSGSIZE;
1753 }
1754 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1755 #endif
1756
1757 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1758 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1759 {
1760         struct nlattr *mp;
1761
1762         mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1763         if (!mp)
1764                 goto nla_put_failure;
1765
1766         if (unlikely(fi->nh)) {
1767                 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1768                         goto nla_put_failure;
1769                 goto mp_end;
1770         }
1771
1772         for_nexthops(fi) {
1773                 u32 nh_tclassid = 0;
1774 #ifdef CONFIG_IP_ROUTE_CLASSID
1775                 nh_tclassid = nh->nh_tclassid;
1776 #endif
1777                 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1778                                     AF_INET, nh_tclassid) < 0)
1779                         goto nla_put_failure;
1780         } endfor_nexthops(fi);
1781
1782 mp_end:
1783         nla_nest_end(skb, mp);
1784
1785         return 0;
1786
1787 nla_put_failure:
1788         return -EMSGSIZE;
1789 }
1790 #else
1791 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1792 {
1793         return 0;
1794 }
1795 #endif
1796
1797 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1798                   const struct fib_rt_info *fri, unsigned int flags)
1799 {
1800         unsigned int nhs = fib_info_num_path(fri->fi);
1801         struct fib_info *fi = fri->fi;
1802         u32 tb_id = fri->tb_id;
1803         struct nlmsghdr *nlh;
1804         struct rtmsg *rtm;
1805
1806         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1807         if (!nlh)
1808                 return -EMSGSIZE;
1809
1810         rtm = nlmsg_data(nlh);
1811         rtm->rtm_family = AF_INET;
1812         rtm->rtm_dst_len = fri->dst_len;
1813         rtm->rtm_src_len = 0;
1814         rtm->rtm_tos = fri->tos;
1815         if (tb_id < 256)
1816                 rtm->rtm_table = tb_id;
1817         else
1818                 rtm->rtm_table = RT_TABLE_COMPAT;
1819         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1820                 goto nla_put_failure;
1821         rtm->rtm_type = fri->type;
1822         rtm->rtm_flags = fi->fib_flags;
1823         rtm->rtm_scope = fi->fib_scope;
1824         rtm->rtm_protocol = fi->fib_protocol;
1825
1826         if (rtm->rtm_dst_len &&
1827             nla_put_in_addr(skb, RTA_DST, fri->dst))
1828                 goto nla_put_failure;
1829         if (fi->fib_priority &&
1830             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1831                 goto nla_put_failure;
1832         if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1833                 goto nla_put_failure;
1834
1835         if (fi->fib_prefsrc &&
1836             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1837                 goto nla_put_failure;
1838
1839         if (fi->nh) {
1840                 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1841                         goto nla_put_failure;
1842                 if (nexthop_is_blackhole(fi->nh))
1843                         rtm->rtm_type = RTN_BLACKHOLE;
1844                 if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1845                         goto offload;
1846         }
1847
1848         if (nhs == 1) {
1849                 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1850                 unsigned char flags = 0;
1851
1852                 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1853                         goto nla_put_failure;
1854
1855                 rtm->rtm_flags = flags;
1856 #ifdef CONFIG_IP_ROUTE_CLASSID
1857                 if (nhc->nhc_family == AF_INET) {
1858                         struct fib_nh *nh;
1859
1860                         nh = container_of(nhc, struct fib_nh, nh_common);
1861                         if (nh->nh_tclassid &&
1862                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1863                                 goto nla_put_failure;
1864                 }
1865 #endif
1866         } else {
1867                 if (fib_add_multipath(skb, fi) < 0)
1868                         goto nla_put_failure;
1869         }
1870
1871 offload:
1872         if (fri->offload)
1873                 rtm->rtm_flags |= RTM_F_OFFLOAD;
1874         if (fri->trap)
1875                 rtm->rtm_flags |= RTM_F_TRAP;
1876         if (fri->offload_failed)
1877                 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1878
1879         nlmsg_end(skb, nlh);
1880         return 0;
1881
1882 nla_put_failure:
1883         nlmsg_cancel(skb, nlh);
1884         return -EMSGSIZE;
1885 }
1886
1887 /*
1888  * Update FIB if:
1889  * - local address disappeared -> we must delete all the entries
1890  *   referring to it.
1891  * - device went down -> we must shutdown all nexthops going via it.
1892  */
1893 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1894 {
1895         int ret = 0;
1896         unsigned int hash = fib_laddr_hashfn(local);
1897         struct hlist_head *head = &fib_info_laddrhash[hash];
1898         int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1899         struct net *net = dev_net(dev);
1900         struct fib_info *fi;
1901
1902         if (!fib_info_laddrhash || local == 0)
1903                 return 0;
1904
1905         hlist_for_each_entry(fi, head, fib_lhash) {
1906                 if (!net_eq(fi->fib_net, net) ||
1907                     fi->fib_tb_id != tb_id)
1908                         continue;
1909                 if (fi->fib_prefsrc == local) {
1910                         fi->fib_flags |= RTNH_F_DEAD;
1911                         fi->pfsrc_removed = true;
1912                         ret++;
1913                 }
1914         }
1915         return ret;
1916 }
1917
1918 static int call_fib_nh_notifiers(struct fib_nh *nh,
1919                                  enum fib_event_type event_type)
1920 {
1921         bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1922         struct fib_nh_notifier_info info = {
1923                 .fib_nh = nh,
1924         };
1925
1926         switch (event_type) {
1927         case FIB_EVENT_NH_ADD:
1928                 if (nh->fib_nh_flags & RTNH_F_DEAD)
1929                         break;
1930                 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1931                         break;
1932                 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1933                                            &info.info);
1934         case FIB_EVENT_NH_DEL:
1935                 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1936                     (nh->fib_nh_flags & RTNH_F_DEAD))
1937                         return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1938                                                    event_type, &info.info);
1939                 break;
1940         default:
1941                 break;
1942         }
1943
1944         return NOTIFY_DONE;
1945 }
1946
1947 /* Update the PMTU of exceptions when:
1948  * - the new MTU of the first hop becomes smaller than the PMTU
1949  * - the old MTU was the same as the PMTU, and it limited discovery of
1950  *   larger MTUs on the path. With that limit raised, we can now
1951  *   discover larger MTUs
1952  * A special case is locked exceptions, for which the PMTU is smaller
1953  * than the minimal accepted PMTU:
1954  * - if the new MTU is greater than the PMTU, don't make any change
1955  * - otherwise, unlock and set PMTU
1956  */
1957 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1958 {
1959         struct fnhe_hash_bucket *bucket;
1960         int i;
1961
1962         bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1963         if (!bucket)
1964                 return;
1965
1966         for (i = 0; i < FNHE_HASH_SIZE; i++) {
1967                 struct fib_nh_exception *fnhe;
1968
1969                 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1970                      fnhe;
1971                      fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1972                         if (fnhe->fnhe_mtu_locked) {
1973                                 if (new <= fnhe->fnhe_pmtu) {
1974                                         fnhe->fnhe_pmtu = new;
1975                                         fnhe->fnhe_mtu_locked = false;
1976                                 }
1977                         } else if (new < fnhe->fnhe_pmtu ||
1978                                    orig == fnhe->fnhe_pmtu) {
1979                                 fnhe->fnhe_pmtu = new;
1980                         }
1981                 }
1982         }
1983 }
1984
1985 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1986 {
1987         struct hlist_head *head = fib_info_devhash_bucket(dev);
1988         struct fib_nh *nh;
1989
1990         hlist_for_each_entry(nh, head, nh_hash) {
1991                 if (nh->fib_nh_dev == dev)
1992                         fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1993         }
1994 }
1995
1996 /* Event              force Flags           Description
1997  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1998  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1999  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
2000  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
2001  *
2002  * only used when fib_nh is built into fib_info
2003  */
2004 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
2005 {
2006         struct hlist_head *head = fib_info_devhash_bucket(dev);
2007         struct fib_info *prev_fi = NULL;
2008         int scope = RT_SCOPE_NOWHERE;
2009         struct fib_nh *nh;
2010         int ret = 0;
2011
2012         if (force)
2013                 scope = -1;
2014
2015         hlist_for_each_entry(nh, head, nh_hash) {
2016                 struct fib_info *fi = nh->nh_parent;
2017                 int dead;
2018
2019                 BUG_ON(!fi->fib_nhs);
2020                 if (nh->fib_nh_dev != dev || fi == prev_fi)
2021                         continue;
2022                 prev_fi = fi;
2023                 dead = 0;
2024                 change_nexthops(fi) {
2025                         if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2026                                 dead++;
2027                         else if (nexthop_nh->fib_nh_dev == dev &&
2028                                  nexthop_nh->fib_nh_scope != scope) {
2029                                 switch (event) {
2030                                 case NETDEV_DOWN:
2031                                 case NETDEV_UNREGISTER:
2032                                         nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2033                                         fallthrough;
2034                                 case NETDEV_CHANGE:
2035                                         nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2036                                         break;
2037                                 }
2038                                 call_fib_nh_notifiers(nexthop_nh,
2039                                                       FIB_EVENT_NH_DEL);
2040                                 dead++;
2041                         }
2042 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2043                         if (event == NETDEV_UNREGISTER &&
2044                             nexthop_nh->fib_nh_dev == dev) {
2045                                 dead = fi->fib_nhs;
2046                                 break;
2047                         }
2048 #endif
2049                 } endfor_nexthops(fi)
2050                 if (dead == fi->fib_nhs) {
2051                         switch (event) {
2052                         case NETDEV_DOWN:
2053                         case NETDEV_UNREGISTER:
2054                                 fi->fib_flags |= RTNH_F_DEAD;
2055                                 fallthrough;
2056                         case NETDEV_CHANGE:
2057                                 fi->fib_flags |= RTNH_F_LINKDOWN;
2058                                 break;
2059                         }
2060                         ret++;
2061                 }
2062
2063                 fib_rebalance(fi);
2064         }
2065
2066         return ret;
2067 }
2068
2069 /* Must be invoked inside of an RCU protected region.  */
2070 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2071 {
2072         struct fib_info *fi = NULL, *last_resort = NULL;
2073         struct hlist_head *fa_head = res->fa_head;
2074         struct fib_table *tb = res->table;
2075         u8 slen = 32 - res->prefixlen;
2076         int order = -1, last_idx = -1;
2077         struct fib_alias *fa, *fa1 = NULL;
2078         u32 last_prio = res->fi->fib_priority;
2079         u8 last_tos = 0;
2080
2081         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2082                 struct fib_info *next_fi = fa->fa_info;
2083                 struct fib_nh_common *nhc;
2084
2085                 if (fa->fa_slen != slen)
2086                         continue;
2087                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2088                         continue;
2089                 if (fa->tb_id != tb->tb_id)
2090                         continue;
2091                 if (next_fi->fib_priority > last_prio &&
2092                     fa->fa_tos == last_tos) {
2093                         if (last_tos)
2094                                 continue;
2095                         break;
2096                 }
2097                 if (next_fi->fib_flags & RTNH_F_DEAD)
2098                         continue;
2099                 last_tos = fa->fa_tos;
2100                 last_prio = next_fi->fib_priority;
2101
2102                 if (next_fi->fib_scope != res->scope ||
2103                     fa->fa_type != RTN_UNICAST)
2104                         continue;
2105
2106                 nhc = fib_info_nhc(next_fi, 0);
2107                 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2108                         continue;
2109
2110                 fib_alias_accessed(fa);
2111
2112                 if (!fi) {
2113                         if (next_fi != res->fi)
2114                                 break;
2115                         fa1 = fa;
2116                 } else if (!fib_detect_death(fi, order, &last_resort,
2117                                              &last_idx, fa1->fa_default)) {
2118                         fib_result_assign(res, fi);
2119                         fa1->fa_default = order;
2120                         goto out;
2121                 }
2122                 fi = next_fi;
2123                 order++;
2124         }
2125
2126         if (order <= 0 || !fi) {
2127                 if (fa1)
2128                         fa1->fa_default = -1;
2129                 goto out;
2130         }
2131
2132         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2133                               fa1->fa_default)) {
2134                 fib_result_assign(res, fi);
2135                 fa1->fa_default = order;
2136                 goto out;
2137         }
2138
2139         if (last_idx >= 0)
2140                 fib_result_assign(res, last_resort);
2141         fa1->fa_default = last_idx;
2142 out:
2143         return;
2144 }
2145
2146 /*
2147  * Dead device goes up. We wake up dead nexthops.
2148  * It takes sense only on multipath routes.
2149  *
2150  * only used when fib_nh is built into fib_info
2151  */
2152 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2153 {
2154         struct fib_info *prev_fi;
2155         struct hlist_head *head;
2156         struct fib_nh *nh;
2157         int ret;
2158
2159         if (!(dev->flags & IFF_UP))
2160                 return 0;
2161
2162         if (nh_flags & RTNH_F_DEAD) {
2163                 unsigned int flags = dev_get_flags(dev);
2164
2165                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2166                         nh_flags |= RTNH_F_LINKDOWN;
2167         }
2168
2169         prev_fi = NULL;
2170         head = fib_info_devhash_bucket(dev);
2171         ret = 0;
2172
2173         hlist_for_each_entry(nh, head, nh_hash) {
2174                 struct fib_info *fi = nh->nh_parent;
2175                 int alive;
2176
2177                 BUG_ON(!fi->fib_nhs);
2178                 if (nh->fib_nh_dev != dev || fi == prev_fi)
2179                         continue;
2180
2181                 prev_fi = fi;
2182                 alive = 0;
2183                 change_nexthops(fi) {
2184                         if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2185                                 alive++;
2186                                 continue;
2187                         }
2188                         if (!nexthop_nh->fib_nh_dev ||
2189                             !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2190                                 continue;
2191                         if (nexthop_nh->fib_nh_dev != dev ||
2192                             !__in_dev_get_rtnl(dev))
2193                                 continue;
2194                         alive++;
2195                         nexthop_nh->fib_nh_flags &= ~nh_flags;
2196                         call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2197                 } endfor_nexthops(fi)
2198
2199                 if (alive > 0) {
2200                         fi->fib_flags &= ~nh_flags;
2201                         ret++;
2202                 }
2203
2204                 fib_rebalance(fi);
2205         }
2206
2207         return ret;
2208 }
2209
2210 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2211 static bool fib_good_nh(const struct fib_nh *nh)
2212 {
2213         int state = NUD_REACHABLE;
2214
2215         if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2216                 struct neighbour *n;
2217
2218                 rcu_read_lock_bh();
2219
2220                 if (likely(nh->fib_nh_gw_family == AF_INET))
2221                         n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2222                                                    (__force u32)nh->fib_nh_gw4);
2223                 else if (nh->fib_nh_gw_family == AF_INET6)
2224                         n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2225                                                            &nh->fib_nh_gw6);
2226                 else
2227                         n = NULL;
2228                 if (n)
2229                         state = n->nud_state;
2230
2231                 rcu_read_unlock_bh();
2232         }
2233
2234         return !!(state & NUD_VALID);
2235 }
2236
2237 void fib_select_multipath(struct fib_result *res, int hash)
2238 {
2239         struct fib_info *fi = res->fi;
2240         struct net *net = fi->fib_net;
2241         bool first = false;
2242
2243         if (unlikely(res->fi->nh)) {
2244                 nexthop_path_fib_result(res, hash);
2245                 return;
2246         }
2247
2248         change_nexthops(fi) {
2249                 if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2250                         if (!fib_good_nh(nexthop_nh))
2251                                 continue;
2252                         if (!first) {
2253                                 res->nh_sel = nhsel;
2254                                 res->nhc = &nexthop_nh->nh_common;
2255                                 first = true;
2256                         }
2257                 }
2258
2259                 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2260                         continue;
2261
2262                 res->nh_sel = nhsel;
2263                 res->nhc = &nexthop_nh->nh_common;
2264                 return;
2265         } endfor_nexthops(fi);
2266 }
2267 #endif
2268
2269 void fib_select_path(struct net *net, struct fib_result *res,
2270                      struct flowi4 *fl4, const struct sk_buff *skb)
2271 {
2272         if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2273                 goto check_saddr;
2274
2275 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2276         if (fib_info_num_path(res->fi) > 1) {
2277                 int h = fib_multipath_hash(net, fl4, skb, NULL);
2278
2279                 fib_select_multipath(res, h);
2280         }
2281         else
2282 #endif
2283         if (!res->prefixlen &&
2284             res->table->tb_num_default > 1 &&
2285             res->type == RTN_UNICAST)
2286                 fib_select_default(fl4, res);
2287
2288 check_saddr:
2289         if (!fl4->saddr)
2290                 fl4->saddr = fib_result_prefsrc(net, res);
2291 }