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)
 161{
 162        struct dsthash_ent *ent;
 163
 164        spin_lock(&ht->lock);
 165        /* initialize hash with random val at the time we allocate
 166         * the first hashtable entry */
 167        if (unlikely(!ht->rnd_initialized)) {
 168                get_random_bytes(&ht->rnd, sizeof(ht->rnd));
 169                ht->rnd_initialized = true;
 170        }
 171
 172        if (ht->cfg.max && ht->count >= ht->cfg.max) {
 173                /* FIXME: do something. question is what.. */
 174                net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
 175                ent = NULL;
 176        } else
 177                ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
 178        if (ent) {
 179                memcpy(&ent->dst, dst, sizeof(ent->dst));
 180                spin_lock_init(&ent->lock);
 181
 182                spin_lock(&ent->lock);
 183                hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
 184                ht->count++;
 185        }
 186        spin_unlock(&ht->lock);
 187        return ent;
 188}
 189
 190static void dsthash_free_rcu(struct rcu_head *head)
 191{
 192        struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
 193
 194        kmem_cache_free(hashlimit_cachep, ent);
 195}
 196
 197static inline void
 198dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
 199{
 200        hlist_del_rcu(&ent->node);
 201        call_rcu_bh(&ent->rcu, dsthash_free_rcu);
 202        ht->count--;
 203}
 204static void htable_gc(unsigned long htlong);
 205
 206static int htable_create(struct net *net, struct xt_hashlimit_mtinfo1 *minfo,
 207                         u_int8_t family)
 208{
 209        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 210        struct xt_hashlimit_htable *hinfo;
 211        unsigned int size;
 212        unsigned int i;
 213
 214        if (minfo->cfg.size) {
 215                size = minfo->cfg.size;
 216        } else {
 217                size = (totalram_pages << PAGE_SHIFT) / 16384 /
 218                       sizeof(struct list_head);
 219                if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
 220                        size = 8192;
 221                if (size < 16)
 222                        size = 16;
 223        }
 224        /* FIXME: don't use vmalloc() here or anywhere else -HW */
 225        hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
 226                        sizeof(struct list_head) * size);
 227        if (hinfo == NULL)
 228                return -ENOMEM;
 229        minfo->hinfo = hinfo;
 230
 231        /* copy match config into hashtable config */
 232        memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
 233        hinfo->cfg.size = size;
 234        if (hinfo->cfg.max == 0)
 235                hinfo->cfg.max = 8 * hinfo->cfg.size;
 236        else if (hinfo->cfg.max < hinfo->cfg.size)
 237                hinfo->cfg.max = hinfo->cfg.size;
 238
 239        for (i = 0; i < hinfo->cfg.size; i++)
 240                INIT_HLIST_HEAD(&hinfo->hash[i]);
 241
 242        hinfo->use = 1;
 243        hinfo->count = 0;
 244        hinfo->family = family;
 245        hinfo->rnd_initialized = false;
 246        spin_lock_init(&hinfo->lock);
 247
 248        hinfo->pde = proc_create_data(minfo->name, 0,
 249                (family == NFPROTO_IPV4) ?
 250                hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
 251                &dl_file_ops, hinfo);
 252        if (hinfo->pde == NULL) {
 253                vfree(hinfo);
 254                return -ENOMEM;
 255        }
 256        hinfo->net = net;
 257
 258        setup_timer(&hinfo->timer, htable_gc, (unsigned long)hinfo);
 259        hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
 260        add_timer(&hinfo->timer);
 261
 262        hlist_add_head(&hinfo->node, &hashlimit_net->htables);
 263
 264        return 0;
 265}
 266
 267static bool select_all(const struct xt_hashlimit_htable *ht,
 268                       const struct dsthash_ent *he)
 269{
 270        return 1;
 271}
 272
 273static bool select_gc(const struct xt_hashlimit_htable *ht,
 274                      const struct dsthash_ent *he)
 275{
 276        return time_after_eq(jiffies, he->expires);
 277}
 278
 279static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
 280                        bool (*select)(const struct xt_hashlimit_htable *ht,
 281                                      const struct dsthash_ent *he))
 282{
 283        unsigned int i;
 284
 285        /* lock hash table and iterate over it */
 286        spin_lock_bh(&ht->lock);
 287        for (i = 0; i < ht->cfg.size; i++) {
 288                struct dsthash_ent *dh;
 289                struct hlist_node *pos, *n;
 290                hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
 291                        if ((*select)(ht, dh))
 292                                dsthash_free(ht, dh);
 293                }
 294        }
 295        spin_unlock_bh(&ht->lock);
 296}
 297
 298/* hash table garbage collector, run by timer */
 299static void htable_gc(unsigned long htlong)
 300{
 301        struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
 302
 303        htable_selective_cleanup(ht, select_gc);
 304
 305        /* re-add the timer accordingly */
 306        ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
 307        add_timer(&ht->timer);
 308}
 309
 310static void htable_destroy(struct xt_hashlimit_htable *hinfo)
 311{
 312        struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
 313        struct proc_dir_entry *parent;
 314
 315        del_timer_sync(&hinfo->timer);
 316
 317        if (hinfo->family == NFPROTO_IPV4)
 318                parent = hashlimit_net->ipt_hashlimit;
 319        else
 320                parent = hashlimit_net->ip6t_hashlimit;
 321        remove_proc_entry(hinfo->pde->name, parent);
 322        htable_selective_cleanup(hinfo, select_all);
 323        vfree(hinfo);
 324}
 325
 326static struct xt_hashlimit_htable *htable_find_get(struct net *net,
 327                                                   const char *name,
 328                                                   u_int8_t family)
 329{
 330        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 331        struct xt_hashlimit_htable *hinfo;
 332        struct hlist_node *pos;
 333
 334        hlist_for_each_entry(hinfo, pos, &hashlimit_net->htables, node) {
 335                if (!strcmp(name, hinfo->pde->name) &&
 336                    hinfo->family == family) {
 337                        hinfo->use++;
 338                        return hinfo;
 339                }
 340        }
 341        return NULL;
 342}
 343
 344static void htable_put(struct xt_hashlimit_htable *hinfo)
 345{
 346        mutex_lock(&hashlimit_mutex);
 347        if (--hinfo->use == 0) {
 348                hlist_del(&hinfo->node);
 349                htable_destroy(hinfo);
 350        }
 351        mutex_unlock(&hashlimit_mutex);
 352}
 353
 354/* The algorithm used is the Simple Token Bucket Filter (TBF)
 355 * see net/sched/sch_tbf.c in the linux source tree
 356 */
 357
 358/* Rusty: This is my (non-mathematically-inclined) understanding of
 359   this algorithm.  The `average rate' in jiffies becomes your initial
 360   amount of credit `credit' and the most credit you can ever have
 361   `credit_cap'.  The `peak rate' becomes the cost of passing the
 362   test, `cost'.
 363
 364   `prev' tracks the last packet hit: you gain one credit per jiffy.
 365   If you get credit balance more than this, the extra credit is
 366   discarded.  Every time the match passes, you lose `cost' credits;
 367   if you don't have that many, the test fails.
 368
 369   See Alexey's formal explanation in net/sched/sch_tbf.c.
 370
 371   To get the maximum range, we multiply by this factor (ie. you get N
 372   credits per jiffy).  We want to allow a rate as low as 1 per day
 373   (slowest userspace tool allows), which means
 374   CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
 375*/
 376#define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
 377
 378/* Repeated shift and or gives us all 1s, final shift and add 1 gives
 379 * us the power of 2 below the theoretical max, so GCC simply does a
 380 * shift. */
 381#define _POW2_BELOW2(x) ((x)|((x)>>1))
 382#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
 383#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
 384#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
 385#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
 386#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
 387
 388#define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
 389
 390/* in byte mode, the lowest possible rate is one packet/second.
 391 * credit_cap is used as a counter that tells us how many times we can
 392 * refill the "credits available" counter when it becomes empty.
 393 */
 394#define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
 395#define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
 396
 397static u32 xt_hashlimit_len_to_chunks(u32 len)
 398{
 399        return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
 400}
 401
 402/* Precision saver. */
 403static u32 user2credits(u32 user)
 404{
 405        /* If multiplying would overflow... */
 406        if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
 407                /* Divide first. */
 408                return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
 409
 410        return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
 411}
 412
 413static u32 user2credits_byte(u32 user)
 414{
 415        u64 us = user;
 416        us *= HZ * CREDITS_PER_JIFFY_BYTES;
 417        return (u32) (us >> 32);
 418}
 419
 420static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now, u32 mode)
 421{
 422        unsigned long delta = now - dh->rateinfo.prev;
 423        u32 cap;
 424
 425        if (delta == 0)
 426                return;
 427
 428        dh->rateinfo.prev = now;
 429
 430        if (mode & XT_HASHLIMIT_BYTES) {
 431                u32 tmp = dh->rateinfo.credit;
 432                dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
 433                cap = CREDITS_PER_JIFFY_BYTES * HZ;
 434                if (tmp >= dh->rateinfo.credit) {/* overflow */
 435                        dh->rateinfo.credit = cap;
 436                        return;
 437                }
 438        } else {
 439                dh->rateinfo.credit += delta * CREDITS_PER_JIFFY;
 440                cap = dh->rateinfo.credit_cap;
 441        }
 442        if (dh->rateinfo.credit > cap)
 443                dh->rateinfo.credit = cap;
 444}
 445
 446static void rateinfo_init(struct dsthash_ent *dh,
 447                          struct xt_hashlimit_htable *hinfo)
 448{
 449        dh->rateinfo.prev = jiffies;
 450        if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
 451                dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
 452                dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
 453                dh->rateinfo.credit_cap = hinfo->cfg.burst;
 454        } else {
 455                dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
 456                                                   hinfo->cfg.burst);
 457                dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
 458                dh->rateinfo.credit_cap = dh->rateinfo.credit;
 459        }
 460}
 461
 462static inline __be32 maskl(__be32 a, unsigned int l)
 463{
 464        return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
 465}
 466
 467#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 468static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
 469{
 470        switch (p) {
 471        case 0 ... 31:
 472                i[0] = maskl(i[0], p);
 473                i[1] = i[2] = i[3] = 0;
 474                break;
 475        case 32 ... 63:
 476                i[1] = maskl(i[1], p - 32);
 477                i[2] = i[3] = 0;
 478                break;
 479        case 64 ... 95:
 480                i[2] = maskl(i[2], p - 64);
 481                i[3] = 0;
 482                break;
 483        case 96 ... 127:
 484                i[3] = maskl(i[3], p - 96);
 485                break;
 486        case 128:
 487                break;
 488        }
 489}
 490#endif
 491
 492static int
 493hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
 494                   struct dsthash_dst *dst,
 495                   const struct sk_buff *skb, unsigned int protoff)
 496{
 497        __be16 _ports[2], *ports;
 498        u8 nexthdr;
 499        int poff;
 500
 501        memset(dst, 0, sizeof(*dst));
 502
 503        switch (hinfo->family) {
 504        case NFPROTO_IPV4:
 505                if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
 506                        dst->ip.dst = maskl(ip_hdr(skb)->daddr,
 507                                      hinfo->cfg.dstmask);
 508                if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
 509                        dst->ip.src = maskl(ip_hdr(skb)->saddr,
 510                                      hinfo->cfg.srcmask);
 511
 512                if (!(hinfo->cfg.mode &
 513                      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 514                        return 0;
 515                nexthdr = ip_hdr(skb)->protocol;
 516                break;
 517#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 518        case NFPROTO_IPV6:
 519        {
 520                __be16 frag_off;
 521
 522                if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
 523                        memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
 524                               sizeof(dst->ip6.dst));
 525                        hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
 526                }
 527                if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
 528                        memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
 529                               sizeof(dst->ip6.src));
 530                        hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
 531                }
 532
 533                if (!(hinfo->cfg.mode &
 534                      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 535                        return 0;
 536                nexthdr = ipv6_hdr(skb)->nexthdr;
 537                protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
 538                if ((int)protoff < 0)
 539                        return -1;
 540                break;
 541        }
 542#endif
 543        default:
 544                BUG();
 545                return 0;
 546        }
 547
 548        poff = proto_ports_offset(nexthdr);
 549        if (poff >= 0) {
 550                ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
 551                                           &_ports);
 552        } else {
 553                _ports[0] = _ports[1] = 0;
 554                ports = _ports;
 555        }
 556        if (!ports)
 557                return -1;
 558        if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
 559                dst->src_port = ports[0];
 560        if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
 561                dst->dst_port = ports[1];
 562        return 0;
 563}
 564
 565static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
 566{
 567        u64 tmp = xt_hashlimit_len_to_chunks(len);
 568        tmp = tmp * dh->rateinfo.cost;
 569
 570        if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
 571                tmp = CREDITS_PER_JIFFY_BYTES * HZ;
 572
 573        if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
 574                dh->rateinfo.credit_cap--;
 575                dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
 576        }
 577        return (u32) tmp;
 578}
 579
 580static bool
 581hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
 582{
 583        const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 584        struct xt_hashlimit_htable *hinfo = info->hinfo;
 585        unsigned long now = jiffies;
 586        struct dsthash_ent *dh;
 587        struct dsthash_dst dst;
 588        u32 cost;
 589
 590        if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
 591                goto hotdrop;
 592
 593        rcu_read_lock_bh();
 594        dh = dsthash_find(hinfo, &dst);
 595        if (dh == NULL) {
 596                dh = dsthash_alloc_init(hinfo, &dst);
 597                if (dh == NULL) {
 598                        rcu_read_unlock_bh();
 599                        goto hotdrop;
 600                }
 601                dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
 602                rateinfo_init(dh, hinfo);
 603        } else {
 604                /* update expiration timeout */
 605                dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 606                rateinfo_recalc(dh, now, hinfo->cfg.mode);
 607        }
 608
 609        if (info->cfg.mode & XT_HASHLIMIT_BYTES)
 610                cost = hashlimit_byte_cost(skb->len, dh);
 611        else
 612                cost = dh->rateinfo.cost;
 613
 614        if (dh->rateinfo.credit >= cost) {
 615                /* below the limit */
 616                dh->rateinfo.credit -= cost;
 617                spin_unlock(&dh->lock);
 618                rcu_read_unlock_bh();
 619                return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
 620        }
 621
 622        spin_unlock(&dh->lock);
 623        rcu_read_unlock_bh();
 624        /* default match is underlimit - so over the limit, we need to invert */
 625        return info->cfg.mode & XT_HASHLIMIT_INVERT;
 626
 627 hotdrop:
 628        par->hotdrop = true;
 629        return false;
 630}
 631
 632static int hashlimit_mt_check(const struct xt_mtchk_param *par)
 633{
 634        struct net *net = par->net;
 635        struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 636        int ret;
 637
 638        if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
 639                return -EINVAL;
 640        if (info->name[sizeof(info->name)-1] != '\0')
 641                return -EINVAL;
 642        if (par->family == NFPROTO_IPV4) {
 643                if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
 644                        return -EINVAL;
 645        } else {
 646                if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
 647                        return -EINVAL;
 648        }
 649
 650        if (info->cfg.mode & ~XT_HASHLIMIT_ALL) {
 651                pr_info("Unknown mode mask %X, kernel too old?\n",
 652                                                info->cfg.mode);
 653                return -EINVAL;
 654        }
 655
 656        /* Check for overflow. */
 657        if (info->cfg.mode & XT_HASHLIMIT_BYTES) {
 658                if (user2credits_byte(info->cfg.avg) == 0) {
 659                        pr_info("overflow, rate too high: %u\n", info->cfg.avg);
 660                        return -EINVAL;
 661                }
 662        } else if (info->cfg.burst == 0 ||
 663                    user2credits(info->cfg.avg * info->cfg.burst) <
 664                    user2credits(info->cfg.avg)) {
 665                        pr_info("overflow, try lower: %u/%u\n",
 666                                info->cfg.avg, info->cfg.burst);
 667                        return -ERANGE;
 668        }
 669
 670        mutex_lock(&hashlimit_mutex);
 671        info->hinfo = htable_find_get(net, info->name, par->family);
 672        if (info->hinfo == NULL) {
 673                ret = htable_create(net, info, par->family);
 674                if (ret < 0) {
 675                        mutex_unlock(&hashlimit_mutex);
 676                        return ret;
 677                }
 678        }
 679        mutex_unlock(&hashlimit_mutex);
 680        return 0;
 681}
 682
 683static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
 684{
 685        const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 686
 687        htable_put(info->hinfo);
 688}
 689
 690static struct xt_match hashlimit_mt_reg[] __read_mostly = {
 691        {
 692                .name           = "hashlimit",
 693                .revision       = 1,
 694                .family         = NFPROTO_IPV4,
 695                .match          = hashlimit_mt,
 696                .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 697                .checkentry     = hashlimit_mt_check,
 698                .destroy        = hashlimit_mt_destroy,
 699                .me             = THIS_MODULE,
 700        },
 701#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 702        {
 703                .name           = "hashlimit",
 704                .revision       = 1,
 705                .family         = NFPROTO_IPV6,
 706                .match          = hashlimit_mt,
 707                .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 708                .checkentry     = hashlimit_mt_check,
 709                .destroy        = hashlimit_mt_destroy,
 710                .me             = THIS_MODULE,
 711        },
 712#endif
 713};
 714
 715/* PROC stuff */
 716static void *dl_seq_start(struct seq_file *s, loff_t *pos)
 717        __acquires(htable->lock)
 718{
 719        struct xt_hashlimit_htable *htable = s->private;
 720        unsigned int *bucket;
 721
 722        spin_lock_bh(&htable->lock);
 723        if (*pos >= htable->cfg.size)
 724                return NULL;
 725
 726        bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
 727        if (!bucket)
 728                return ERR_PTR(-ENOMEM);
 729
 730        *bucket = *pos;
 731        return bucket;
 732}
 733
 734static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
 735{
 736        struct xt_hashlimit_htable *htable = s->private;
 737        unsigned int *bucket = (unsigned int *)v;
 738
 739        *pos = ++(*bucket);
 740        if (*pos >= htable->cfg.size) {
 741                kfree(v);
 742                return NULL;
 743        }
 744        return bucket;
 745}
 746
 747static void dl_seq_stop(struct seq_file *s, void *v)
 748        __releases(htable->lock)
 749{
 750        struct xt_hashlimit_htable *htable = s->private;
 751        unsigned int *bucket = (unsigned int *)v;
 752
 753        if (!IS_ERR(bucket))
 754                kfree(bucket);
 755        spin_unlock_bh(&htable->lock);
 756}
 757
 758static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
 759                                   struct seq_file *s)
 760{
 761        int res;
 762        const struct xt_hashlimit_htable *ht = s->private;
 763
 764        spin_lock(&ent->lock);
 765        /* recalculate to show accurate numbers */
 766        rateinfo_recalc(ent, jiffies, ht->cfg.mode);
 767
 768        switch (family) {
 769        case NFPROTO_IPV4:
 770                res = seq_printf(s, "%ld %pI4:%u->%pI4:%u %u %u %u\n",
 771                                 (long)(ent->expires - jiffies)/HZ,
 772                                 &ent->dst.ip.src,
 773                                 ntohs(ent->dst.src_port),
 774                                 &ent->dst.ip.dst,
 775                                 ntohs(ent->dst.dst_port),
 776                                 ent->rateinfo.credit, ent->rateinfo.credit_cap,
 777                                 ent->rateinfo.cost);
 778                break;
 779#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 780        case NFPROTO_IPV6:
 781                res = seq_printf(s, "%ld %pI6:%u->%pI6:%u %u %u %u\n",
 782                                 (long)(ent->expires - jiffies)/HZ,
 783                                 &ent->dst.ip6.src,
 784                                 ntohs(ent->dst.src_port),
 785                                 &ent->dst.ip6.dst,
 786                                 ntohs(ent->dst.dst_port),
 787                                 ent->rateinfo.credit, ent->rateinfo.credit_cap,
 788                                 ent->rateinfo.cost);
 789                break;
 790#endif
 791        default:
 792                BUG();
 793                res = 0;
 794        }
 795        spin_unlock(&ent->lock);
 796        return res;
 797}
 798
 799static int dl_seq_show(struct seq_file *s, void *v)
 800{
 801        struct xt_hashlimit_htable *htable = s->private;
 802        unsigned int *bucket = (unsigned int *)v;
 803        struct dsthash_ent *ent;
 804        struct hlist_node *pos;
 805
 806        if (!hlist_empty(&htable->hash[*bucket])) {
 807                hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
 808                        if (dl_seq_real_show(ent, htable->family, s))
 809                                return -1;
 810        }
 811        return 0;
 812}
 813
 814static const struct seq_operations dl_seq_ops = {
 815        .start = dl_seq_start,
 816        .next  = dl_seq_next,
 817        .stop  = dl_seq_stop,
 818        .show  = dl_seq_show
 819};
 820
 821static int dl_proc_open(struct inode *inode, struct file *file)
 822{
 823        int ret = seq_open(file, &dl_seq_ops);
 824
 825        if (!ret) {
 826                struct seq_file *sf = file->private_data;
 827                sf->private = PDE(inode)->data;
 828        }
 829        return ret;
 830}
 831
 832static const struct file_operations dl_file_ops = {
 833        .owner   = THIS_MODULE,
 834        .open    = dl_proc_open,
 835        .read    = seq_read,
 836        .llseek  = seq_lseek,
 837        .release = seq_release
 838};
 839
 840static int __net_init hashlimit_proc_net_init(struct net *net)
 841{
 842        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 843
 844        hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
 845        if (!hashlimit_net->ipt_hashlimit)
 846                return -ENOMEM;
 847#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 848        hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
 849        if (!hashlimit_net->ip6t_hashlimit) {
 850                proc_net_remove(net, "ipt_hashlimit");
 851                return -ENOMEM;
 852        }
 853#endif
 854        return 0;
 855}
 856
 857static void __net_exit hashlimit_proc_net_exit(struct net *net)
 858{
 859        proc_net_remove(net, "ipt_hashlimit");
 860#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 861        proc_net_remove(net, "ip6t_hashlimit");
 862#endif
 863}
 864
 865static int __net_init hashlimit_net_init(struct net *net)
 866{
 867        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 868
 869        INIT_HLIST_HEAD(&hashlimit_net->htables);
 870        return hashlimit_proc_net_init(net);
 871}
 872
 873static void __net_exit hashlimit_net_exit(struct net *net)
 874{
 875        struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 876
 877        BUG_ON(!hlist_empty(&hashlimit_net->htables));
 878        hashlimit_proc_net_exit(net);
 879}
 880
 881static struct pernet_operations hashlimit_net_ops = {
 882        .init   = hashlimit_net_init,
 883        .exit   = hashlimit_net_exit,
 884        .id     = &hashlimit_net_id,
 885        .size   = sizeof(struct hashlimit_net),
 886};
 887
 888static int __init hashlimit_mt_init(void)
 889{
 890        int err;
 891
 892        err = register_pernet_subsys(&hashlimit_net_ops);
 893        if (err < 0)
 894                return err;
 895        err = xt_register_matches(hashlimit_mt_reg,
 896              ARRAY_SIZE(hashlimit_mt_reg));
 897        if (err < 0)
 898                goto err1;
 899
 900        err = -ENOMEM;
 901        hashlimit_cachep = kmem_cache_create("xt_hashlimit",
 902                                            sizeof(struct dsthash_ent), 0, 0,
 903                                            NULL);
 904        if (!hashlimit_cachep) {
 905                pr_warning("unable to create slab cache\n");
 906                goto err2;
 907        }
 908        return 0;
 909
 910err2:
 911        xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
 912err1:
 913        unregister_pernet_subsys(&hashlimit_net_ops);
 914        return err;
 915
 916}
 917
 918static void __exit hashlimit_mt_exit(void)
 919{
 920        xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
 921        unregister_pernet_subsys(&hashlimit_net_ops);
 922
 923        rcu_barrier_bh();
 924        kmem_cache_destroy(hashlimit_cachep);
 925}
 926
 927module_init(hashlimit_mt_init);
 928module_exit(hashlimit_mt_exit);
 929
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.