net/netfilter/xt_hashlimit.c

   1 /*
   2  *      xt_hashlimit - Netfilter module to limit the number of packets per time
   3  *      seperately 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 #include <linux/module.h>
  11 #include <linux/spinlock.h>
  12 #include <linux/random.h>
  13 #include <linux/jhash.h>
  14 #include <linux/slab.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/proc_fs.h>
  17 #include <linux/seq_file.h>
  18 #include <linux/list.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/mm.h>
  21 #include <linux/in.h>
  22 #include <linux/ip.h>
  23 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
  24 #include <linux/ipv6.h>
  25 #include <net/ipv6.h>
  26 #endif
  27
  28 #include <net/net_namespace.h>
  29
  30 #include <linux/netfilter/x_tables.h>
  31 #include <linux/netfilter_ipv4/ip_tables.h>
  32 #include <linux/netfilter_ipv6/ip6_tables.h>
  33 #include <linux/netfilter/xt_hashlimit.h>
  34 #include <linux/mutex.h>
  35
  36 MODULE_LICENSE("GPL");
  37 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
  38 MODULE_AUTHOR("Jan Engelhardt <jengelh@computergmbh.de>");
  39 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
  40 MODULE_ALIAS("ipt_hashlimit");
  41 MODULE_ALIAS("ip6t_hashlimit");
  42
  43 /* need to declare this at the top */
  44 static struct proc_dir_entry *hashlimit_procdir4;
  45 static struct proc_dir_entry *hashlimit_procdir6;
  46 static const struct file_operations dl_file_ops;
  47
  48 /* hash table crap */
  49 struct dsthash_dst {
  50         union {
  51                 struct {
  52                         __be32 src;
  53                         __be32 dst;
  54                 } ip;
  55 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
  56                 struct {
  57                         __be32 src[4];
  58                         __be32 dst[4];
  59                 } ip6;
  60 #endif
  61         };
  62         __be16 src_port;
  63         __be16 dst_port;
  64 };
  65
  66 struct dsthash_ent {
  67         /* static / read-only parts in the beginning */
  68         struct hlist_node node;
  69         struct dsthash_dst dst;
  70
  71         /* modified structure members in the end */
  72         unsigned long expires;          /* precalculated expiry time */
  73         struct {
  74                 unsigned long prev;     /* last modification */
  75                 u_int32_t credit;
  76                 u_int32_t credit_cap, cost;
  77         } rateinfo;
  78 };
  79
  80 struct xt_hashlimit_htable {
  81         struct hlist_node node;         /* global list of all htables */
  82         atomic_t use;
  83         u_int8_t family;
  84
  85         struct hashlimit_cfg1 cfg;      /* config */
  86
  87         /* used internally */
  88         spinlock_t lock;                /* lock for list_head */
  89         u_int32_t rnd;                  /* random seed for hash */
  90         int rnd_initialized;
  91         unsigned int count;             /* number entries in table */
  92         struct timer_list timer;        /* timer for gc */
  93
  94         /* seq_file stuff */
  95         struct proc_dir_entry *pde;
  96
  97         struct hlist_head hash[0];      /* hashtable itself */
  98 };
  99
 100 static DEFINE_SPINLOCK(hashlimit_lock); /* protects htables list */
 101 static DEFINE_MUTEX(hlimit_mutex);      /* additional checkentry protection */
 102 static HLIST_HEAD(hashlimit_htables);
 103 static struct kmem_cache *hashlimit_cachep __read_mostly;
 104
 105 static inline bool dst_cmp(const struct dsthash_ent *ent,
 106                            const struct dsthash_dst *b)
 107 {
 108         return !memcmp(&ent->dst, b, sizeof(ent->dst));
 109 }
 110
 111 static u_int32_t
 112 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
 113 {
 114         u_int32_t hash = jhash2((const u32 *)dst,
 115                                 sizeof(*dst)/sizeof(u32),
 116                                 ht->rnd);
 117         /*
 118          * Instead of returning hash % ht->cfg.size (implying a divide)
 119          * we return the high 32 bits of the (hash * ht->cfg.size) that will
 120          * give results between [0 and cfg.size-1] and same hash distribution,
 121          * but using a multiply, less expensive than a divide
 122          */
 123         return ((u64)hash * ht->cfg.size) >> 32;
 124 }
 125
 126 static struct dsthash_ent *
 127 dsthash_find(const struct xt_hashlimit_htable *ht,
 128              const struct dsthash_dst *dst)
 129 {
 130         struct dsthash_ent *ent;
 131         struct hlist_node *pos;
 132         u_int32_t hash = hash_dst(ht, dst);
 133
 134         if (!hlist_empty(&ht->hash[hash])) {
 135                 hlist_for_each_entry(ent, pos, &ht->hash[hash], node)
 136                         if (dst_cmp(ent, dst))
 137                                 return ent;
 138         }
 139         return NULL;
 140 }
 141
 142 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
 143 static struct dsthash_ent *
 144 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
 145                    const struct dsthash_dst *dst)
 146 {
 147         struct dsthash_ent *ent;
 148
 149         /* initialize hash with random val at the time we allocate
 150          * the first hashtable entry */
 151         if (!ht->rnd_initialized) {
 152                 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
 153                 ht->rnd_initialized = 1;
 154         }
 155
 156         if (ht->cfg.max && ht->count >= ht->cfg.max) {
 157                 /* FIXME: do something. question is what.. */
 158                 if (net_ratelimit())
 159                         printk(KERN_WARNING
 160                                 "xt_hashlimit: max count of %u reached\n",
 161                                 ht->cfg.max);
 162                 return NULL;
 163         }
 164
 165         ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
 166         if (!ent) {
 167                 if (net_ratelimit())
 168                         printk(KERN_ERR
 169                                 "xt_hashlimit: can't allocate dsthash_ent\n");
 170                 return NULL;
 171         }
 172         memcpy(&ent->dst, dst, sizeof(ent->dst));
 173
 174         hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
 175         ht->count++;
 176         return ent;
 177 }
 178
 179 static inline void
 180 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
 181 {
 182         hlist_del(&ent->node);
 183         kmem_cache_free(hashlimit_cachep, ent);
 184         ht->count--;
 185 }
 186 static void htable_gc(unsigned long htlong);
 187
 188 static int htable_create_v0(struct xt_hashlimit_info *minfo, u_int8_t family)
 189 {
 190         struct xt_hashlimit_htable *hinfo;
 191         unsigned int size;
 192         unsigned int i;
 193
 194         if (minfo->cfg.size)
 195                 size = minfo->cfg.size;
 196         else {
 197                 size = ((num_physpages << PAGE_SHIFT) / 16384) /
 198                        sizeof(struct list_head);
 199                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
 200                         size = 8192;
 201                 if (size < 16)
 202                         size = 16;
 203         }
 204         /* FIXME: don't use vmalloc() here or anywhere else -HW */
 205         hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
 206                         sizeof(struct list_head) * size);
 207         if (!hinfo) {
 208                 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
 209                 return -1;
 210         }
 211         minfo->hinfo = hinfo;
 212
 213         /* copy match config into hashtable config */
 214         hinfo->cfg.mode        = minfo->cfg.mode;
 215         hinfo->cfg.avg         = minfo->cfg.avg;
 216         hinfo->cfg.burst       = minfo->cfg.burst;
 217         hinfo->cfg.max         = minfo->cfg.max;
 218         hinfo->cfg.gc_interval = minfo->cfg.gc_interval;
 219         hinfo->cfg.expire      = minfo->cfg.expire;
 220
 221         if (family == NFPROTO_IPV4)
 222                 hinfo->cfg.srcmask = hinfo->cfg.dstmask = 32;
 223         else
 224                 hinfo->cfg.srcmask = hinfo->cfg.dstmask = 128;
 225
 226         hinfo->cfg.size = size;
 227         if (!hinfo->cfg.max)
 228                 hinfo->cfg.max = 8 * hinfo->cfg.size;
 229         else if (hinfo->cfg.max < hinfo->cfg.size)
 230                 hinfo->cfg.max = hinfo->cfg.size;
 231
 232         for (i = 0; i < hinfo->cfg.size; i++)
 233                 INIT_HLIST_HEAD(&hinfo->hash[i]);
 234
 235         atomic_set(&hinfo->use, 1);
 236         hinfo->count = 0;
 237         hinfo->family = family;
 238         hinfo->rnd_initialized = 0;
 239         spin_lock_init(&hinfo->lock);
 240         hinfo->pde = proc_create_data(minfo->name, 0,
 241                 (family == NFPROTO_IPV4) ?
 242                 hashlimit_procdir4 : hashlimit_procdir6,
 243                 &dl_file_ops, hinfo);
 244         if (!hinfo->pde) {
 245                 vfree(hinfo);
 246                 return -1;
 247         }
 248
 249         setup_timer(&hinfo->timer, htable_gc, (unsigned long )hinfo);
 250         hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
 251         add_timer(&hinfo->timer);
 252
 253         spin_lock_bh(&hashlimit_lock);
 254         hlist_add_head(&hinfo->node, &hashlimit_htables);
 255         spin_unlock_bh(&hashlimit_lock);
 256
 257         return 0;
 258 }
 259
 260 static int htable_create(struct xt_hashlimit_mtinfo1 *minfo, u_int8_t family)
 261 {
 262         struct xt_hashlimit_htable *hinfo;
 263         unsigned int size;
 264         unsigned int i;
 265
 266         if (minfo->cfg.size) {
 267                 size = minfo->cfg.size;
 268         } else {
 269                 size = (num_physpages << PAGE_SHIFT) / 16384 /
 270                        sizeof(struct list_head);
 271                 if (num_physpages > 1024 * 1024 * 1024 / PAGE_SIZE)
 272                         size = 8192;
 273                 if (size < 16)
 274                         size = 16;
 275         }
 276         /* FIXME: don't use vmalloc() here or anywhere else -HW */
 277         hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
 278                         sizeof(struct list_head) * size);
 279         if (hinfo == NULL) {
 280                 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
 281                 return -1;
 282         }
 283         minfo->hinfo = hinfo;
 284
 285         /* copy match config into hashtable config */
 286         memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
 287         hinfo->cfg.size = size;
 288         if (hinfo->cfg.max == 0)
 289                 hinfo->cfg.max = 8 * hinfo->cfg.size;
 290         else if (hinfo->cfg.max < hinfo->cfg.size)
 291                 hinfo->cfg.max = hinfo->cfg.size;
 292
 293         for (i = 0; i < hinfo->cfg.size; i++)
 294                 INIT_HLIST_HEAD(&hinfo->hash[i]);
 295
 296         atomic_set(&hinfo->use, 1);
 297         hinfo->count = 0;
 298         hinfo->family = family;
 299         hinfo->rnd_initialized = 0;
 300         spin_lock_init(&hinfo->lock);
 301
 302         hinfo->pde = proc_create_data(minfo->name, 0,
 303                 (family == NFPROTO_IPV4) ?
 304                 hashlimit_procdir4 : hashlimit_procdir6,
 305                 &dl_file_ops, hinfo);
 306         if (hinfo->pde == NULL) {
 307                 vfree(hinfo);
 308                 return -1;
 309         }
 310
 311         setup_timer(&hinfo->timer, htable_gc, (unsigned long)hinfo);
 312         hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
 313         add_timer(&hinfo->timer);
 314
 315         spin_lock_bh(&hashlimit_lock);
 316         hlist_add_head(&hinfo->node, &hashlimit_htables);
 317         spin_unlock_bh(&hashlimit_lock);
 318
 319         return 0;
 320 }
 321
 322 static bool select_all(const struct xt_hashlimit_htable *ht,
 323                        const struct dsthash_ent *he)
 324 {
 325         return 1;
 326 }
 327
 328 static bool select_gc(const struct xt_hashlimit_htable *ht,
 329                       const struct dsthash_ent *he)
 330 {
 331         return time_after_eq(jiffies, he->expires);
 332 }
 333
 334 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
 335                         bool (*select)(const struct xt_hashlimit_htable *ht,
 336                                       const struct dsthash_ent *he))
 337 {
 338         unsigned int i;
 339
 340         /* lock hash table and iterate over it */
 341         spin_lock_bh(&ht->lock);
 342         for (i = 0; i < ht->cfg.size; i++) {
 343                 struct dsthash_ent *dh;
 344                 struct hlist_node *pos, *n;
 345                 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
 346                         if ((*select)(ht, dh))
 347                                 dsthash_free(ht, dh);
 348                 }
 349         }
 350         spin_unlock_bh(&ht->lock);
 351 }
 352
 353 /* hash table garbage collector, run by timer */
 354 static void htable_gc(unsigned long htlong)
 355 {
 356         struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
 357
 358         htable_selective_cleanup(ht, select_gc);
 359
 360         /* re-add the timer accordingly */
 361         ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
 362         add_timer(&ht->timer);
 363 }
 364
 365 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
 366 {
 367         del_timer_sync(&hinfo->timer);
 368
 369         /* remove proc entry */
 370         remove_proc_entry(hinfo->pde->name,
 371                           hinfo->family == NFPROTO_IPV4 ? hashlimit_procdir4 :
 372                                                      hashlimit_procdir6);
 373         htable_selective_cleanup(hinfo, select_all);
 374         vfree(hinfo);
 375 }
 376
 377 static struct xt_hashlimit_htable *htable_find_get(const char *name,
 378                                                    u_int8_t family)
 379 {
 380         struct xt_hashlimit_htable *hinfo;
 381         struct hlist_node *pos;
 382
 383         spin_lock_bh(&hashlimit_lock);
 384         hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) {
 385                 if (!strcmp(name, hinfo->pde->name) &&
 386                     hinfo->family == family) {
 387                         atomic_inc(&hinfo->use);
 388                         spin_unlock_bh(&hashlimit_lock);
 389                         return hinfo;
 390                 }
 391         }
 392         spin_unlock_bh(&hashlimit_lock);
 393         return NULL;
 394 }
 395
 396 static void htable_put(struct xt_hashlimit_htable *hinfo)
 397 {
 398         if (atomic_dec_and_test(&hinfo->use)) {
 399                 spin_lock_bh(&hashlimit_lock);
 400                 hlist_del(&hinfo->node);
 401                 spin_unlock_bh(&hashlimit_lock);
 402                 htable_destroy(hinfo);
 403         }
 404 }
 405
 406 /* The algorithm used is the Simple Token Bucket Filter (TBF)
 407  * see net/sched/sch_tbf.c in the linux source tree
 408  */
 409
 410 /* Rusty: This is my (non-mathematically-inclined) understanding of
 411    this algorithm.  The `average rate' in jiffies becomes your initial
 412    amount of credit `credit' and the most credit you can ever have
 413    `credit_cap'.  The `peak rate' becomes the cost of passing the
 414    test, `cost'.
 415
 416    `prev' tracks the last packet hit: you gain one credit per jiffy.
 417    If you get credit balance more than this, the extra credit is
 418    discarded.  Every time the match passes, you lose `cost' credits;
 419    if you don't have that many, the test fails.
 420
 421    See Alexey's formal explanation in net/sched/sch_tbf.c.
 422
 423    To get the maximum range, we multiply by this factor (ie. you get N
 424    credits per jiffy).  We want to allow a rate as low as 1 per day
 425    (slowest userspace tool allows), which means
 426    CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
 427 */
 428 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
 429
 430 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
 431  * us the power of 2 below the theoretical max, so GCC simply does a
 432  * shift. */
 433 #define _POW2_BELOW2(x) ((x)|((x)>>1))
 434 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
 435 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
 436 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
 437 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
 438 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
 439
 440 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
 441
 442 /* Precision saver. */
 443 static inline u_int32_t
 444 user2credits(u_int32_t user)
 445 {
 446         /* If multiplying would overflow... */
 447         if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
 448                 /* Divide first. */
 449                 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
 450
 451         return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
 452 }
 453
 454 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
 455 {
 456         dh->rateinfo.credit += (now - dh->rateinfo.prev) * CREDITS_PER_JIFFY;
 457         if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
 458                 dh->rateinfo.credit = dh->rateinfo.credit_cap;
 459         dh->rateinfo.prev = now;
 460 }
 461
 462 static inline __be32 maskl(__be32 a, unsigned int l)
 463 {
 464         return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
 465 }
 466
 467 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 468 static 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         case 96 ... 127:
 483                 i[3] = maskl(i[3], p - 96);
 484                 break;
 485         case 128:
 486                 break;
 487         }
 488 }
 489 #endif
 490
 491 static int
 492 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
 493                    struct dsthash_dst *dst,
 494                    const struct sk_buff *skb, unsigned int protoff)
 495 {
 496         __be16 _ports[2], *ports;
 497         u8 nexthdr;
 498
 499         memset(dst, 0, sizeof(*dst));
 500
 501         switch (hinfo->family) {
 502         case NFPROTO_IPV4:
 503                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
 504                         dst->ip.dst = maskl(ip_hdr(skb)->daddr,
 505                                       hinfo->cfg.dstmask);
 506                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
 507                         dst->ip.src = maskl(ip_hdr(skb)->saddr,
 508                                       hinfo->cfg.srcmask);
 509
 510                 if (!(hinfo->cfg.mode &
 511                       (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 512                         return 0;
 513                 nexthdr = ip_hdr(skb)->protocol;
 514                 break;
 515 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 516         case NFPROTO_IPV6:
 517                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
 518                         memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
 519                                sizeof(dst->ip6.dst));
 520                         hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
 521                 }
 522                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
 523                         memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
 524                                sizeof(dst->ip6.src));
 525                         hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
 526                 }
 527
 528                 if (!(hinfo->cfg.mode &
 529                       (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 530                         return 0;
 531                 nexthdr = ipv6_hdr(skb)->nexthdr;
 532                 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
 533                 if ((int)protoff < 0)
 534                         return -1;
 535                 break;
 536 #endif
 537         default:
 538                 BUG();
 539                 return 0;
 540         }
 541
 542         switch (nexthdr) {
 543         case IPPROTO_TCP:
 544         case IPPROTO_UDP:
 545         case IPPROTO_UDPLITE:
 546         case IPPROTO_SCTP:
 547         case IPPROTO_DCCP:
 548                 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
 549                                            &_ports);
 550                 break;
 551         default:
 552                 _ports[0] = _ports[1] = 0;
 553                 ports = _ports;
 554                 break;
 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
 565 static bool
 566 hashlimit_mt_v0(const struct sk_buff *skb, const struct xt_match_param *par)
 567 {
 568         const struct xt_hashlimit_info *r = par->matchinfo;
 569         struct xt_hashlimit_htable *hinfo = r->hinfo;
 570         unsigned long now = jiffies;
 571         struct dsthash_ent *dh;
 572         struct dsthash_dst dst;
 573
 574         if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
 575                 goto hotdrop;
 576
 577         spin_lock_bh(&hinfo->lock);
 578         dh = dsthash_find(hinfo, &dst);
 579         if (!dh) {
 580                 dh = dsthash_alloc_init(hinfo, &dst);
 581                 if (!dh) {
 582                         spin_unlock_bh(&hinfo->lock);
 583                         goto hotdrop;
 584                 }
 585
 586                 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
 587                 dh->rateinfo.prev = jiffies;
 588                 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
 589                                                    hinfo->cfg.burst);
 590                 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
 591                                                        hinfo->cfg.burst);
 592                 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
 593         } else {
 594                 /* update expiration timeout */
 595                 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 596                 rateinfo_recalc(dh, now);
 597         }
 598
 599         if (dh->rateinfo.credit >= dh->rateinfo.cost) {
 600                 /* We're underlimit. */
 601                 dh->rateinfo.credit -= dh->rateinfo.cost;
 602                 spin_unlock_bh(&hinfo->lock);
 603                 return true;
 604         }
 605
 606         spin_unlock_bh(&hinfo->lock);
 607
 608         /* default case: we're overlimit, thus don't match */
 609         return false;
 610
 611 hotdrop:
 612         *par->hotdrop = true;
 613         return false;
 614 }
 615
 616 static bool
 617 hashlimit_mt(const struct sk_buff *skb, const struct xt_match_param *par)
 618 {
 619         const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 620         struct xt_hashlimit_htable *hinfo = info->hinfo;
 621         unsigned long now = jiffies;
 622         struct dsthash_ent *dh;
 623         struct dsthash_dst dst;
 624
 625         if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
 626                 goto hotdrop;
 627
 628         spin_lock_bh(&hinfo->lock);
 629         dh = dsthash_find(hinfo, &dst);
 630         if (dh == NULL) {
 631                 dh = dsthash_alloc_init(hinfo, &dst);
 632                 if (dh == NULL) {
 633                         spin_unlock_bh(&hinfo->lock);
 634                         goto hotdrop;
 635                 }
 636
 637                 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
 638                 dh->rateinfo.prev = jiffies;
 639                 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
 640                                       hinfo->cfg.burst);
 641                 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
 642                                           hinfo->cfg.burst);
 643                 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
 644         } else {
 645                 /* update expiration timeout */
 646                 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 647                 rateinfo_recalc(dh, now);
 648         }
 649
 650         if (dh->rateinfo.credit >= dh->rateinfo.cost) {
 651                 /* below the limit */
 652                 dh->rateinfo.credit -= dh->rateinfo.cost;
 653                 spin_unlock_bh(&hinfo->lock);
 654                 return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
 655         }
 656
 657         spin_unlock_bh(&hinfo->lock);
 658         /* default match is underlimit - so over the limit, we need to invert */
 659         return info->cfg.mode & XT_HASHLIMIT_INVERT;
 660
 661  hotdrop:
 662         *par->hotdrop = true;
 663         return false;
 664 }
 665
 666 static bool hashlimit_mt_check_v0(const struct xt_mtchk_param *par)
 667 {
 668         struct xt_hashlimit_info *r = par->matchinfo;
 669
 670         /* Check for overflow. */
 671         if (r->cfg.burst == 0 ||
 672             user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
 673                 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
 674                        r->cfg.avg, r->cfg.burst);
 675                 return false;
 676         }
 677         if (r->cfg.mode == 0 ||
 678             r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
 679                            XT_HASHLIMIT_HASH_DIP |
 680                            XT_HASHLIMIT_HASH_SIP |
 681                            XT_HASHLIMIT_HASH_SPT))
 682                 return false;
 683         if (!r->cfg.gc_interval)
 684                 return false;
 685         if (!r->cfg.expire)
 686                 return false;
 687         if (r->name[sizeof(r->name) - 1] != '\0')
 688                 return false;
 689
 690         /* This is the best we've got: We cannot release and re-grab lock,
 691          * since checkentry() is called before x_tables.c grabs xt_mutex.
 692          * We also cannot grab the hashtable spinlock, since htable_create will
 693          * call vmalloc, and that can sleep.  And we cannot just re-search
 694          * the list of htable's in htable_create(), since then we would
 695          * create duplicate proc files. -HW */
 696         mutex_lock(&hlimit_mutex);
 697         r->hinfo = htable_find_get(r->name, par->match->family);
 698         if (!r->hinfo && htable_create_v0(r, par->match->family) != 0) {
 699                 mutex_unlock(&hlimit_mutex);
 700                 return false;
 701         }
 702         mutex_unlock(&hlimit_mutex);
 703
 704         return true;
 705 }
 706
 707 static bool hashlimit_mt_check(const struct xt_mtchk_param *par)
 708 {
 709         struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 710
 711         /* Check for overflow. */
 712         if (info->cfg.burst == 0 ||
 713             user2credits(info->cfg.avg * info->cfg.burst) <
 714             user2credits(info->cfg.avg)) {
 715                 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
 716                        info->cfg.avg, info->cfg.burst);
 717                 return false;
 718         }
 719         if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
 720                 return false;
 721         if (info->name[sizeof(info->name)-1] != '\0')
 722                 return false;
 723         if (par->match->family == NFPROTO_IPV4) {
 724                 if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
 725                         return false;
 726         } else {
 727                 if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
 728                         return false;
 729         }
 730
 731         /* This is the best we've got: We cannot release and re-grab lock,
 732          * since checkentry() is called before x_tables.c grabs xt_mutex.
 733          * We also cannot grab the hashtable spinlock, since htable_create will
 734          * call vmalloc, and that can sleep.  And we cannot just re-search
 735          * the list of htable's in htable_create(), since then we would
 736          * create duplicate proc files. -HW */
 737         mutex_lock(&hlimit_mutex);
 738         info->hinfo = htable_find_get(info->name, par->match->family);
 739         if (!info->hinfo && htable_create(info, par->match->family) != 0) {
 740                 mutex_unlock(&hlimit_mutex);
 741                 return false;
 742         }
 743         mutex_unlock(&hlimit_mutex);
 744         return true;
 745 }
 746
 747 static void
 748 hashlimit_mt_destroy_v0(const struct xt_mtdtor_param *par)
 749 {
 750         const struct xt_hashlimit_info *r = par->matchinfo;
 751
 752         htable_put(r->hinfo);
 753 }
 754
 755 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
 756 {
 757         const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 758
 759         htable_put(info->hinfo);
 760 }
 761
 762 #ifdef CONFIG_COMPAT
 763 struct compat_xt_hashlimit_info {
 764         char name[IFNAMSIZ];
 765         struct hashlimit_cfg cfg;
 766         compat_uptr_t hinfo;
 767         compat_uptr_t master;
 768 };
 769
 770 static void hashlimit_mt_compat_from_user(void *dst, void *src)
 771 {
 772         int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
 773
 774         memcpy(dst, src, off);
 775         memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
 776 }
 777
 778 static int hashlimit_mt_compat_to_user(void __user *dst, void *src)
 779 {
 780         int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
 781
 782         return copy_to_user(dst, src, off) ? -EFAULT : 0;
 783 }
 784 #endif
 785
 786 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
 787         {
 788                 .name           = "hashlimit",
 789                 .revision       = 0,
 790                 .family         = NFPROTO_IPV4,
 791                 .match          = hashlimit_mt_v0,
 792                 .matchsize      = sizeof(struct xt_hashlimit_info),
 793 #ifdef CONFIG_COMPAT
 794                 .compatsize     = sizeof(struct compat_xt_hashlimit_info),
 795                 .compat_from_user = hashlimit_mt_compat_from_user,
 796                 .compat_to_user = hashlimit_mt_compat_to_user,
 797 #endif
 798                 .checkentry     = hashlimit_mt_check_v0,
 799                 .destroy        = hashlimit_mt_destroy_v0,
 800                 .me             = THIS_MODULE
 801         },
 802         {
 803                 .name           = "hashlimit",
 804                 .revision       = 1,
 805                 .family         = NFPROTO_IPV4,
 806                 .match          = hashlimit_mt,
 807                 .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 808                 .checkentry     = hashlimit_mt_check,
 809                 .destroy        = hashlimit_mt_destroy,
 810                 .me             = THIS_MODULE,
 811         },
 812 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 813         {
 814                 .name           = "hashlimit",
 815                 .family         = NFPROTO_IPV6,
 816                 .match          = hashlimit_mt_v0,
 817                 .matchsize      = sizeof(struct xt_hashlimit_info),
 818 #ifdef CONFIG_COMPAT
 819                 .compatsize     = sizeof(struct compat_xt_hashlimit_info),
 820                 .compat_from_user = hashlimit_mt_compat_from_user,
 821                 .compat_to_user = hashlimit_mt_compat_to_user,
 822 #endif
 823                 .checkentry     = hashlimit_mt_check_v0,
 824                 .destroy        = hashlimit_mt_destroy_v0,
 825                 .me             = THIS_MODULE
 826         },
 827         {
 828                 .name           = "hashlimit",
 829                 .revision       = 1,
 830                 .family         = NFPROTO_IPV6,
 831                 .match          = hashlimit_mt,
 832                 .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 833                 .checkentry     = hashlimit_mt_check,
 834                 .destroy        = hashlimit_mt_destroy,
 835                 .me             = THIS_MODULE,
 836         },
 837 #endif
 838 };
 839
 840 /* PROC stuff */
 841 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
 842         __acquires(htable->lock)
 843 {
 844         struct proc_dir_entry *pde = s->private;
 845         struct xt_hashlimit_htable *htable = pde->data;
 846         unsigned int *bucket;
 847
 848         spin_lock_bh(&htable->lock);
 849         if (*pos >= htable->cfg.size)
 850                 return NULL;
 851
 852         bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
 853         if (!bucket)
 854                 return ERR_PTR(-ENOMEM);
 855
 856         *bucket = *pos;
 857         return bucket;
 858 }
 859
 860 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
 861 {
 862         struct proc_dir_entry *pde = s->private;
 863         struct xt_hashlimit_htable *htable = pde->data;
 864         unsigned int *bucket = (unsigned int *)v;
 865
 866         *pos = ++(*bucket);
 867         if (*pos >= htable->cfg.size) {
 868                 kfree(v);
 869                 return NULL;
 870         }
 871         return bucket;
 872 }
 873
 874 static void dl_seq_stop(struct seq_file *s, void *v)
 875         __releases(htable->lock)
 876 {
 877         struct proc_dir_entry *pde = s->private;
 878         struct xt_hashlimit_htable *htable = pde->data;
 879         unsigned int *bucket = (unsigned int *)v;
 880
 881         kfree(bucket);
 882         spin_unlock_bh(&htable->lock);
 883 }
 884
 885 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
 886                                    struct seq_file *s)
 887 {
 888         /* recalculate to show accurate numbers */
 889         rateinfo_recalc(ent, jiffies);
 890
 891         switch (family) {
 892         case NFPROTO_IPV4:
 893                 return seq_printf(s, "%ld %pI4:%u->%pI4:%u %u %u %u\n",
 894                                  (long)(ent->expires - jiffies)/HZ,
 895                                  &ent->dst.ip.src,
 896                                  ntohs(ent->dst.src_port),
 897                                  &ent->dst.ip.dst,
 898                                  ntohs(ent->dst.dst_port),
 899                                  ent->rateinfo.credit, ent->rateinfo.credit_cap,
 900                                  ent->rateinfo.cost);
 901 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 902         case NFPROTO_IPV6:
 903                 return seq_printf(s, "%ld %pI6:%u->%pI6:%u %u %u %u\n",
 904                                  (long)(ent->expires - jiffies)/HZ,
 905                                  &ent->dst.ip6.src,
 906                                  ntohs(ent->dst.src_port),
 907                                  &ent->dst.ip6.dst,
 908                                  ntohs(ent->dst.dst_port),
 909                                  ent->rateinfo.credit, ent->rateinfo.credit_cap,
 910                                  ent->rateinfo.cost);
 911 #endif
 912         default:
 913                 BUG();
 914                 return 0;
 915         }
 916 }
 917
 918 static int dl_seq_show(struct seq_file *s, void *v)
 919 {
 920         struct proc_dir_entry *pde = s->private;
 921         struct xt_hashlimit_htable *htable = pde->data;
 922         unsigned int *bucket = (unsigned int *)v;
 923         struct dsthash_ent *ent;
 924         struct hlist_node *pos;
 925
 926         if (!hlist_empty(&htable->hash[*bucket])) {
 927                 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
 928                         if (dl_seq_real_show(ent, htable->family, s))
 929                                 return -1;
 930         }
 931         return 0;
 932 }
 933
 934 static const struct seq_operations dl_seq_ops = {
 935         .start = dl_seq_start,
 936         .next  = dl_seq_next,
 937         .stop  = dl_seq_stop,
 938         .show  = dl_seq_show
 939 };
 940
 941 static int dl_proc_open(struct inode *inode, struct file *file)
 942 {
 943         int ret = seq_open(file, &dl_seq_ops);
 944
 945         if (!ret) {
 946                 struct seq_file *sf = file->private_data;
 947                 sf->private = PDE(inode);
 948         }
 949         return ret;
 950 }
 951
 952 static const struct file_operations dl_file_ops = {
 953         .owner   = THIS_MODULE,
 954         .open    = dl_proc_open,
 955         .read    = seq_read,
 956         .llseek  = seq_lseek,
 957         .release = seq_release
 958 };
 959
 960 static int __init hashlimit_mt_init(void)
 961 {
 962         int err;
 963
 964         err = xt_register_matches(hashlimit_mt_reg,
 965               ARRAY_SIZE(hashlimit_mt_reg));
 966         if (err < 0)
 967                 goto err1;
 968
 969         err = -ENOMEM;
 970         hashlimit_cachep = kmem_cache_create("xt_hashlimit",
 971                                             sizeof(struct dsthash_ent), 0, 0,
 972                                             NULL);
 973         if (!hashlimit_cachep) {
 974                 printk(KERN_ERR "xt_hashlimit: unable to create slab cache\n");
 975                 goto err2;
 976         }
 977         hashlimit_procdir4 = proc_mkdir("ipt_hashlimit", init_net.proc_net);
 978         if (!hashlimit_procdir4) {
 979                 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
 980                                 "entry\n");
 981                 goto err3;
 982         }
 983         err = 0;
 984 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 985         hashlimit_procdir6 = proc_mkdir("ip6t_hashlimit", init_net.proc_net);
 986         if (!hashlimit_procdir6) {
 987                 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
 988                                 "entry\n");
 989                 err = -ENOMEM;
 990         }
 991 #endif
 992         if (!err)
 993                 return 0;
 994         remove_proc_entry("ipt_hashlimit", init_net.proc_net);
 995 err3:
 996         kmem_cache_destroy(hashlimit_cachep);
 997 err2:
 998         xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
 999 err1:
1000         return err;
1001
1002 }
1003
1004 static void __exit hashlimit_mt_exit(void)
1005 {
1006         remove_proc_entry("ipt_hashlimit", init_net.proc_net);
1007 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
1008         remove_proc_entry("ip6t_hashlimit", init_net.proc_net);
1009 #endif
1010         kmem_cache_destroy(hashlimit_cachep);
1011         xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1012 }
1013
1014 module_init(hashlimit_mt_init);
1015 module_exit(hashlimit_mt_exit);