linux/net/netfilter/xt_hashlimit.c
<<
>>
Prefs
   1/*
   2 *      xt_hashlimit - Netfilter module to limit the number of packets per time
   3 *      separately for each hashbucket (sourceip/sourceport/dstip/dstport)
   4 *
   5 *      (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
   6 *      (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   7 *      Copyright © CC Computer Consultants GmbH, 2007 - 2008
   8 *
   9 * Development of this code was funded by Astaro AG, http://www.astaro.com/
  10 */
  11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12#include <linux/module.h>
  13#include <linux/spinlock.h>
  14#include <linux/random.h>
  15#include <linux/jhash.h>
  16#include <linux/slab.h>
  17#include <linux/vmalloc.h>
  18#include <linux/proc_fs.h>
  19#include <linux/seq_file.h>
  20#include <linux/list.h>
  21#include <linux/skbuff.h>
  22#include <linux/mm.h>
  23#include <linux/in.h>
  24#include <linux/ip.h>
  25#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
  26#include <linux/ipv6.h>
  27#include <net/ipv6.h>
  28#endif
  29
  30#include <net/net_namespace.h>
  31#include <net/netns/generic.h>
  32
  33#include <linux/netfilter/x_tables.h>
  34#include <linux/netfilter_ipv4/ip_tables.h>
  35#include <linux/netfilter_ipv6/ip6_tables.h>
  36#include <linux/netfilter/xt_hashlimit.h>
  37#include <linux/mutex.h>
  38
  39MODULE_LICENSE("GPL");
  40MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
  41MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
  42MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
  43MODULE_ALIAS("ipt_hashlimit");
  44MODULE_ALIAS("ip6t_hashlimit");
  45
  46struct hashlimit_net {
  47        struct hlist_head       htables;
  48        struct proc_dir_entry   *ipt_hashlimit;
  49        struct proc_dir_entry   *ip6t_hashlimit;
  50};
  51
  52static int hashlimit_net_id;
  53static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
  54{
  55        return net_generic(net, hashlimit_net_id);
  56}
  57
  58/* need to declare this at the top */
  59static const struct file_operations dl_file_ops;
  60
  61/* hash table crap */
  62struct dsthash_dst {
  63        union {
  64                struct {
  65                        __be32 src;
  66                        __be32 dst;
  67                } ip;
  68#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
  69                struct {
  70                        __be32 src[4];
  71                        __be32 dst[4];
  72                } ip6;
  73#endif
  74        };
  75        __be16 src_port;
  76        __be16 dst_port;
  77};
  78
  79struct dsthash_ent {
  80        /* static / read-only parts in the beginning */
  81        struct hlist_node node;
  82        struct dsthash_dst dst;
  83
  84        /* modified structure members in the end */
  85        spinlock_t lock;
  86        unsigned long expires;          /* precalculated expiry time */
  87        struct {
  88                unsigned long prev;     /* last modification */
  89                u_int32_t credit;
  90                u_int32_t credit_cap, cost;
  91        } rateinfo;
  92        struct rcu_head rcu;
  93};
  94
  95struct xt_hashlimit_htable {
  96        struct hlist_node node;         /* global list of all htables */
  97        int use;
  98        u_int8_t family;
  99        bool rnd_initialized;
 100
 101        struct hashlimit_cfg1 cfg;      /* config */
 102
 103        /* used internally */
 104        spinlock_t lock;                /* lock for list_head */
 105        u_int32_t rnd;                  /* random seed for hash */
 106        unsigned int count;             /* number entries in table */
 107        struct timer_list timer;        /* timer for gc */
 108
 109        /* seq_file stuff */
 110        struct proc_dir_entry *pde;
 111        const char *name;
 112        struct net *net;
 113
 114        struct hlist_head hash[0];      /* hashtable itself */
 115};
 116
 117static DEFINE_MUTEX(hashlimit_mutex);   /* protects htables list */
 118static struct kmem_cache *hashlimit_cachep __read_mostly;
 119
 120static inline bool dst_cmp(const struct dsthash_ent *ent,
 121                           const struct dsthash_dst *b)
 122{
 123        return !memcmp(&ent->dst, b, sizeof(ent->dst));
 124}
 125
 126static u_int32_t
 127hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
 128{
 129        u_int32_t hash = jhash2((const u32 *)dst,
 130                                sizeof(*dst)/sizeof(u32),
 131                                ht->rnd);
 132        /*
 133         * Instead of returning hash % ht->cfg.size (implying a divide)
 134         * we return the high 32 bits of the (hash * ht->cfg.size) that will
 135         * give results between [0 and cfg.size-1] and same hash distribution,
 136         * but using a multiply, less expensive than a divide
 137         */
 138        return reciprocal_scale(hash, ht->cfg.size);
 139}
 140
 141static struct dsthash_ent *
 142dsthash_find(const struct xt_hashlimit_htable *ht,
 143             const struct dsthash_dst *dst)
 144{
 145        struct dsthash_ent *ent;
 146        u_int32_t hash = hash_dst(ht, dst);
 147
 148        if (!hlist_empty(&ht->hash[hash])) {
 149                hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
 150                        if (dst_cmp(ent, dst)) {
 151                                spin_lock(&ent->lock);
 152                                return ent;
 153                        }
 154        }
 155        return NULL;
 156}
 157
 158/* allocate dsthash_ent, initialize dst, put in htable and lock it */
 159static struct dsthash_ent *
 160dsthash_alloc_init(struct xt_hashlimit_htable *ht,
 161                   const struct dsthash_dst *dst, bool *race)
 162{
 163        struct dsthash_ent *ent;
 164
 165        spin_lock(&ht->lock);
 166
 167        /* Two or more packets may race to create the same entry in the
 168         * hashtable, double check if this packet lost race.
 169         */
 170        ent = dsthash_find(ht, dst);
 171        if (ent != NULL) {
 172                spin_unlock(&ht->lock);
 173                *race = true;
 174                return ent;
 175        }
 176
 177        /* initialize hash with random val at the time we allocate
 178         * the first hashtable entry */
 179        if (unlikely(!ht->rnd_initialized)) {
 180                get_random_bytes(&ht->rnd, sizeof(ht->rnd));
 181                ht->rnd_initialized = true;
 182        }
 183
 184        if (ht->cfg.max && ht->count >= ht->cfg.max) {
 185                /* FIXME: do something. question is what.. */
 186                net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
 187                ent = NULL;
 188        } else
 189                ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
 190        if (ent) {
 191                memcpy(&ent->dst, dst, sizeof(ent->dst));
 192                spin_lock_init(&ent->lock);
 193
 194                spin_lock(&ent->lock);
 195                hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
 196                ht->count++;
 197        }
 198        spin_unlock(&ht->lock);
 199        return ent;
 200}
 201
 202static void dsthash_free_rcu(struct rcu_head *head)
 203{
 204        struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
 205
 206        kmem_cache_free(hashlimit_cachep, ent);
 207}
 208
 209static inline void
 210dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
 211{
 212        hlist_del_rcu(&ent->node);
 213        call_rcu_bh(&ent->rcu, dsthash_free_rcu);
 214        ht->count--;
 215}
 216static void htable_gc(unsigned long htlong);
 217
 218static int htable_create(struct net *net, struct xt_hashlimit_mtinfo1 *minfo,
 219                         u_int8_t family)
 220{
 221        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 222        struct xt_hashlimit_htable *hinfo;
 223        unsigned int size;
 224        unsigned int i;
 225
 226        if (minfo->cfg.size) {
 227                size = minfo->cfg.size;
 228        } else {
 229                size = (totalram_pages << PAGE_SHIFT) / 16384 /
 230                       sizeof(struct list_head);
 231                if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
 232                        size = 8192;
 233                if (size < 16)
 234                        size = 16;
 235        }
 236        /* FIXME: don't use vmalloc() here or anywhere else -HW */
 237        hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
 238                        sizeof(struct list_head) * size);
 239        if (hinfo == NULL)
 240                return -ENOMEM;
 241        minfo->hinfo = hinfo;
 242
 243        /* copy match config into hashtable config */
 244        memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
 245        hinfo->cfg.size = size;
 246        if (hinfo->cfg.max == 0)
 247                hinfo->cfg.max = 8 * hinfo->cfg.size;
 248        else if (hinfo->cfg.max < hinfo->cfg.size)
 249                hinfo->cfg.max = hinfo->cfg.size;
 250
 251        for (i = 0; i < hinfo->cfg.size; i++)
 252                INIT_HLIST_HEAD(&hinfo->hash[i]);
 253
 254        hinfo->use = 1;
 255        hinfo->count = 0;
 256        hinfo->family = family;
 257        hinfo->rnd_initialized = false;
 258        hinfo->name = kstrdup(minfo->name, GFP_KERNEL);
 259        if (!hinfo->name) {
 260                vfree(hinfo);
 261                return -ENOMEM;
 262        }
 263        spin_lock_init(&hinfo->lock);
 264
 265        hinfo->pde = proc_create_data(minfo->name, 0,
 266                (family == NFPROTO_IPV4) ?
 267                hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
 268                &dl_file_ops, hinfo);
 269        if (hinfo->pde == NULL) {
 270                kfree(hinfo->name);
 271                vfree(hinfo);
 272                return -ENOMEM;
 273        }
 274        hinfo->net = net;
 275
 276        setup_timer(&hinfo->timer, htable_gc, (unsigned long)hinfo);
 277        hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
 278        add_timer(&hinfo->timer);
 279
 280        hlist_add_head(&hinfo->node, &hashlimit_net->htables);
 281
 282        return 0;
 283}
 284
 285static bool select_all(const struct xt_hashlimit_htable *ht,
 286                       const struct dsthash_ent *he)
 287{
 288        return 1;
 289}
 290
 291static bool select_gc(const struct xt_hashlimit_htable *ht,
 292                      const struct dsthash_ent *he)
 293{
 294        return time_after_eq(jiffies, he->expires);
 295}
 296
 297static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
 298                        bool (*select)(const struct xt_hashlimit_htable *ht,
 299                                      const struct dsthash_ent *he))
 300{
 301        unsigned int i;
 302
 303        /* lock hash table and iterate over it */
 304        spin_lock_bh(&ht->lock);
 305        for (i = 0; i < ht->cfg.size; i++) {
 306                struct dsthash_ent *dh;
 307                struct hlist_node *n;
 308                hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
 309                        if ((*select)(ht, dh))
 310                                dsthash_free(ht, dh);
 311                }
 312        }
 313        spin_unlock_bh(&ht->lock);
 314}
 315
 316/* hash table garbage collector, run by timer */
 317static void htable_gc(unsigned long htlong)
 318{
 319        struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
 320
 321        htable_selective_cleanup(ht, select_gc);
 322
 323        /* re-add the timer accordingly */
 324        ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
 325        add_timer(&ht->timer);
 326}
 327
 328static void htable_destroy(struct xt_hashlimit_htable *hinfo)
 329{
 330        struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
 331        struct proc_dir_entry *parent;
 332
 333        del_timer_sync(&hinfo->timer);
 334
 335        if (hinfo->family == NFPROTO_IPV4)
 336                parent = hashlimit_net->ipt_hashlimit;
 337        else
 338                parent = hashlimit_net->ip6t_hashlimit;
 339
 340        if(parent != NULL)
 341                remove_proc_entry(hinfo->name, parent);
 342
 343        htable_selective_cleanup(hinfo, select_all);
 344        kfree(hinfo->name);
 345        vfree(hinfo);
 346}
 347
 348static struct xt_hashlimit_htable *htable_find_get(struct net *net,
 349                                                   const char *name,
 350                                                   u_int8_t family)
 351{
 352        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 353        struct xt_hashlimit_htable *hinfo;
 354
 355        hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
 356                if (!strcmp(name, hinfo->name) &&
 357                    hinfo->family == family) {
 358                        hinfo->use++;
 359                        return hinfo;
 360                }
 361        }
 362        return NULL;
 363}
 364
 365static void htable_put(struct xt_hashlimit_htable *hinfo)
 366{
 367        mutex_lock(&hashlimit_mutex);
 368        if (--hinfo->use == 0) {
 369                hlist_del(&hinfo->node);
 370                htable_destroy(hinfo);
 371        }
 372        mutex_unlock(&hashlimit_mutex);
 373}
 374
 375/* The algorithm used is the Simple Token Bucket Filter (TBF)
 376 * see net/sched/sch_tbf.c in the linux source tree
 377 */
 378
 379/* Rusty: This is my (non-mathematically-inclined) understanding of
 380   this algorithm.  The `average rate' in jiffies becomes your initial
 381   amount of credit `credit' and the most credit you can ever have
 382   `credit_cap'.  The `peak rate' becomes the cost of passing the
 383   test, `cost'.
 384
 385   `prev' tracks the last packet hit: you gain one credit per jiffy.
 386   If you get credit balance more than this, the extra credit is
 387   discarded.  Every time the match passes, you lose `cost' credits;
 388   if you don't have that many, the test fails.
 389
 390   See Alexey's formal explanation in net/sched/sch_tbf.c.
 391
 392   To get the maximum range, we multiply by this factor (ie. you get N
 393   credits per jiffy).  We want to allow a rate as low as 1 per day
 394   (slowest userspace tool allows), which means
 395   CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
 396*/
 397#define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
 398
 399/* Repeated shift and or gives us all 1s, final shift and add 1 gives
 400 * us the power of 2 below the theoretical max, so GCC simply does a
 401 * shift. */
 402#define _POW2_BELOW2(x) ((x)|((x)>>1))
 403#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
 404#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
 405#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
 406#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
 407#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
 408
 409#define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
 410
 411/* in byte mode, the lowest possible rate is one packet/second.
 412 * credit_cap is used as a counter that tells us how many times we can
 413 * refill the "credits available" counter when it becomes empty.
 414 */
 415#define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
 416#define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
 417
 418static u32 xt_hashlimit_len_to_chunks(u32 len)
 419{
 420        return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
 421}
 422
 423/* Precision saver. */
 424static u32 user2credits(u32 user)
 425{
 426        /* If multiplying would overflow... */
 427        if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
 428                /* Divide first. */
 429                return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
 430
 431        return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
 432}
 433
 434static u32 user2credits_byte(u32 user)
 435{
 436        u64 us = user;
 437        us *= HZ * CREDITS_PER_JIFFY_BYTES;
 438        return (u32) (us >> 32);
 439}
 440
 441static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now, u32 mode)
 442{
 443        unsigned long delta = now - dh->rateinfo.prev;
 444        u32 cap;
 445
 446        if (delta == 0)
 447                return;
 448
 449        dh->rateinfo.prev = now;
 450
 451        if (mode & XT_HASHLIMIT_BYTES) {
 452                u32 tmp = dh->rateinfo.credit;
 453                dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
 454                cap = CREDITS_PER_JIFFY_BYTES * HZ;
 455                if (tmp >= dh->rateinfo.credit) {/* overflow */
 456                        dh->rateinfo.credit = cap;
 457                        return;
 458                }
 459        } else {
 460                dh->rateinfo.credit += delta * CREDITS_PER_JIFFY;
 461                cap = dh->rateinfo.credit_cap;
 462        }
 463        if (dh->rateinfo.credit > cap)
 464                dh->rateinfo.credit = cap;
 465}
 466
 467static void rateinfo_init(struct dsthash_ent *dh,
 468                          struct xt_hashlimit_htable *hinfo)
 469{
 470        dh->rateinfo.prev = jiffies;
 471        if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
 472                dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
 473                dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
 474                dh->rateinfo.credit_cap = hinfo->cfg.burst;
 475        } else {
 476                dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
 477                                                   hinfo->cfg.burst);
 478                dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
 479                dh->rateinfo.credit_cap = dh->rateinfo.credit;
 480        }
 481}
 482
 483static inline __be32 maskl(__be32 a, unsigned int l)
 484{
 485        return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
 486}
 487
 488#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 489static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
 490{
 491        switch (p) {
 492        case 0 ... 31:
 493                i[0] = maskl(i[0], p);
 494                i[1] = i[2] = i[3] = 0;
 495                break;
 496        case 32 ... 63:
 497                i[1] = maskl(i[1], p - 32);
 498                i[2] = i[3] = 0;
 499                break;
 500        case 64 ... 95:
 501                i[2] = maskl(i[2], p - 64);
 502                i[3] = 0;
 503                break;
 504        case 96 ... 127:
 505                i[3] = maskl(i[3], p - 96);
 506                break;
 507        case 128:
 508                break;
 509        }
 510}
 511#endif
 512
 513static int
 514hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
 515                   struct dsthash_dst *dst,
 516                   const struct sk_buff *skb, unsigned int protoff)
 517{
 518        __be16 _ports[2], *ports;
 519        u8 nexthdr;
 520        int poff;
 521
 522        memset(dst, 0, sizeof(*dst));
 523
 524        switch (hinfo->family) {
 525        case NFPROTO_IPV4:
 526                if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
 527                        dst->ip.dst = maskl(ip_hdr(skb)->daddr,
 528                                      hinfo->cfg.dstmask);
 529                if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
 530                        dst->ip.src = maskl(ip_hdr(skb)->saddr,
 531                                      hinfo->cfg.srcmask);
 532
 533                if (!(hinfo->cfg.mode &
 534                      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 535                        return 0;
 536                nexthdr = ip_hdr(skb)->protocol;
 537                break;
 538#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 539        case NFPROTO_IPV6:
 540        {
 541                __be16 frag_off;
 542
 543                if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
 544                        memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
 545                               sizeof(dst->ip6.dst));
 546                        hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
 547                }
 548                if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
 549                        memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
 550                               sizeof(dst->ip6.src));
 551                        hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
 552                }
 553
 554                if (!(hinfo->cfg.mode &
 555                      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 556                        return 0;
 557                nexthdr = ipv6_hdr(skb)->nexthdr;
 558                protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
 559                if ((int)protoff < 0)
 560                        return -1;
 561                break;
 562        }
 563#endif
 564        default:
 565                BUG();
 566                return 0;
 567        }
 568
 569        poff = proto_ports_offset(nexthdr);
 570        if (poff >= 0) {
 571                ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
 572                                           &_ports);
 573        } else {
 574                _ports[0] = _ports[1] = 0;
 575                ports = _ports;
 576        }
 577        if (!ports)
 578                return -1;
 579        if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
 580                dst->src_port = ports[0];
 581        if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
 582                dst->dst_port = ports[1];
 583        return 0;
 584}
 585
 586static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
 587{
 588        u64 tmp = xt_hashlimit_len_to_chunks(len);
 589        tmp = tmp * dh->rateinfo.cost;
 590
 591        if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
 592                tmp = CREDITS_PER_JIFFY_BYTES * HZ;
 593
 594        if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
 595                dh->rateinfo.credit_cap--;
 596                dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
 597        }
 598        return (u32) tmp;
 599}
 600
 601static bool
 602hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
 603{
 604        const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 605        struct xt_hashlimit_htable *hinfo = info->hinfo;
 606        unsigned long now = jiffies;
 607        struct dsthash_ent *dh;
 608        struct dsthash_dst dst;
 609        bool race = false;
 610        u32 cost;
 611
 612        if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
 613                goto hotdrop;
 614
 615        rcu_read_lock_bh();
 616        dh = dsthash_find(hinfo, &dst);
 617        if (dh == NULL) {
 618                dh = dsthash_alloc_init(hinfo, &dst, &race);
 619                if (dh == NULL) {
 620                        rcu_read_unlock_bh();
 621                        goto hotdrop;
 622                } else if (race) {
 623                        /* Already got an entry, update expiration timeout */
 624                        dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 625                        rateinfo_recalc(dh, now, hinfo->cfg.mode);
 626                } else {
 627                        dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
 628                        rateinfo_init(dh, hinfo);
 629                }
 630        } else {
 631                /* update expiration timeout */
 632                dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 633                rateinfo_recalc(dh, now, hinfo->cfg.mode);
 634        }
 635
 636        if (info->cfg.mode & XT_HASHLIMIT_BYTES)
 637                cost = hashlimit_byte_cost(skb->len, dh);
 638        else
 639                cost = dh->rateinfo.cost;
 640
 641        if (dh->rateinfo.credit >= cost) {
 642                /* below the limit */
 643                dh->rateinfo.credit -= cost;
 644                spin_unlock(&dh->lock);
 645                rcu_read_unlock_bh();
 646                return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
 647        }
 648
 649        spin_unlock(&dh->lock);
 650        rcu_read_unlock_bh();
 651        /* default match is underlimit - so over the limit, we need to invert */
 652        return info->cfg.mode & XT_HASHLIMIT_INVERT;
 653
 654 hotdrop:
 655        par->hotdrop = true;
 656        return false;
 657}
 658
 659static int hashlimit_mt_check(const struct xt_mtchk_param *par)
 660{
 661        struct net *net = par->net;
 662        struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 663        int ret;
 664
 665        if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
 666                return -EINVAL;
 667        if (info->name[sizeof(info->name)-1] != '\0')
 668                return -EINVAL;
 669        if (par->family == NFPROTO_IPV4) {
 670                if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
 671                        return -EINVAL;
 672        } else {
 673                if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
 674                        return -EINVAL;
 675        }
 676
 677        if (info->cfg.mode & ~XT_HASHLIMIT_ALL) {
 678                pr_info("Unknown mode mask %X, kernel too old?\n",
 679                                                info->cfg.mode);
 680                return -EINVAL;
 681        }
 682
 683        /* Check for overflow. */
 684        if (info->cfg.mode & XT_HASHLIMIT_BYTES) {
 685                if (user2credits_byte(info->cfg.avg) == 0) {
 686                        pr_info("overflow, rate too high: %u\n", info->cfg.avg);
 687                        return -EINVAL;
 688                }
 689        } else if (info->cfg.burst == 0 ||
 690                    user2credits(info->cfg.avg * info->cfg.burst) <
 691                    user2credits(info->cfg.avg)) {
 692                        pr_info("overflow, try lower: %u/%u\n",
 693                                info->cfg.avg, info->cfg.burst);
 694                        return -ERANGE;
 695        }
 696
 697        mutex_lock(&hashlimit_mutex);
 698        info->hinfo = htable_find_get(net, info->name, par->family);
 699        if (info->hinfo == NULL) {
 700                ret = htable_create(net, info, par->family);
 701                if (ret < 0) {
 702                        mutex_unlock(&hashlimit_mutex);
 703                        return ret;
 704                }
 705        }
 706        mutex_unlock(&hashlimit_mutex);
 707        return 0;
 708}
 709
 710static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
 711{
 712        const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 713
 714        htable_put(info->hinfo);
 715}
 716
 717static struct xt_match hashlimit_mt_reg[] __read_mostly = {
 718        {
 719                .name           = "hashlimit",
 720                .revision       = 1,
 721                .family         = NFPROTO_IPV4,
 722                .match          = hashlimit_mt,
 723                .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 724                .checkentry     = hashlimit_mt_check,
 725                .destroy        = hashlimit_mt_destroy,
 726                .me             = THIS_MODULE,
 727        },
 728#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 729        {
 730                .name           = "hashlimit",
 731                .revision       = 1,
 732                .family         = NFPROTO_IPV6,
 733                .match          = hashlimit_mt,
 734                .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 735                .checkentry     = hashlimit_mt_check,
 736                .destroy        = hashlimit_mt_destroy,
 737                .me             = THIS_MODULE,
 738        },
 739#endif
 740};
 741
 742/* PROC stuff */
 743static void *dl_seq_start(struct seq_file *s, loff_t *pos)
 744        __acquires(htable->lock)
 745{
 746        struct xt_hashlimit_htable *htable = s->private;
 747        unsigned int *bucket;
 748
 749        spin_lock_bh(&htable->lock);
 750        if (*pos >= htable->cfg.size)
 751                return NULL;
 752
 753        bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
 754        if (!bucket)
 755                return ERR_PTR(-ENOMEM);
 756
 757        *bucket = *pos;
 758        return bucket;
 759}
 760
 761static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
 762{
 763        struct xt_hashlimit_htable *htable = s->private;
 764        unsigned int *bucket = (unsigned int *)v;
 765
 766        *pos = ++(*bucket);
 767        if (*pos >= htable->cfg.size) {
 768                kfree(v);
 769                return NULL;
 770        }
 771        return bucket;
 772}
 773
 774static void dl_seq_stop(struct seq_file *s, void *v)
 775        __releases(htable->lock)
 776{
 777        struct xt_hashlimit_htable *htable = s->private;
 778        unsigned int *bucket = (unsigned int *)v;
 779
 780        if (!IS_ERR(bucket))
 781                kfree(bucket);
 782        spin_unlock_bh(&htable->lock);
 783}
 784
 785static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
 786                                   struct seq_file *s)
 787{
 788        int res;
 789        const struct xt_hashlimit_htable *ht = s->private;
 790
 791        spin_lock(&ent->lock);
 792        /* recalculate to show accurate numbers */
 793        rateinfo_recalc(ent, jiffies, ht->cfg.mode);
 794
 795        switch (family) {
 796        case NFPROTO_IPV4:
 797                res = seq_printf(s, "%ld %pI4:%u->%pI4:%u %u %u %u\n",
 798                                 (long)(ent->expires - jiffies)/HZ,
 799                                 &ent->dst.ip.src,
 800                                 ntohs(ent->dst.src_port),
 801                                 &ent->dst.ip.dst,
 802                                 ntohs(ent->dst.dst_port),
 803                                 ent->rateinfo.credit, ent->rateinfo.credit_cap,
 804                                 ent->rateinfo.cost);
 805                break;
 806#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 807        case NFPROTO_IPV6:
 808                res = seq_printf(s, "%ld %pI6:%u->%pI6:%u %u %u %u\n",
 809                                 (long)(ent->expires - jiffies)/HZ,
 810                                 &ent->dst.ip6.src,
 811                                 ntohs(ent->dst.src_port),
 812                                 &ent->dst.ip6.dst,
 813                                 ntohs(ent->dst.dst_port),
 814                                 ent->rateinfo.credit, ent->rateinfo.credit_cap,
 815                                 ent->rateinfo.cost);
 816                break;
 817#endif
 818        default:
 819                BUG();
 820                res = 0;
 821        }
 822        spin_unlock(&ent->lock);
 823        return res;
 824}
 825
 826static int dl_seq_show(struct seq_file *s, void *v)
 827{
 828        struct xt_hashlimit_htable *htable = s->private;
 829        unsigned int *bucket = (unsigned int *)v;
 830        struct dsthash_ent *ent;
 831
 832        if (!hlist_empty(&htable->hash[*bucket])) {
 833                hlist_for_each_entry(ent, &htable->hash[*bucket], node)
 834                        if (dl_seq_real_show(ent, htable->family, s))
 835                                return -1;
 836        }
 837        return 0;
 838}
 839
 840static const struct seq_operations dl_seq_ops = {
 841        .start = dl_seq_start,
 842        .next  = dl_seq_next,
 843        .stop  = dl_seq_stop,
 844        .show  = dl_seq_show
 845};
 846
 847static int dl_proc_open(struct inode *inode, struct file *file)
 848{
 849        int ret = seq_open(file, &dl_seq_ops);
 850
 851        if (!ret) {
 852                struct seq_file *sf = file->private_data;
 853                sf->private = PDE_DATA(inode);
 854        }
 855        return ret;
 856}
 857
 858static const struct file_operations dl_file_ops = {
 859        .owner   = THIS_MODULE,
 860        .open    = dl_proc_open,
 861        .read    = seq_read,
 862        .llseek  = seq_lseek,
 863        .release = seq_release
 864};
 865
 866static int __net_init hashlimit_proc_net_init(struct net *net)
 867{
 868        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 869
 870        hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
 871        if (!hashlimit_net->ipt_hashlimit)
 872                return -ENOMEM;
 873#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 874        hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
 875        if (!hashlimit_net->ip6t_hashlimit) {
 876                remove_proc_entry("ipt_hashlimit", net->proc_net);
 877                return -ENOMEM;
 878        }
 879#endif
 880        return 0;
 881}
 882
 883static void __net_exit hashlimit_proc_net_exit(struct net *net)
 884{
 885        struct xt_hashlimit_htable *hinfo;
 886        struct proc_dir_entry *pde;
 887        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 888
 889        /* recent_net_exit() is called before recent_mt_destroy(). Make sure
 890         * that the parent xt_recent proc entry is is empty before trying to
 891         * remove it.
 892         */
 893        mutex_lock(&hashlimit_mutex);
 894        pde = hashlimit_net->ipt_hashlimit;
 895        if (pde == NULL)
 896                pde = hashlimit_net->ip6t_hashlimit;
 897
 898        hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
 899                remove_proc_entry(hinfo->name, pde);
 900
 901        hashlimit_net->ipt_hashlimit = NULL;
 902        hashlimit_net->ip6t_hashlimit = NULL;
 903        mutex_unlock(&hashlimit_mutex);
 904
 905        remove_proc_entry("ipt_hashlimit", net->proc_net);
 906#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 907        remove_proc_entry("ip6t_hashlimit", net->proc_net);
 908#endif
 909}
 910
 911static int __net_init hashlimit_net_init(struct net *net)
 912{
 913        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 914
 915        INIT_HLIST_HEAD(&hashlimit_net->htables);
 916        return hashlimit_proc_net_init(net);
 917}
 918
 919static void __net_exit hashlimit_net_exit(struct net *net)
 920{
 921        hashlimit_proc_net_exit(net);
 922}
 923
 924static struct pernet_operations hashlimit_net_ops = {
 925        .init   = hashlimit_net_init,
 926        .exit   = hashlimit_net_exit,
 927        .id     = &hashlimit_net_id,
 928        .size   = sizeof(struct hashlimit_net),
 929};
 930
 931static int __init hashlimit_mt_init(void)
 932{
 933        int err;
 934
 935        err = register_pernet_subsys(&hashlimit_net_ops);
 936        if (err < 0)
 937                return err;
 938        err = xt_register_matches(hashlimit_mt_reg,
 939              ARRAY_SIZE(hashlimit_mt_reg));
 940        if (err < 0)
 941                goto err1;
 942
 943        err = -ENOMEM;
 944        hashlimit_cachep = kmem_cache_create("xt_hashlimit",
 945                                            sizeof(struct dsthash_ent), 0, 0,
 946                                            NULL);
 947        if (!hashlimit_cachep) {
 948                pr_warning("unable to create slab cache\n");
 949                goto err2;
 950        }
 951        return 0;
 952
 953err2:
 954        xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
 955err1:
 956        unregister_pernet_subsys(&hashlimit_net_ops);
 957        return err;
 958
 959}
 960
 961static void __exit hashlimit_mt_exit(void)
 962{
 963        xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
 964        unregister_pernet_subsys(&hashlimit_net_ops);
 965
 966        rcu_barrier_bh();
 967        kmem_cache_destroy(hashlimit_cachep);
 968}
 969
 970module_init(hashlimit_mt_init);
 971module_exit(hashlimit_mt_exit);
 972