linux/net/ipv4/fib_semantics.c
<<
>>
Prefs
   1/*
   2 * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3 *              operating system.  INET is implemented using the  BSD Socket
   4 *              interface as the means of communication with the user level.
   5 *
   6 *              IPv4 Forwarding Information Base: semantics.
   7 *
   8 * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
   9 *
  10 *              This program is free software; you can redistribute it and/or
  11 *              modify it under the terms of the GNU General Public License
  12 *              as published by the Free Software Foundation; either version
  13 *              2 of the License, or (at your option) any later version.
  14 */
  15
  16#include <asm/uaccess.h>
  17#include <linux/bitops.h>
  18#include <linux/types.h>
  19#include <linux/kernel.h>
  20#include <linux/jiffies.h>
  21#include <linux/mm.h>
  22#include <linux/string.h>
  23#include <linux/socket.h>
  24#include <linux/sockios.h>
  25#include <linux/errno.h>
  26#include <linux/in.h>
  27#include <linux/inet.h>
  28#include <linux/inetdevice.h>
  29#include <linux/netdevice.h>
  30#include <linux/if_arp.h>
  31#include <linux/proc_fs.h>
  32#include <linux/skbuff.h>
  33#include <linux/init.h>
  34#include <linux/slab.h>
  35
  36#include <net/arp.h>
  37#include <net/ip.h>
  38#include <net/protocol.h>
  39#include <net/route.h>
  40#include <net/tcp.h>
  41#include <net/sock.h>
  42#include <net/ip_fib.h>
  43#include <net/netlink.h>
  44#include <net/nexthop.h>
  45
  46#include "fib_lookup.h"
  47
  48static DEFINE_SPINLOCK(fib_info_lock);
  49static struct hlist_head *fib_info_hash;
  50static struct hlist_head *fib_info_laddrhash;
  51static unsigned int fib_info_hash_size;
  52static unsigned int fib_info_cnt;
  53
  54#define DEVINDEX_HASHBITS 8
  55#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  56static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  57
  58#ifdef CONFIG_IP_ROUTE_MULTIPATH
  59
  60static DEFINE_SPINLOCK(fib_multipath_lock);
  61
  62#define for_nexthops(fi) {                                              \
  63        int nhsel; const struct fib_nh *nh;                             \
  64        for (nhsel = 0, nh = (fi)->fib_nh;                              \
  65             nhsel < (fi)->fib_nhs;                                     \
  66             nh++, nhsel++)
  67
  68#define change_nexthops(fi) {                                           \
  69        int nhsel; struct fib_nh *nexthop_nh;                           \
  70        for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  71             nhsel < (fi)->fib_nhs;                                     \
  72             nexthop_nh++, nhsel++)
  73
  74#else /* CONFIG_IP_ROUTE_MULTIPATH */
  75
  76/* Hope, that gcc will optimize it to get rid of dummy loop */
  77
  78#define for_nexthops(fi) {                                              \
  79        int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  80        for (nhsel = 0; nhsel < 1; nhsel++)
  81
  82#define change_nexthops(fi) {                                           \
  83        int nhsel;                                                      \
  84        struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  85        for (nhsel = 0; nhsel < 1; nhsel++)
  86
  87#endif /* CONFIG_IP_ROUTE_MULTIPATH */
  88
  89#define endfor_nexthops(fi) }
  90
  91
  92const struct fib_prop fib_props[RTN_MAX + 1] = {
  93        [RTN_UNSPEC] = {
  94                .error  = 0,
  95                .scope  = RT_SCOPE_NOWHERE,
  96        },
  97        [RTN_UNICAST] = {
  98                .error  = 0,
  99                .scope  = RT_SCOPE_UNIVERSE,
 100        },
 101        [RTN_LOCAL] = {
 102                .error  = 0,
 103                .scope  = RT_SCOPE_HOST,
 104        },
 105        [RTN_BROADCAST] = {
 106                .error  = 0,
 107                .scope  = RT_SCOPE_LINK,
 108        },
 109        [RTN_ANYCAST] = {
 110                .error  = 0,
 111                .scope  = RT_SCOPE_LINK,
 112        },
 113        [RTN_MULTICAST] = {
 114                .error  = 0,
 115                .scope  = RT_SCOPE_UNIVERSE,
 116        },
 117        [RTN_BLACKHOLE] = {
 118                .error  = -EINVAL,
 119                .scope  = RT_SCOPE_UNIVERSE,
 120        },
 121        [RTN_UNREACHABLE] = {
 122                .error  = -EHOSTUNREACH,
 123                .scope  = RT_SCOPE_UNIVERSE,
 124        },
 125        [RTN_PROHIBIT] = {
 126                .error  = -EACCES,
 127                .scope  = RT_SCOPE_UNIVERSE,
 128        },
 129        [RTN_THROW] = {
 130                .error  = -EAGAIN,
 131                .scope  = RT_SCOPE_UNIVERSE,
 132        },
 133        [RTN_NAT] = {
 134                .error  = -EINVAL,
 135                .scope  = RT_SCOPE_NOWHERE,
 136        },
 137        [RTN_XRESOLVE] = {
 138                .error  = -EINVAL,
 139                .scope  = RT_SCOPE_NOWHERE,
 140        },
 141};
 142
 143static void rt_fibinfo_free(struct rtable __rcu **rtp)
 144{
 145        struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 146
 147        if (!rt)
 148                return;
 149
 150        /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 151         * because we waited an RCU grace period before calling
 152         * free_fib_info_rcu()
 153         */
 154
 155        dst_free(&rt->dst);
 156}
 157
 158static void free_nh_exceptions(struct fib_nh *nh)
 159{
 160        struct fnhe_hash_bucket *hash = nh->nh_exceptions;
 161        int i;
 162
 163        for (i = 0; i < FNHE_HASH_SIZE; i++) {
 164                struct fib_nh_exception *fnhe;
 165
 166                fnhe = rcu_dereference_protected(hash[i].chain, 1);
 167                while (fnhe) {
 168                        struct fib_nh_exception *next;
 169                        
 170                        next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 171
 172                        rt_fibinfo_free(&fnhe->fnhe_rth);
 173
 174                        kfree(fnhe);
 175
 176                        fnhe = next;
 177                }
 178        }
 179        kfree(hash);
 180}
 181
 182static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 183{
 184        int cpu;
 185
 186        if (!rtp)
 187                return;
 188
 189        for_each_possible_cpu(cpu) {
 190                struct rtable *rt;
 191
 192                rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 193                if (rt)
 194                        dst_free(&rt->dst);
 195        }
 196        free_percpu(rtp);
 197}
 198
 199/* Release a nexthop info record */
 200static void free_fib_info_rcu(struct rcu_head *head)
 201{
 202        struct fib_info *fi = container_of(head, struct fib_info, rcu);
 203
 204        change_nexthops(fi) {
 205                if (nexthop_nh->nh_dev)
 206                        dev_put(nexthop_nh->nh_dev);
 207                if (nexthop_nh->nh_exceptions)
 208                        free_nh_exceptions(nexthop_nh);
 209                rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 210                rt_fibinfo_free(&nexthop_nh->nh_rth_input);
 211        } endfor_nexthops(fi);
 212
 213        release_net(fi->fib_net);
 214        if (fi->fib_metrics != (u32 *) dst_default_metrics)
 215                kfree(fi->fib_metrics);
 216        kfree(fi);
 217}
 218
 219void free_fib_info(struct fib_info *fi)
 220{
 221        if (fi->fib_dead == 0) {
 222                pr_warn("Freeing alive fib_info %p\n", fi);
 223                return;
 224        }
 225        fib_info_cnt--;
 226#ifdef CONFIG_IP_ROUTE_CLASSID
 227        change_nexthops(fi) {
 228                if (nexthop_nh->nh_tclassid)
 229                        fi->fib_net->ipv4.fib_num_tclassid_users--;
 230        } endfor_nexthops(fi);
 231#endif
 232        call_rcu(&fi->rcu, free_fib_info_rcu);
 233}
 234
 235void fib_release_info(struct fib_info *fi)
 236{
 237        spin_lock_bh(&fib_info_lock);
 238        if (fi && --fi->fib_treeref == 0) {
 239                hlist_del(&fi->fib_hash);
 240                if (fi->fib_prefsrc)
 241                        hlist_del(&fi->fib_lhash);
 242                change_nexthops(fi) {
 243                        if (!nexthop_nh->nh_dev)
 244                                continue;
 245                        hlist_del(&nexthop_nh->nh_hash);
 246                } endfor_nexthops(fi)
 247                fi->fib_dead = 1;
 248                fib_info_put(fi);
 249        }
 250        spin_unlock_bh(&fib_info_lock);
 251}
 252
 253static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 254{
 255        const struct fib_nh *onh = ofi->fib_nh;
 256
 257        for_nexthops(fi) {
 258                if (nh->nh_oif != onh->nh_oif ||
 259                    nh->nh_gw  != onh->nh_gw ||
 260                    nh->nh_scope != onh->nh_scope ||
 261#ifdef CONFIG_IP_ROUTE_MULTIPATH
 262                    nh->nh_weight != onh->nh_weight ||
 263#endif
 264#ifdef CONFIG_IP_ROUTE_CLASSID
 265                    nh->nh_tclassid != onh->nh_tclassid ||
 266#endif
 267                    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
 268                        return -1;
 269                onh++;
 270        } endfor_nexthops(fi);
 271        return 0;
 272}
 273
 274static inline unsigned int fib_devindex_hashfn(unsigned int val)
 275{
 276        unsigned int mask = DEVINDEX_HASHSIZE - 1;
 277
 278        return (val ^
 279                (val >> DEVINDEX_HASHBITS) ^
 280                (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 281}
 282
 283static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 284{
 285        unsigned int mask = (fib_info_hash_size - 1);
 286        unsigned int val = fi->fib_nhs;
 287
 288        val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 289        val ^= (__force u32)fi->fib_prefsrc;
 290        val ^= fi->fib_priority;
 291        for_nexthops(fi) {
 292                val ^= fib_devindex_hashfn(nh->nh_oif);
 293        } endfor_nexthops(fi)
 294
 295        return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 296}
 297
 298static struct fib_info *fib_find_info(const struct fib_info *nfi)
 299{
 300        struct hlist_head *head;
 301        struct hlist_node *node;
 302        struct fib_info *fi;
 303        unsigned int hash;
 304
 305        hash = fib_info_hashfn(nfi);
 306        head = &fib_info_hash[hash];
 307
 308        hlist_for_each_entry(fi, node, head, fib_hash) {
 309                if (!net_eq(fi->fib_net, nfi->fib_net))
 310                        continue;
 311                if (fi->fib_nhs != nfi->fib_nhs)
 312                        continue;
 313                if (nfi->fib_protocol == fi->fib_protocol &&
 314                    nfi->fib_scope == fi->fib_scope &&
 315                    nfi->fib_prefsrc == fi->fib_prefsrc &&
 316                    nfi->fib_priority == fi->fib_priority &&
 317                    memcmp(nfi->fib_metrics, fi->fib_metrics,
 318                           sizeof(u32) * RTAX_MAX) == 0 &&
 319                    ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
 320                    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 321                        return fi;
 322        }
 323
 324        return NULL;
 325}
 326
 327/* Check, that the gateway is already configured.
 328 * Used only by redirect accept routine.
 329 */
 330int ip_fib_check_default(__be32 gw, struct net_device *dev)
 331{
 332        struct hlist_head *head;
 333        struct hlist_node *node;
 334        struct fib_nh *nh;
 335        unsigned int hash;
 336
 337        spin_lock(&fib_info_lock);
 338
 339        hash = fib_devindex_hashfn(dev->ifindex);
 340        head = &fib_info_devhash[hash];
 341        hlist_for_each_entry(nh, node, head, nh_hash) {
 342                if (nh->nh_dev == dev &&
 343                    nh->nh_gw == gw &&
 344                    !(nh->nh_flags & RTNH_F_DEAD)) {
 345                        spin_unlock(&fib_info_lock);
 346                        return 0;
 347                }
 348        }
 349
 350        spin_unlock(&fib_info_lock);
 351
 352        return -1;
 353}
 354
 355static inline size_t fib_nlmsg_size(struct fib_info *fi)
 356{
 357        size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 358                         + nla_total_size(4) /* RTA_TABLE */
 359                         + nla_total_size(4) /* RTA_DST */
 360                         + nla_total_size(4) /* RTA_PRIORITY */
 361                         + nla_total_size(4); /* RTA_PREFSRC */
 362
 363        /* space for nested metrics */
 364        payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 365
 366        if (fi->fib_nhs) {
 367                /* Also handles the special case fib_nhs == 1 */
 368
 369                /* each nexthop is packed in an attribute */
 370                size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 371
 372                /* may contain flow and gateway attribute */
 373                nhsize += 2 * nla_total_size(4);
 374
 375                /* all nexthops are packed in a nested attribute */
 376                payload += nla_total_size(fi->fib_nhs * nhsize);
 377        }
 378
 379        return payload;
 380}
 381
 382void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 383               int dst_len, u32 tb_id, struct nl_info *info,
 384               unsigned int nlm_flags)
 385{
 386        struct sk_buff *skb;
 387        u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 388        int err = -ENOBUFS;
 389
 390        skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 391        if (skb == NULL)
 392                goto errout;
 393
 394        err = fib_dump_info(skb, info->pid, seq, event, tb_id,
 395                            fa->fa_type, key, dst_len,
 396                            fa->fa_tos, fa->fa_info, nlm_flags);
 397        if (err < 0) {
 398                /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 399                WARN_ON(err == -EMSGSIZE);
 400                kfree_skb(skb);
 401                goto errout;
 402        }
 403        rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
 404                    info->nlh, GFP_KERNEL);
 405        return;
 406errout:
 407        if (err < 0)
 408                rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 409}
 410
 411/* Return the first fib alias matching TOS with
 412 * priority less than or equal to PRIO.
 413 */
 414struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
 415{
 416        if (fah) {
 417                struct fib_alias *fa;
 418                list_for_each_entry(fa, fah, fa_list) {
 419                        if (fa->fa_tos > tos)
 420                                continue;
 421                        if (fa->fa_info->fib_priority >= prio ||
 422                            fa->fa_tos < tos)
 423                                return fa;
 424                }
 425        }
 426        return NULL;
 427}
 428
 429int fib_detect_death(struct fib_info *fi, int order,
 430                     struct fib_info **last_resort, int *last_idx, int dflt)
 431{
 432        struct neighbour *n;
 433        int state = NUD_NONE;
 434
 435        n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 436        if (n) {
 437                state = n->nud_state;
 438                neigh_release(n);
 439        }
 440        if (state == NUD_REACHABLE)
 441                return 0;
 442        if ((state & NUD_VALID) && order != dflt)
 443                return 0;
 444        if ((state & NUD_VALID) ||
 445            (*last_idx < 0 && order > dflt)) {
 446                *last_resort = fi;
 447                *last_idx = order;
 448        }
 449        return 1;
 450}
 451
 452#ifdef CONFIG_IP_ROUTE_MULTIPATH
 453
 454static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 455{
 456        int nhs = 0;
 457
 458        while (rtnh_ok(rtnh, remaining)) {
 459                nhs++;
 460                rtnh = rtnh_next(rtnh, &remaining);
 461        }
 462
 463        /* leftover implies invalid nexthop configuration, discard it */
 464        return remaining > 0 ? 0 : nhs;
 465}
 466
 467static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 468                       int remaining, struct fib_config *cfg)
 469{
 470        change_nexthops(fi) {
 471                int attrlen;
 472
 473                if (!rtnh_ok(rtnh, remaining))
 474                        return -EINVAL;
 475
 476                nexthop_nh->nh_flags =
 477                        (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 478                nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 479                nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 480
 481                attrlen = rtnh_attrlen(rtnh);
 482                if (attrlen > 0) {
 483                        struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 484
 485                        nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 486                        nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 487#ifdef CONFIG_IP_ROUTE_CLASSID
 488                        nla = nla_find(attrs, attrlen, RTA_FLOW);
 489                        nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 490                        if (nexthop_nh->nh_tclassid)
 491                                fi->fib_net->ipv4.fib_num_tclassid_users++;
 492#endif
 493                }
 494
 495                rtnh = rtnh_next(rtnh, &remaining);
 496        } endfor_nexthops(fi);
 497
 498        return 0;
 499}
 500
 501#endif
 502
 503int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 504{
 505#ifdef CONFIG_IP_ROUTE_MULTIPATH
 506        struct rtnexthop *rtnh;
 507        int remaining;
 508#endif
 509
 510        if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 511                return 1;
 512
 513        if (cfg->fc_oif || cfg->fc_gw) {
 514                if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 515                    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 516                        return 0;
 517                return 1;
 518        }
 519
 520#ifdef CONFIG_IP_ROUTE_MULTIPATH
 521        if (cfg->fc_mp == NULL)
 522                return 0;
 523
 524        rtnh = cfg->fc_mp;
 525        remaining = cfg->fc_mp_len;
 526
 527        for_nexthops(fi) {
 528                int attrlen;
 529
 530                if (!rtnh_ok(rtnh, remaining))
 531                        return -EINVAL;
 532
 533                if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 534                        return 1;
 535
 536                attrlen = rtnh_attrlen(rtnh);
 537                if (attrlen < 0) {
 538                        struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 539
 540                        nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 541                        if (nla && nla_get_be32(nla) != nh->nh_gw)
 542                                return 1;
 543#ifdef CONFIG_IP_ROUTE_CLASSID
 544                        nla = nla_find(attrs, attrlen, RTA_FLOW);
 545                        if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 546                                return 1;
 547#endif
 548                }
 549
 550                rtnh = rtnh_next(rtnh, &remaining);
 551        } endfor_nexthops(fi);
 552#endif
 553        return 0;
 554}
 555
 556
 557/*
 558 * Picture
 559 * -------
 560 *
 561 * Semantics of nexthop is very messy by historical reasons.
 562 * We have to take into account, that:
 563 * a) gateway can be actually local interface address,
 564 *    so that gatewayed route is direct.
 565 * b) gateway must be on-link address, possibly
 566 *    described not by an ifaddr, but also by a direct route.
 567 * c) If both gateway and interface are specified, they should not
 568 *    contradict.
 569 * d) If we use tunnel routes, gateway could be not on-link.
 570 *
 571 * Attempt to reconcile all of these (alas, self-contradictory) conditions
 572 * results in pretty ugly and hairy code with obscure logic.
 573 *
 574 * I chose to generalized it instead, so that the size
 575 * of code does not increase practically, but it becomes
 576 * much more general.
 577 * Every prefix is assigned a "scope" value: "host" is local address,
 578 * "link" is direct route,
 579 * [ ... "site" ... "interior" ... ]
 580 * and "universe" is true gateway route with global meaning.
 581 *
 582 * Every prefix refers to a set of "nexthop"s (gw, oif),
 583 * where gw must have narrower scope. This recursion stops
 584 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 585 * which means that gw is forced to be on link.
 586 *
 587 * Code is still hairy, but now it is apparently logically
 588 * consistent and very flexible. F.e. as by-product it allows
 589 * to co-exists in peace independent exterior and interior
 590 * routing processes.
 591 *
 592 * Normally it looks as following.
 593 *
 594 * {universe prefix}  -> (gw, oif) [scope link]
 595 *                |
 596 *                |-> {link prefix} -> (gw, oif) [scope local]
 597 *                                      |
 598 *                                      |-> {local prefix} (terminal node)
 599 */
 600static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 601                        struct fib_nh *nh)
 602{
 603        int err;
 604        struct net *net;
 605        struct net_device *dev;
 606
 607        net = cfg->fc_nlinfo.nl_net;
 608        if (nh->nh_gw) {
 609                struct fib_result res;
 610
 611                if (nh->nh_flags & RTNH_F_ONLINK) {
 612
 613                        if (cfg->fc_scope >= RT_SCOPE_LINK)
 614                                return -EINVAL;
 615                        if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 616                                return -EINVAL;
 617                        dev = __dev_get_by_index(net, nh->nh_oif);
 618                        if (!dev)
 619                                return -ENODEV;
 620                        if (!(dev->flags & IFF_UP))
 621                                return -ENETDOWN;
 622                        nh->nh_dev = dev;
 623                        dev_hold(dev);
 624                        nh->nh_scope = RT_SCOPE_LINK;
 625                        return 0;
 626                }
 627                rcu_read_lock();
 628                {
 629                        struct flowi4 fl4 = {
 630                                .daddr = nh->nh_gw,
 631                                .flowi4_scope = cfg->fc_scope + 1,
 632                                .flowi4_oif = nh->nh_oif,
 633                        };
 634
 635                        /* It is not necessary, but requires a bit of thinking */
 636                        if (fl4.flowi4_scope < RT_SCOPE_LINK)
 637                                fl4.flowi4_scope = RT_SCOPE_LINK;
 638                        err = fib_lookup(net, &fl4, &res);
 639                        if (err) {
 640                                rcu_read_unlock();
 641                                return err;
 642                        }
 643                }
 644                err = -EINVAL;
 645                if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 646                        goto out;
 647                nh->nh_scope = res.scope;
 648                nh->nh_oif = FIB_RES_OIF(res);
 649                nh->nh_dev = dev = FIB_RES_DEV(res);
 650                if (!dev)
 651                        goto out;
 652                dev_hold(dev);
 653                err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 654        } else {
 655                struct in_device *in_dev;
 656
 657                if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 658                        return -EINVAL;
 659
 660                rcu_read_lock();
 661                err = -ENODEV;
 662                in_dev = inetdev_by_index(net, nh->nh_oif);
 663                if (in_dev == NULL)
 664                        goto out;
 665                err = -ENETDOWN;
 666                if (!(in_dev->dev->flags & IFF_UP))
 667                        goto out;
 668                nh->nh_dev = in_dev->dev;
 669                dev_hold(nh->nh_dev);
 670                nh->nh_scope = RT_SCOPE_HOST;
 671                err = 0;
 672        }
 673out:
 674        rcu_read_unlock();
 675        return err;
 676}
 677
 678static inline unsigned int fib_laddr_hashfn(__be32 val)
 679{
 680        unsigned int mask = (fib_info_hash_size - 1);
 681
 682        return ((__force u32)val ^
 683                ((__force u32)val >> 7) ^
 684                ((__force u32)val >> 14)) & mask;
 685}
 686
 687static struct hlist_head *fib_info_hash_alloc(int bytes)
 688{
 689        if (bytes <= PAGE_SIZE)
 690                return kzalloc(bytes, GFP_KERNEL);
 691        else
 692                return (struct hlist_head *)
 693                        __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 694                                         get_order(bytes));
 695}
 696
 697static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 698{
 699        if (!hash)
 700                return;
 701
 702        if (bytes <= PAGE_SIZE)
 703                kfree(hash);
 704        else
 705                free_pages((unsigned long) hash, get_order(bytes));
 706}
 707
 708static void fib_info_hash_move(struct hlist_head *new_info_hash,
 709                               struct hlist_head *new_laddrhash,
 710                               unsigned int new_size)
 711{
 712        struct hlist_head *old_info_hash, *old_laddrhash;
 713        unsigned int old_size = fib_info_hash_size;
 714        unsigned int i, bytes;
 715
 716        spin_lock_bh(&fib_info_lock);
 717        old_info_hash = fib_info_hash;
 718        old_laddrhash = fib_info_laddrhash;
 719        fib_info_hash_size = new_size;
 720
 721        for (i = 0; i < old_size; i++) {
 722                struct hlist_head *head = &fib_info_hash[i];
 723                struct hlist_node *node, *n;
 724                struct fib_info *fi;
 725
 726                hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
 727                        struct hlist_head *dest;
 728                        unsigned int new_hash;
 729
 730                        hlist_del(&fi->fib_hash);
 731
 732                        new_hash = fib_info_hashfn(fi);
 733                        dest = &new_info_hash[new_hash];
 734                        hlist_add_head(&fi->fib_hash, dest);
 735                }
 736        }
 737        fib_info_hash = new_info_hash;
 738
 739        for (i = 0; i < old_size; i++) {
 740                struct hlist_head *lhead = &fib_info_laddrhash[i];
 741                struct hlist_node *node, *n;
 742                struct fib_info *fi;
 743
 744                hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
 745                        struct hlist_head *ldest;
 746                        unsigned int new_hash;
 747
 748                        hlist_del(&fi->fib_lhash);
 749
 750                        new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 751                        ldest = &new_laddrhash[new_hash];
 752                        hlist_add_head(&fi->fib_lhash, ldest);
 753                }
 754        }
 755        fib_info_laddrhash = new_laddrhash;
 756
 757        spin_unlock_bh(&fib_info_lock);
 758
 759        bytes = old_size * sizeof(struct hlist_head *);
 760        fib_info_hash_free(old_info_hash, bytes);
 761        fib_info_hash_free(old_laddrhash, bytes);
 762}
 763
 764__be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 765{
 766        nh->nh_saddr = inet_select_addr(nh->nh_dev,
 767                                        nh->nh_gw,
 768                                        nh->nh_parent->fib_scope);
 769        nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 770
 771        return nh->nh_saddr;
 772}
 773
 774struct fib_info *fib_create_info(struct fib_config *cfg)
 775{
 776        int err;
 777        struct fib_info *fi = NULL;
 778        struct fib_info *ofi;
 779        int nhs = 1;
 780        struct net *net = cfg->fc_nlinfo.nl_net;
 781
 782        if (cfg->fc_type > RTN_MAX)
 783                goto err_inval;
 784
 785        /* Fast check to catch the most weird cases */
 786        if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 787                goto err_inval;
 788
 789#ifdef CONFIG_IP_ROUTE_MULTIPATH
 790        if (cfg->fc_mp) {
 791                nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 792                if (nhs == 0)
 793                        goto err_inval;
 794        }
 795#endif
 796
 797        err = -ENOBUFS;
 798        if (fib_info_cnt >= fib_info_hash_size) {
 799                unsigned int new_size = fib_info_hash_size << 1;
 800                struct hlist_head *new_info_hash;
 801                struct hlist_head *new_laddrhash;
 802                unsigned int bytes;
 803
 804                if (!new_size)
 805                        new_size = 1;
 806                bytes = new_size * sizeof(struct hlist_head *);
 807                new_info_hash = fib_info_hash_alloc(bytes);
 808                new_laddrhash = fib_info_hash_alloc(bytes);
 809                if (!new_info_hash || !new_laddrhash) {
 810                        fib_info_hash_free(new_info_hash, bytes);
 811                        fib_info_hash_free(new_laddrhash, bytes);
 812                } else
 813                        fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 814
 815                if (!fib_info_hash_size)
 816                        goto failure;
 817        }
 818
 819        fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 820        if (fi == NULL)
 821                goto failure;
 822        if (cfg->fc_mx) {
 823                fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 824                if (!fi->fib_metrics)
 825                        goto failure;
 826        } else
 827                fi->fib_metrics = (u32 *) dst_default_metrics;
 828        fib_info_cnt++;
 829
 830        fi->fib_net = hold_net(net);
 831        fi->fib_protocol = cfg->fc_protocol;
 832        fi->fib_scope = cfg->fc_scope;
 833        fi->fib_flags = cfg->fc_flags;
 834        fi->fib_priority = cfg->fc_priority;
 835        fi->fib_prefsrc = cfg->fc_prefsrc;
 836
 837        fi->fib_nhs = nhs;
 838        change_nexthops(fi) {
 839                nexthop_nh->nh_parent = fi;
 840                nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
 841        } endfor_nexthops(fi)
 842
 843        if (cfg->fc_mx) {
 844                struct nlattr *nla;
 845                int remaining;
 846
 847                nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 848                        int type = nla_type(nla);
 849
 850                        if (type) {
 851                                u32 val;
 852
 853                                if (type > RTAX_MAX)
 854                                        goto err_inval;
 855                                val = nla_get_u32(nla);
 856                                if (type == RTAX_ADVMSS && val > 65535 - 40)
 857                                        val = 65535 - 40;
 858                                if (type == RTAX_MTU && val > 65535 - 15)
 859                                        val = 65535 - 15;
 860                                fi->fib_metrics[type - 1] = val;
 861                        }
 862                }
 863        }
 864
 865        if (cfg->fc_mp) {
 866#ifdef CONFIG_IP_ROUTE_MULTIPATH
 867                err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 868                if (err != 0)
 869                        goto failure;
 870                if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 871                        goto err_inval;
 872                if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 873                        goto err_inval;
 874#ifdef CONFIG_IP_ROUTE_CLASSID
 875                if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 876                        goto err_inval;
 877#endif
 878#else
 879                goto err_inval;
 880#endif
 881        } else {
 882                struct fib_nh *nh = fi->fib_nh;
 883
 884                nh->nh_oif = cfg->fc_oif;
 885                nh->nh_gw = cfg->fc_gw;
 886                nh->nh_flags = cfg->fc_flags;
 887#ifdef CONFIG_IP_ROUTE_CLASSID
 888                nh->nh_tclassid = cfg->fc_flow;
 889                if (nh->nh_tclassid)
 890                        fi->fib_net->ipv4.fib_num_tclassid_users++;
 891#endif
 892#ifdef CONFIG_IP_ROUTE_MULTIPATH
 893                nh->nh_weight = 1;
 894#endif
 895        }
 896
 897        if (fib_props[cfg->fc_type].error) {
 898                if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 899                        goto err_inval;
 900                goto link_it;
 901        } else {
 902                switch (cfg->fc_type) {
 903                case RTN_UNICAST:
 904                case RTN_LOCAL:
 905                case RTN_BROADCAST:
 906                case RTN_ANYCAST:
 907                case RTN_MULTICAST:
 908                        break;
 909                default:
 910                        goto err_inval;
 911                }
 912        }
 913
 914        if (cfg->fc_scope > RT_SCOPE_HOST)
 915                goto err_inval;
 916
 917        if (cfg->fc_scope == RT_SCOPE_HOST) {
 918                struct fib_nh *nh = fi->fib_nh;
 919
 920                /* Local address is added. */
 921                if (nhs != 1 || nh->nh_gw)
 922                        goto err_inval;
 923                nh->nh_scope = RT_SCOPE_NOWHERE;
 924                nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 925                err = -ENODEV;
 926                if (nh->nh_dev == NULL)
 927                        goto failure;
 928        } else {
 929                change_nexthops(fi) {
 930                        err = fib_check_nh(cfg, fi, nexthop_nh);
 931                        if (err != 0)
 932                                goto failure;
 933                } endfor_nexthops(fi)
 934        }
 935
 936        if (fi->fib_prefsrc) {
 937                if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 938                    fi->fib_prefsrc != cfg->fc_dst)
 939                        if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 940                                goto err_inval;
 941        }
 942
 943        change_nexthops(fi) {
 944                fib_info_update_nh_saddr(net, nexthop_nh);
 945        } endfor_nexthops(fi)
 946
 947link_it:
 948        ofi = fib_find_info(fi);
 949        if (ofi) {
 950                fi->fib_dead = 1;
 951                free_fib_info(fi);
 952                ofi->fib_treeref++;
 953                return ofi;
 954        }
 955
 956        fi->fib_treeref++;
 957        atomic_inc(&fi->fib_clntref);
 958        spin_lock_bh(&fib_info_lock);
 959        hlist_add_head(&fi->fib_hash,
 960                       &fib_info_hash[fib_info_hashfn(fi)]);
 961        if (fi->fib_prefsrc) {
 962                struct hlist_head *head;
 963
 964                head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 965                hlist_add_head(&fi->fib_lhash, head);
 966        }
 967        change_nexthops(fi) {
 968                struct hlist_head *head;
 969                unsigned int hash;
 970
 971                if (!nexthop_nh->nh_dev)
 972                        continue;
 973                hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 974                head = &fib_info_devhash[hash];
 975                hlist_add_head(&nexthop_nh->nh_hash, head);
 976        } endfor_nexthops(fi)
 977        spin_unlock_bh(&fib_info_lock);
 978        return fi;
 979
 980err_inval:
 981        err = -EINVAL;
 982
 983failure:
 984        if (fi) {
 985                fi->fib_dead = 1;
 986                free_fib_info(fi);
 987        }
 988
 989        return ERR_PTR(err);
 990}
 991
 992int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 993                  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
 994                  struct fib_info *fi, unsigned int flags)
 995{
 996        struct nlmsghdr *nlh;
 997        struct rtmsg *rtm;
 998
 999        nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
1000        if (nlh == NULL)
1001                return -EMSGSIZE;
1002
1003        rtm = nlmsg_data(nlh);
1004        rtm->rtm_family = AF_INET;
1005        rtm->rtm_dst_len = dst_len;
1006        rtm->rtm_src_len = 0;
1007        rtm->rtm_tos = tos;
1008        if (tb_id < 256)
1009                rtm->rtm_table = tb_id;
1010        else
1011                rtm->rtm_table = RT_TABLE_COMPAT;
1012        if (nla_put_u32(skb, RTA_TABLE, tb_id))
1013                goto nla_put_failure;
1014        rtm->rtm_type = type;
1015        rtm->rtm_flags = fi->fib_flags;
1016        rtm->rtm_scope = fi->fib_scope;
1017        rtm->rtm_protocol = fi->fib_protocol;
1018
1019        if (rtm->rtm_dst_len &&
1020            nla_put_be32(skb, RTA_DST, dst))
1021                goto nla_put_failure;
1022        if (fi->fib_priority &&
1023            nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1024                goto nla_put_failure;
1025        if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1026                goto nla_put_failure;
1027
1028        if (fi->fib_prefsrc &&
1029            nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc))
1030                goto nla_put_failure;
1031        if (fi->fib_nhs == 1) {
1032                if (fi->fib_nh->nh_gw &&
1033                    nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1034                        goto nla_put_failure;
1035                if (fi->fib_nh->nh_oif &&
1036                    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1037                        goto nla_put_failure;
1038#ifdef CONFIG_IP_ROUTE_CLASSID
1039                if (fi->fib_nh[0].nh_tclassid &&
1040                    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1041                        goto nla_put_failure;
1042#endif
1043        }
1044#ifdef CONFIG_IP_ROUTE_MULTIPATH
1045        if (fi->fib_nhs > 1) {
1046                struct rtnexthop *rtnh;
1047                struct nlattr *mp;
1048
1049                mp = nla_nest_start(skb, RTA_MULTIPATH);
1050                if (mp == NULL)
1051                        goto nla_put_failure;
1052
1053                for_nexthops(fi) {
1054                        rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1055                        if (rtnh == NULL)
1056                                goto nla_put_failure;
1057
1058                        rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1059                        rtnh->rtnh_hops = nh->nh_weight - 1;
1060                        rtnh->rtnh_ifindex = nh->nh_oif;
1061
1062                        if (nh->nh_gw &&
1063                            nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw))
1064                                goto nla_put_failure;
1065#ifdef CONFIG_IP_ROUTE_CLASSID
1066                        if (nh->nh_tclassid &&
1067                            nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1068                                goto nla_put_failure;
1069#endif
1070                        /* length of rtnetlink header + attributes */
1071                        rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1072                } endfor_nexthops(fi);
1073
1074                nla_nest_end(skb, mp);
1075        }
1076#endif
1077        return nlmsg_end(skb, nlh);
1078
1079nla_put_failure:
1080        nlmsg_cancel(skb, nlh);
1081        return -EMSGSIZE;
1082}
1083
1084/*
1085 * Update FIB if:
1086 * - local address disappeared -> we must delete all the entries
1087 *   referring to it.
1088 * - device went down -> we must shutdown all nexthops going via it.
1089 */
1090int fib_sync_down_addr(struct net *net, __be32 local)
1091{
1092        int ret = 0;
1093        unsigned int hash = fib_laddr_hashfn(local);
1094        struct hlist_head *head = &fib_info_laddrhash[hash];
1095        struct hlist_node *node;
1096        struct fib_info *fi;
1097
1098        if (fib_info_laddrhash == NULL || local == 0)
1099                return 0;
1100
1101        hlist_for_each_entry(fi, node, head, fib_lhash) {
1102                if (!net_eq(fi->fib_net, net))
1103                        continue;
1104                if (fi->fib_prefsrc == local) {
1105                        fi->fib_flags |= RTNH_F_DEAD;
1106                        ret++;
1107                }
1108        }
1109        return ret;
1110}
1111
1112int fib_sync_down_dev(struct net_device *dev, int force)
1113{
1114        int ret = 0;
1115        int scope = RT_SCOPE_NOWHERE;
1116        struct fib_info *prev_fi = NULL;
1117        unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1118        struct hlist_head *head = &fib_info_devhash[hash];
1119        struct hlist_node *node;
1120        struct fib_nh *nh;
1121
1122        if (force)
1123                scope = -1;
1124
1125        hlist_for_each_entry(nh, node, head, nh_hash) {
1126                struct fib_info *fi = nh->nh_parent;
1127                int dead;
1128
1129                BUG_ON(!fi->fib_nhs);
1130                if (nh->nh_dev != dev || fi == prev_fi)
1131                        continue;
1132                prev_fi = fi;
1133                dead = 0;
1134                change_nexthops(fi) {
1135                        if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1136                                dead++;
1137                        else if (nexthop_nh->nh_dev == dev &&
1138                                 nexthop_nh->nh_scope != scope) {
1139                                nexthop_nh->nh_flags |= RTNH_F_DEAD;
1140#ifdef CONFIG_IP_ROUTE_MULTIPATH
1141                                spin_lock_bh(&fib_multipath_lock);
1142                                fi->fib_power -= nexthop_nh->nh_power;
1143                                nexthop_nh->nh_power = 0;
1144                                spin_unlock_bh(&fib_multipath_lock);
1145#endif
1146                                dead++;
1147                        }
1148#ifdef CONFIG_IP_ROUTE_MULTIPATH
1149                        if (force > 1 && nexthop_nh->nh_dev == dev) {
1150                                dead = fi->fib_nhs;
1151                                break;
1152                        }
1153#endif
1154                } endfor_nexthops(fi)
1155                if (dead == fi->fib_nhs) {
1156                        fi->fib_flags |= RTNH_F_DEAD;
1157                        ret++;
1158                }
1159        }
1160
1161        return ret;
1162}
1163
1164/* Must be invoked inside of an RCU protected region.  */
1165void fib_select_default(struct fib_result *res)
1166{
1167        struct fib_info *fi = NULL, *last_resort = NULL;
1168        struct list_head *fa_head = res->fa_head;
1169        struct fib_table *tb = res->table;
1170        int order = -1, last_idx = -1;
1171        struct fib_alias *fa;
1172
1173        list_for_each_entry_rcu(fa, fa_head, fa_list) {
1174                struct fib_info *next_fi = fa->fa_info;
1175
1176                if (next_fi->fib_scope != res->scope ||
1177                    fa->fa_type != RTN_UNICAST)
1178                        continue;
1179
1180                if (next_fi->fib_priority > res->fi->fib_priority)
1181                        break;
1182                if (!next_fi->fib_nh[0].nh_gw ||
1183                    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1184                        continue;
1185
1186                fib_alias_accessed(fa);
1187
1188                if (fi == NULL) {
1189                        if (next_fi != res->fi)
1190                                break;
1191                } else if (!fib_detect_death(fi, order, &last_resort,
1192                                             &last_idx, tb->tb_default)) {
1193                        fib_result_assign(res, fi);
1194                        tb->tb_default = order;
1195                        goto out;
1196                }
1197                fi = next_fi;
1198                order++;
1199        }
1200
1201        if (order <= 0 || fi == NULL) {
1202                tb->tb_default = -1;
1203                goto out;
1204        }
1205
1206        if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1207                                tb->tb_default)) {
1208                fib_result_assign(res, fi);
1209                tb->tb_default = order;
1210                goto out;
1211        }
1212
1213        if (last_idx >= 0)
1214                fib_result_assign(res, last_resort);
1215        tb->tb_default = last_idx;
1216out:
1217        return;
1218}
1219
1220#ifdef CONFIG_IP_ROUTE_MULTIPATH
1221
1222/*
1223 * Dead device goes up. We wake up dead nexthops.
1224 * It takes sense only on multipath routes.
1225 */
1226int fib_sync_up(struct net_device *dev)
1227{
1228        struct fib_info *prev_fi;
1229        unsigned int hash;
1230        struct hlist_head *head;
1231        struct hlist_node *node;
1232        struct fib_nh *nh;
1233        int ret;
1234
1235        if (!(dev->flags & IFF_UP))
1236                return 0;
1237
1238        prev_fi = NULL;
1239        hash = fib_devindex_hashfn(dev->ifindex);
1240        head = &fib_info_devhash[hash];
1241        ret = 0;
1242
1243        hlist_for_each_entry(nh, node, head, nh_hash) {
1244                struct fib_info *fi = nh->nh_parent;
1245                int alive;
1246
1247                BUG_ON(!fi->fib_nhs);
1248                if (nh->nh_dev != dev || fi == prev_fi)
1249                        continue;
1250
1251                prev_fi = fi;
1252                alive = 0;
1253                change_nexthops(fi) {
1254                        if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1255                                alive++;
1256                                continue;
1257                        }
1258                        if (nexthop_nh->nh_dev == NULL ||
1259                            !(nexthop_nh->nh_dev->flags & IFF_UP))
1260                                continue;
1261                        if (nexthop_nh->nh_dev != dev ||
1262                            !__in_dev_get_rtnl(dev))
1263                                continue;
1264                        alive++;
1265                        spin_lock_bh(&fib_multipath_lock);
1266                        nexthop_nh->nh_power = 0;
1267                        nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1268                        spin_unlock_bh(&fib_multipath_lock);
1269                } endfor_nexthops(fi)
1270
1271                if (alive > 0) {
1272                        fi->fib_flags &= ~RTNH_F_DEAD;
1273                        ret++;
1274                }
1275        }
1276
1277        return ret;
1278}
1279
1280/*
1281 * The algorithm is suboptimal, but it provides really
1282 * fair weighted route distribution.
1283 */
1284void fib_select_multipath(struct fib_result *res)
1285{
1286        struct fib_info *fi = res->fi;
1287        int w;
1288
1289        spin_lock_bh(&fib_multipath_lock);
1290        if (fi->fib_power <= 0) {
1291                int power = 0;
1292                change_nexthops(fi) {
1293                        if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1294                                power += nexthop_nh->nh_weight;
1295                                nexthop_nh->nh_power = nexthop_nh->nh_weight;
1296                        }
1297                } endfor_nexthops(fi);
1298                fi->fib_power = power;
1299                if (power <= 0) {
1300                        spin_unlock_bh(&fib_multipath_lock);
1301                        /* Race condition: route has just become dead. */
1302                        res->nh_sel = 0;
1303                        return;
1304                }
1305        }
1306
1307
1308        /* w should be random number [0..fi->fib_power-1],
1309         * it is pretty bad approximation.
1310         */
1311
1312        w = jiffies % fi->fib_power;
1313
1314        change_nexthops(fi) {
1315                if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1316                    nexthop_nh->nh_power) {
1317                        w -= nexthop_nh->nh_power;
1318                        if (w <= 0) {
1319                                nexthop_nh->nh_power--;
1320                                fi->fib_power--;
1321                                res->nh_sel = nhsel;
1322                                spin_unlock_bh(&fib_multipath_lock);
1323                                return;
1324                        }
1325                }
1326        } endfor_nexthops(fi);
1327
1328        /* Race condition: route has just become dead. */
1329        res->nh_sel = 0;
1330        spin_unlock_bh(&fib_multipath_lock);
1331}
1332#endif
1333
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.