net/ipv4/netfilter/ipt_hashlimit.c

   1 /* iptables match extension to limit the number of packets per second
   2  * seperately for each hashbucket (sourceip/sourceport/dstip/dstport)
   3  *
   4  * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
   5  *
   6  * $Id: ipt_hashlimit.c 3244 2004-10-20 16:24:29Z laforge@netfilter.org $
   7  *
   8  * Development of this code was funded by Astaro AG, http://www.astaro.com/
   9  *
  10  * based on ipt_limit.c by:
  11  * Jérôme de Vivie    <devivie@info.enserb.u-bordeaux.fr>
  12  * Hervé Eychenne      <eychenne@info.enserb.u-bordeaux.fr>
  13  * Rusty Russell        <rusty@rustcorp.com.au>
  14  *
  15  * The general idea is to create a hash table for every dstip and have a
  16  * seperate limit counter per tuple.  This way you can do something like 'limit
  17  * the number of syn packets for each of my internal addresses.
  18  *
  19  * Ideally this would just be implemented as a general 'hash' match, which would
  20  * allow us to attach any iptables target to it's hash buckets.  But this is
  21  * not possible in the current iptables architecture.  As always, pkttables for
  22  * 2.7.x will help ;)
  23  */
  24 #include <linux/module.h>
  25 #include <linux/skbuff.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/random.h>
  28 #include <linux/jhash.h>
  29 #include <linux/slab.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/list.h>
  34
  35 #include <linux/netfilter_ipv4/ip_tables.h>
  36 #include <linux/netfilter_ipv4/ipt_hashlimit.h>
  37
  38 /* FIXME: this is just for IP_NF_ASSERRT */
  39 #include <linux/netfilter_ipv4/ip_conntrack.h>
  40 #include <linux/mutex.h>
  41
  42 MODULE_LICENSE("GPL");
  43 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
  44 MODULE_DESCRIPTION("iptables match for limiting per hash-bucket");
  45
  46 /* need to declare this at the top */
  47 static struct proc_dir_entry *hashlimit_procdir;
  48 static struct file_operations dl_file_ops;
  49
  50 /* hash table crap */
  51
  52 struct dsthash_dst {
  53         u_int32_t src_ip;
  54         u_int32_t dst_ip;
  55         /* ports have to be consecutive !!! */
  56         u_int16_t src_port;
  57         u_int16_t dst_port;
  58 };
  59
  60 struct dsthash_ent {
  61         /* static / read-only parts in the beginning */
  62         struct hlist_node node;
  63         struct dsthash_dst dst;
  64
  65         /* modified structure members in the end */
  66         unsigned long expires;          /* precalculated expiry time */
  67         struct {
  68                 unsigned long prev;     /* last modification */
  69                 u_int32_t credit;
  70                 u_int32_t credit_cap, cost;
  71         } rateinfo;
  72 };
  73
  74 struct ipt_hashlimit_htable {
  75         struct hlist_node node;         /* global list of all htables */
  76         atomic_t use;
  77
  78         struct hashlimit_cfg cfg;       /* config */
  79
  80         /* used internally */
  81         spinlock_t lock;                /* lock for list_head */
  82         u_int32_t rnd;                  /* random seed for hash */
  83         int rnd_initialized;
  84         struct timer_list timer;        /* timer for gc */
  85         atomic_t count;                 /* number entries in table */
  86
  87         /* seq_file stuff */
  88         struct proc_dir_entry *pde;
  89
  90         struct hlist_head hash[0];      /* hashtable itself */
  91 };
  92
  93 static DEFINE_SPINLOCK(hashlimit_lock); /* protects htables list */
  94 static DEFINE_MUTEX(hlimit_mutex);      /* additional checkentry protection */
  95 static HLIST_HEAD(hashlimit_htables);
  96 static kmem_cache_t *hashlimit_cachep __read_mostly;
  97
  98 static inline int dst_cmp(const struct dsthash_ent *ent, struct dsthash_dst *b)
  99 {
 100         return (ent->dst.dst_ip == b->dst_ip
 101                 && ent->dst.dst_port == b->dst_port
 102                 && ent->dst.src_port == b->src_port
 103                 && ent->dst.src_ip == b->src_ip);
 104 }
 105
 106 static inline u_int32_t
 107 hash_dst(const struct ipt_hashlimit_htable *ht, const struct dsthash_dst *dst)
 108 {
 109         return (jhash_3words(dst->dst_ip, (dst->dst_port<<16 | dst->src_port),
 110                              dst->src_ip, ht->rnd) % ht->cfg.size);
 111 }
 112
 113 static inline struct dsthash_ent *
 114 __dsthash_find(const struct ipt_hashlimit_htable *ht, struct dsthash_dst *dst)
 115 {
 116         struct dsthash_ent *ent;
 117         struct hlist_node *pos;
 118         u_int32_t hash = hash_dst(ht, dst);
 119
 120         if (!hlist_empty(&ht->hash[hash]))
 121                 hlist_for_each_entry(ent, pos, &ht->hash[hash], node) {
 122                         if (dst_cmp(ent, dst)) {
 123                                 return ent;
 124                         }
 125                 }
 126
 127         return NULL;
 128 }
 129
 130 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
 131 static struct dsthash_ent *
 132 __dsthash_alloc_init(struct ipt_hashlimit_htable *ht, struct dsthash_dst *dst)
 133 {
 134         struct dsthash_ent *ent;
 135
 136         /* initialize hash with random val at the time we allocate
 137          * the first hashtable entry */
 138         if (!ht->rnd_initialized) {
 139                 get_random_bytes(&ht->rnd, 4);
 140                 ht->rnd_initialized = 1;
 141         }
 142
 143         if (ht->cfg.max &&
 144             atomic_read(&ht->count) >= ht->cfg.max) {
 145                 /* FIXME: do something. question is what.. */
 146                 if (net_ratelimit())
 147                         printk(KERN_WARNING
 148                                 "ipt_hashlimit: max count of %u reached\n",
 149                                 ht->cfg.max);
 150                 return NULL;
 151         }
 152
 153         ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
 154         if (!ent) {
 155                 if (net_ratelimit())
 156                         printk(KERN_ERR
 157                                 "ipt_hashlimit: can't allocate dsthash_ent\n");
 158                 return NULL;
 159         }
 160
 161         atomic_inc(&ht->count);
 162
 163         ent->dst.dst_ip = dst->dst_ip;
 164         ent->dst.dst_port = dst->dst_port;
 165         ent->dst.src_ip = dst->src_ip;
 166         ent->dst.src_port = dst->src_port;
 167
 168         hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
 169
 170         return ent;
 171 }
 172
 173 static inline void
 174 __dsthash_free(struct ipt_hashlimit_htable *ht, struct dsthash_ent *ent)
 175 {
 176         hlist_del(&ent->node);
 177         kmem_cache_free(hashlimit_cachep, ent);
 178         atomic_dec(&ht->count);
 179 }
 180 static void htable_gc(unsigned long htlong);
 181
 182 static int htable_create(struct ipt_hashlimit_info *minfo)
 183 {
 184         int i;
 185         unsigned int size;
 186         struct ipt_hashlimit_htable *hinfo;
 187
 188         if (minfo->cfg.size)
 189                 size = minfo->cfg.size;
 190         else {
 191                 size = (((num_physpages << PAGE_SHIFT) / 16384)
 192                          / sizeof(struct list_head));
 193                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
 194                         size = 8192;
 195                 if (size < 16)
 196                         size = 16;
 197         }
 198         /* FIXME: don't use vmalloc() here or anywhere else -HW */
 199         hinfo = vmalloc(sizeof(struct ipt_hashlimit_htable)
 200                         + (sizeof(struct list_head) * size));
 201         if (!hinfo) {
 202                 printk(KERN_ERR "ipt_hashlimit: Unable to create hashtable\n");
 203                 return -1;
 204         }
 205         minfo->hinfo = hinfo;
 206
 207         /* copy match config into hashtable config */
 208         memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
 209         hinfo->cfg.size = size;
 210         if (!hinfo->cfg.max)
 211                 hinfo->cfg.max = 8 * hinfo->cfg.size;
 212         else if (hinfo->cfg.max < hinfo->cfg.size)
 213                 hinfo->cfg.max = hinfo->cfg.size;
 214
 215         for (i = 0; i < hinfo->cfg.size; i++)
 216                 INIT_HLIST_HEAD(&hinfo->hash[i]);
 217
 218         atomic_set(&hinfo->count, 0);
 219         atomic_set(&hinfo->use, 1);
 220         hinfo->rnd_initialized = 0;
 221         spin_lock_init(&hinfo->lock);
 222         hinfo->pde = create_proc_entry(minfo->name, 0, hashlimit_procdir);
 223         if (!hinfo->pde) {
 224                 vfree(hinfo);
 225                 return -1;
 226         }
 227         hinfo->pde->proc_fops = &dl_file_ops;
 228         hinfo->pde->data = hinfo;
 229
 230         init_timer(&hinfo->timer);
 231         hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
 232         hinfo->timer.data = (unsigned long )hinfo;
 233         hinfo->timer.function = htable_gc;
 234         add_timer(&hinfo->timer);
 235
 236         spin_lock_bh(&hashlimit_lock);
 237         hlist_add_head(&hinfo->node, &hashlimit_htables);
 238         spin_unlock_bh(&hashlimit_lock);
 239
 240         return 0;
 241 }
 242
 243 static int select_all(struct ipt_hashlimit_htable *ht, struct dsthash_ent *he)
 244 {
 245         return 1;
 246 }
 247
 248 static int select_gc(struct ipt_hashlimit_htable *ht, struct dsthash_ent *he)
 249 {
 250         return (jiffies >= he->expires);
 251 }
 252
 253 static void htable_selective_cleanup(struct ipt_hashlimit_htable *ht,
 254                                 int (*select)(struct ipt_hashlimit_htable *ht,
 255                                               struct dsthash_ent *he))
 256 {
 257         int i;
 258
 259         IP_NF_ASSERT(ht->cfg.size && ht->cfg.max);
 260
 261         /* lock hash table and iterate over it */
 262         spin_lock_bh(&ht->lock);
 263         for (i = 0; i < ht->cfg.size; i++) {
 264                 struct dsthash_ent *dh;
 265                 struct hlist_node *pos, *n;
 266                 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
 267                         if ((*select)(ht, dh))
 268                                 __dsthash_free(ht, dh);
 269                 }
 270         }
 271         spin_unlock_bh(&ht->lock);
 272 }
 273
 274 /* hash table garbage collector, run by timer */
 275 static void htable_gc(unsigned long htlong)
 276 {
 277         struct ipt_hashlimit_htable *ht = (struct ipt_hashlimit_htable *)htlong;
 278
 279         htable_selective_cleanup(ht, select_gc);
 280
 281         /* re-add the timer accordingly */
 282         ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
 283         add_timer(&ht->timer);
 284 }
 285
 286 static void htable_destroy(struct ipt_hashlimit_htable *hinfo)
 287 {
 288         /* remove timer, if it is pending */
 289         if (timer_pending(&hinfo->timer))
 290                 del_timer(&hinfo->timer);
 291
 292         /* remove proc entry */
 293         remove_proc_entry(hinfo->pde->name, hashlimit_procdir);
 294
 295         htable_selective_cleanup(hinfo, select_all);
 296         vfree(hinfo);
 297 }
 298
 299 static struct ipt_hashlimit_htable *htable_find_get(char *name)
 300 {
 301         struct ipt_hashlimit_htable *hinfo;
 302         struct hlist_node *pos;
 303
 304         spin_lock_bh(&hashlimit_lock);
 305         hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) {
 306                 if (!strcmp(name, hinfo->pde->name)) {
 307                         atomic_inc(&hinfo->use);
 308                         spin_unlock_bh(&hashlimit_lock);
 309                         return hinfo;
 310                 }
 311         }
 312         spin_unlock_bh(&hashlimit_lock);
 313
 314         return NULL;
 315 }
 316
 317 static void htable_put(struct ipt_hashlimit_htable *hinfo)
 318 {
 319         if (atomic_dec_and_test(&hinfo->use)) {
 320                 spin_lock_bh(&hashlimit_lock);
 321                 hlist_del(&hinfo->node);
 322                 spin_unlock_bh(&hashlimit_lock);
 323                 htable_destroy(hinfo);
 324         }
 325 }
 326
 327
 328 /* The algorithm used is the Simple Token Bucket Filter (TBF)
 329  * see net/sched/sch_tbf.c in the linux source tree
 330  */
 331
 332 /* Rusty: This is my (non-mathematically-inclined) understanding of
 333    this algorithm.  The `average rate' in jiffies becomes your initial
 334    amount of credit `credit' and the most credit you can ever have
 335    `credit_cap'.  The `peak rate' becomes the cost of passing the
 336    test, `cost'.
 337
 338    `prev' tracks the last packet hit: you gain one credit per jiffy.
 339    If you get credit balance more than this, the extra credit is
 340    discarded.  Every time the match passes, you lose `cost' credits;
 341    if you don't have that many, the test fails.
 342
 343    See Alexey's formal explanation in net/sched/sch_tbf.c.
 344
 345    To get the maximum range, we multiply by this factor (ie. you get N
 346    credits per jiffy).  We want to allow a rate as low as 1 per day
 347    (slowest userspace tool allows), which means
 348    CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
 349 */
 350 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
 351
 352 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
 353  * us the power of 2 below the theoretical max, so GCC simply does a
 354  * shift. */
 355 #define _POW2_BELOW2(x) ((x)|((x)>>1))
 356 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
 357 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
 358 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
 359 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
 360 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
 361
 362 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
 363
 364 /* Precision saver. */
 365 static inline u_int32_t
 366 user2credits(u_int32_t user)
 367 {
 368         /* If multiplying would overflow... */
 369         if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
 370                 /* Divide first. */
 371                 return (user / IPT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
 372
 373         return (user * HZ * CREDITS_PER_JIFFY) / IPT_HASHLIMIT_SCALE;
 374 }
 375
 376 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
 377 {
 378         dh->rateinfo.credit += (now - xchg(&dh->rateinfo.prev, now))
 379                                         * CREDITS_PER_JIFFY;
 380         if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
 381                 dh->rateinfo.credit = dh->rateinfo.credit_cap;
 382 }
 383
 384 static int
 385 hashlimit_match(const struct sk_buff *skb,
 386                 const struct net_device *in,
 387                 const struct net_device *out,
 388                 const struct xt_match *match,
 389                 const void *matchinfo,
 390                 int offset,
 391                 unsigned int protoff,
 392                 int *hotdrop)
 393 {
 394         struct ipt_hashlimit_info *r =
 395                 ((struct ipt_hashlimit_info *)matchinfo)->u.master;
 396         struct ipt_hashlimit_htable *hinfo = r->hinfo;
 397         unsigned long now = jiffies;
 398         struct dsthash_ent *dh;
 399         struct dsthash_dst dst;
 400
 401         /* build 'dst' according to hinfo->cfg and current packet */
 402         memset(&dst, 0, sizeof(dst));
 403         if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DIP)
 404                 dst.dst_ip = skb->nh.iph->daddr;
 405         if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SIP)
 406                 dst.src_ip = skb->nh.iph->saddr;
 407         if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DPT
 408             ||hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SPT) {
 409                 u_int16_t _ports[2], *ports;
 410
 411                 switch (skb->nh.iph->protocol) {
 412                 case IPPROTO_TCP:
 413                 case IPPROTO_UDP:
 414                 case IPPROTO_SCTP:
 415                 case IPPROTO_DCCP:
 416                         ports = skb_header_pointer(skb, skb->nh.iph->ihl*4,
 417                                                    sizeof(_ports), &_ports);
 418                         break;
 419                 default:
 420                         _ports[0] = _ports[1] = 0;
 421                         ports = _ports;
 422                         break;
 423                 }
 424                 if (!ports) {
 425                         /* We've been asked to examine this packet, and we
 426                           can't.  Hence, no choice but to drop. */
 427                         *hotdrop = 1;
 428                         return 0;
 429                 }
 430                 if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SPT)
 431                         dst.src_port = ports[0];
 432                 if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DPT)
 433                         dst.dst_port = ports[1];
 434         }
 435
 436         spin_lock_bh(&hinfo->lock);
 437         dh = __dsthash_find(hinfo, &dst);
 438         if (!dh) {
 439                 dh = __dsthash_alloc_init(hinfo, &dst);
 440
 441                 if (!dh) {
 442                         /* enomem... don't match == DROP */
 443                         if (net_ratelimit())
 444                                 printk(KERN_ERR "%s: ENOMEM\n", __FUNCTION__);
 445                         spin_unlock_bh(&hinfo->lock);
 446                         return 0;
 447                 }
 448
 449                 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
 450
 451                 dh->rateinfo.prev = jiffies;
 452                 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
 453                                                         hinfo->cfg.burst);
 454                 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
 455                                                         hinfo->cfg.burst);
 456                 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
 457         } else {
 458                 /* update expiration timeout */
 459                 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 460                 rateinfo_recalc(dh, now);
 461         }
 462
 463         if (dh->rateinfo.credit >= dh->rateinfo.cost) {
 464                 /* We're underlimit. */
 465                 dh->rateinfo.credit -= dh->rateinfo.cost;
 466                 spin_unlock_bh(&hinfo->lock);
 467                 return 1;
 468         }
 469
 470         spin_unlock_bh(&hinfo->lock);
 471
 472         /* default case: we're overlimit, thus don't match */
 473         return 0;
 474 }
 475
 476 static int
 477 hashlimit_checkentry(const char *tablename,
 478                      const void *inf,
 479                      const struct xt_match *match,
 480                      void *matchinfo,
 481                      unsigned int hook_mask)
 482 {
 483         struct ipt_hashlimit_info *r = matchinfo;
 484
 485         /* Check for overflow. */
 486         if (r->cfg.burst == 0
 487             || user2credits(r->cfg.avg * r->cfg.burst) <
 488                                         user2credits(r->cfg.avg)) {
 489                 printk(KERN_ERR "ipt_hashlimit: Overflow, try lower: %u/%u\n",
 490                        r->cfg.avg, r->cfg.burst);
 491                 return 0;
 492         }
 493
 494         if (r->cfg.mode == 0
 495             || r->cfg.mode > (IPT_HASHLIMIT_HASH_DPT
 496                           |IPT_HASHLIMIT_HASH_DIP
 497                           |IPT_HASHLIMIT_HASH_SIP
 498                           |IPT_HASHLIMIT_HASH_SPT))
 499                 return 0;
 500
 501         if (!r->cfg.gc_interval)
 502                 return 0;
 503
 504         if (!r->cfg.expire)
 505                 return 0;
 506
 507         if (r->name[sizeof(r->name) - 1] != '\0')
 508                 return 0;
 509
 510         /* This is the best we've got: We cannot release and re-grab lock,
 511          * since checkentry() is called before ip_tables.c grabs ipt_mutex.
 512          * We also cannot grab the hashtable spinlock, since htable_create will
 513          * call vmalloc, and that can sleep.  And we cannot just re-search
 514          * the list of htable's in htable_create(), since then we would
 515          * create duplicate proc files. -HW */
 516         mutex_lock(&hlimit_mutex);
 517         r->hinfo = htable_find_get(r->name);
 518         if (!r->hinfo && (htable_create(r) != 0)) {
 519                 mutex_unlock(&hlimit_mutex);
 520                 return 0;
 521         }
 522         mutex_unlock(&hlimit_mutex);
 523
 524         /* Ugly hack: For SMP, we only want to use one set */
 525         r->u.master = r;
 526
 527         return 1;
 528 }
 529
 530 static void
 531 hashlimit_destroy(const struct xt_match *match, void *matchinfo)
 532 {
 533         struct ipt_hashlimit_info *r = matchinfo;
 534
 535         htable_put(r->hinfo);
 536 }
 537
 538 #ifdef CONFIG_COMPAT
 539 struct compat_ipt_hashlimit_info {
 540         char name[IFNAMSIZ];
 541         struct hashlimit_cfg cfg;
 542         compat_uptr_t hinfo;
 543         compat_uptr_t master;
 544 };
 545
 546 static void compat_from_user(void *dst, void *src)
 547 {
 548         int off = offsetof(struct compat_ipt_hashlimit_info, hinfo);
 549
 550         memcpy(dst, src, off);
 551         memset(dst + off, 0, sizeof(struct compat_ipt_hashlimit_info) - off);
 552 }
 553
 554 static int compat_to_user(void __user *dst, void *src)
 555 {
 556         int off = offsetof(struct compat_ipt_hashlimit_info, hinfo);
 557
 558         return copy_to_user(dst, src, off) ? -EFAULT : 0;
 559 }
 560 #endif
 561
 562 static struct ipt_match ipt_hashlimit = {
 563         .name           = "hashlimit",
 564         .match          = hashlimit_match,
 565         .matchsize      = sizeof(struct ipt_hashlimit_info),
 566 #ifdef CONFIG_COMPAT
 567         .compatsize     = sizeof(struct compat_ipt_hashlimit_info),
 568         .compat_from_user = compat_from_user,
 569         .compat_to_user = compat_to_user,
 570 #endif
 571         .checkentry     = hashlimit_checkentry,
 572         .destroy        = hashlimit_destroy,
 573         .me             = THIS_MODULE
 574 };
 575
 576 /* PROC stuff */
 577
 578 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
 579 {
 580         struct proc_dir_entry *pde = s->private;
 581         struct ipt_hashlimit_htable *htable = pde->data;
 582         unsigned int *bucket;
 583
 584         spin_lock_bh(&htable->lock);
 585         if (*pos >= htable->cfg.size)
 586                 return NULL;
 587
 588         bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
 589         if (!bucket)
 590                 return ERR_PTR(-ENOMEM);
 591
 592         *bucket = *pos;
 593         return bucket;
 594 }
 595
 596 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
 597 {
 598         struct proc_dir_entry *pde = s->private;
 599         struct ipt_hashlimit_htable *htable = pde->data;
 600         unsigned int *bucket = (unsigned int *)v;
 601
 602         *pos = ++(*bucket);
 603         if (*pos >= htable->cfg.size) {
 604                 kfree(v);
 605                 return NULL;
 606         }
 607         return bucket;
 608 }
 609
 610 static void dl_seq_stop(struct seq_file *s, void *v)
 611 {
 612         struct proc_dir_entry *pde = s->private;
 613         struct ipt_hashlimit_htable *htable = pde->data;
 614         unsigned int *bucket = (unsigned int *)v;
 615
 616         kfree(bucket);
 617
 618         spin_unlock_bh(&htable->lock);
 619 }
 620
 621 static inline int dl_seq_real_show(struct dsthash_ent *ent, struct seq_file *s)
 622 {
 623         /* recalculate to show accurate numbers */
 624         rateinfo_recalc(ent, jiffies);
 625
 626         return seq_printf(s, "%ld %u.%u.%u.%u:%u->%u.%u.%u.%u:%u %u %u %u\n",
 627                         (long)(ent->expires - jiffies)/HZ,
 628                         NIPQUAD(ent->dst.src_ip), ntohs(ent->dst.src_port),
 629                         NIPQUAD(ent->dst.dst_ip), ntohs(ent->dst.dst_port),
 630                         ent->rateinfo.credit, ent->rateinfo.credit_cap,
 631                         ent->rateinfo.cost);
 632 }
 633
 634 static int dl_seq_show(struct seq_file *s, void *v)
 635 {
 636         struct proc_dir_entry *pde = s->private;
 637         struct ipt_hashlimit_htable *htable = pde->data;
 638         unsigned int *bucket = (unsigned int *)v;
 639         struct dsthash_ent *ent;
 640         struct hlist_node *pos;
 641
 642         if (!hlist_empty(&htable->hash[*bucket]))
 643                 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node) {
 644                         if (dl_seq_real_show(ent, s)) {
 645                                 /* buffer was filled and unable to print that tuple */
 646                                 return 1;
 647                         }
 648                 }
 649
 650         return 0;
 651 }
 652
 653 static struct seq_operations dl_seq_ops = {
 654         .start = dl_seq_start,
 655         .next  = dl_seq_next,
 656         .stop  = dl_seq_stop,
 657         .show  = dl_seq_show
 658 };
 659
 660 static int dl_proc_open(struct inode *inode, struct file *file)
 661 {
 662         int ret = seq_open(file, &dl_seq_ops);
 663
 664         if (!ret) {
 665                 struct seq_file *sf = file->private_data;
 666                 sf->private = PDE(inode);
 667         }
 668         return ret;
 669 }
 670
 671 static struct file_operations dl_file_ops = {
 672         .owner   = THIS_MODULE,
 673         .open    = dl_proc_open,
 674         .read    = seq_read,
 675         .llseek  = seq_lseek,
 676         .release = seq_release
 677 };
 678
 679 static int init_or_fini(int fini)
 680 {
 681         int ret = 0;
 682
 683         if (fini)
 684                 goto cleanup;
 685
 686         if (ipt_register_match(&ipt_hashlimit)) {
 687                 ret = -EINVAL;
 688                 goto cleanup_nothing;
 689         }
 690
 691         hashlimit_cachep = kmem_cache_create("ipt_hashlimit",
 692                                             sizeof(struct dsthash_ent), 0,
 693                                             0, NULL, NULL);
 694         if (!hashlimit_cachep) {
 695                 printk(KERN_ERR "Unable to create ipt_hashlimit slab cache\n");
 696                 ret = -ENOMEM;
 697                 goto cleanup_unreg_match;
 698         }
 699
 700         hashlimit_procdir = proc_mkdir("ipt_hashlimit", proc_net);
 701         if (!hashlimit_procdir) {
 702                 printk(KERN_ERR "Unable to create proc dir entry\n");
 703                 ret = -ENOMEM;
 704                 goto cleanup_free_slab;
 705         }
 706
 707         return ret;
 708
 709 cleanup:
 710         remove_proc_entry("ipt_hashlimit", proc_net);
 711 cleanup_free_slab:
 712         kmem_cache_destroy(hashlimit_cachep);
 713 cleanup_unreg_match:
 714         ipt_unregister_match(&ipt_hashlimit);
 715 cleanup_nothing:
 716         return ret;
 717
 718 }
 719
 720 static int __init ipt_hashlimit_init(void)
 721 {
 722         return init_or_fini(0);
 723 }
 724
 725 static void __exit ipt_hashlimit_fini(void)
 726 {
 727         init_or_fini(1);
 728 }
 729
 730 module_init(ipt_hashlimit_init);
 731 module_exit(ipt_hashlimit_fini);