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