]> git.neil.brown.name Git - history.git/commitdiff
[NETFILTER]: Add iptables hashlimit match
authorHarald Welte <laforge@netfilter.org>
Wed, 20 Oct 2004 14:26:24 +0000 (07:26 -0700)
committerDavid S. Miller <davem@nuts.davemloft.net>
Wed, 20 Oct 2004 14:26:24 +0000 (07:26 -0700)
It was previously known as dstlimit, but has now
been generalized one step further.

Signed-off-by: Harald Welte <laforge@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/netfilter_ipv4/ipt_hashlimit.h [new file with mode: 0644]
net/ipv4/netfilter/Kconfig
net/ipv4/netfilter/Makefile
net/ipv4/netfilter/ipt_hashlimit.c [new file with mode: 0644]

diff --git a/include/linux/netfilter_ipv4/ipt_hashlimit.h b/include/linux/netfilter_ipv4/ipt_hashlimit.h
new file mode 100644 (file)
index 0000000..ac2cb64
--- /dev/null
@@ -0,0 +1,40 @@
+#ifndef _IPT_HASHLIMIT_H
+#define _IPT_HASHLIMIT_H
+
+/* timings are in milliseconds. */
+#define IPT_HASHLIMIT_SCALE 10000
+/* 1/10,000 sec period => max of 10,000/sec.  Min rate is then 429490
+   seconds, or one every 59 hours. */
+
+/* details of this structure hidden by the implementation */
+struct ipt_hashlimit_htable;
+
+#define IPT_HASHLIMIT_HASH_DIP 0x0001
+#define IPT_HASHLIMIT_HASH_DPT 0x0002
+#define IPT_HASHLIMIT_HASH_SIP 0x0004
+#define IPT_HASHLIMIT_HASH_SPT 0x0008
+
+struct hashlimit_cfg {
+       u_int32_t mode;   /* bitmask of IPT_HASHLIMIT_HASH_* */
+       u_int32_t avg;    /* Average secs between packets * scale */
+       u_int32_t burst;  /* Period multiplier for upper limit. */
+
+       /* user specified */
+       u_int32_t size;         /* how many buckets */
+       u_int32_t max;          /* max number of entries */
+       u_int32_t gc_interval;  /* gc interval */
+       u_int32_t expire;       /* when do entries expire? */
+};
+
+struct ipt_hashlimit_info {
+       char name [IFNAMSIZ];           /* name */
+       struct hashlimit_cfg cfg;
+       struct ipt_hashlimit_htable *hinfo;
+
+       /* Used internally by the kernel */
+       union {
+               void *ptr;
+               struct ipt_hashlimit_info *master;
+       } u;
+};
+#endif /*_IPT_HASHLIMIT_H*/
index 1d9fc5722adf2a78c20b51c58a1f4c6516febe52..947f17f44f6fe6fab9e1816f5399b95e1dfc7143 100644 (file)
@@ -361,6 +361,20 @@ config IP_NF_MATCH_CONNMARK
          Documentation/modules.txt.  The module will be called
          ipt_connmark.o.  If unsure, say `N'.
 
+config IP_NF_MATCH_HASHLIMIT
+       tristate  'hashlimit match support'
+       depends on IP_NF_IPTABLES
+       help
+         This option adds a new iptables `hashlimit' match.  
+
+         As opposed to `limit', this match dynamically crates a hash table
+         of limit buckets, based on your selection of source/destination
+         ip addresses and/or ports.
+
+         It enables you to express policies like `10kpps for any given
+         destination IP' or `500pps from any given source IP'  with a single
+         IPtables rule.
+
 # `filter', generic and specific targets
 config IP_NF_FILTER
        tristate "Packet filtering"
index c21f1b44b3a723126ab3a12106f0ad0c8791d519..ace5f9e3bcd897b49a36853c0502e2423d001f2a 100644 (file)
@@ -46,6 +46,7 @@ obj-$(CONFIG_IP_NF_RAW) += iptable_raw.o
 # matches
 obj-$(CONFIG_IP_NF_MATCH_HELPER) += ipt_helper.o
 obj-$(CONFIG_IP_NF_MATCH_LIMIT) += ipt_limit.o
+obj-$(CONFIG_IP_NF_MATCH_HASHLIMIT) += ipt_hashlimit.o
 obj-$(CONFIG_IP_NF_MATCH_SCTP) += ipt_sctp.o
 obj-$(CONFIG_IP_NF_MATCH_MARK) += ipt_mark.o
 obj-$(CONFIG_IP_NF_MATCH_MAC) += ipt_mac.o
diff --git a/net/ipv4/netfilter/ipt_hashlimit.c b/net/ipv4/netfilter/ipt_hashlimit.c
new file mode 100644 (file)
index 0000000..ae68244
--- /dev/null
@@ -0,0 +1,716 @@
+/* iptables match extension to limit the number of packets per second
+ * seperately for each hashbucket (sourceip/sourceport/dstip/dstport)
+ *
+ * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
+ *
+ * $Id: ipt_hashlimit.c 3244 2004-10-20 16:24:29Z laforge@netfilter.org $
+ *
+ * Development of this code was funded by Astaro AG, http://www.astaro.com/
+ *
+ * based on ipt_limit.c by:
+ * Jérôme de Vivie     <devivie@info.enserb.u-bordeaux.fr>
+ * Hervé Eychenne      <eychenne@info.enserb.u-bordeaux.fr>
+ * Rusty Russell       <rusty@rustcorp.com.au>
+ *
+ * The general idea is to create a hash table for every dstip and have a
+ * seperate limit counter per tuple.  This way you can do something like 'limit
+ * the number of syn packets for each of my internal addresses.
+ *
+ * Ideally this would just be implemented as a general 'hash' match, which would
+ * allow us to attach any iptables target to it's hash buckets.  But this is
+ * not possible in the current iptables architecture.  As always, pkttables for
+ * 2.7.x will help ;)
+ */
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/spinlock.h>
+#include <linux/random.h>
+#include <linux/jhash.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/tcp.h>
+#include <linux/udp.h>
+#include <linux/sctp.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+
+#define ASSERT_READ_LOCK(x) 
+#define ASSERT_WRITE_LOCK(x) 
+#include <linux/netfilter_ipv4/lockhelp.h>
+#include <linux/netfilter_ipv4/listhelp.h>
+
+#include <linux/netfilter_ipv4/ip_tables.h>
+#include <linux/netfilter_ipv4/ipt_hashlimit.h>
+
+/* FIXME: this is just for IP_NF_ASSERRT */
+#include <linux/netfilter_ipv4/ip_conntrack.h>
+
+#define MS2JIFFIES(x) ((x*HZ)/1000)
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
+MODULE_DESCRIPTION("iptables match for limiting per hash-bucket");
+
+/* need to declare this at the top */
+static struct proc_dir_entry *hashlimit_procdir;
+static struct file_operations dl_file_ops;
+
+/* hash table crap */
+
+struct dsthash_dst {
+       u_int32_t src_ip;
+       u_int32_t dst_ip;
+       /* ports have to be consecutive !!! */
+       u_int16_t src_port;
+       u_int16_t dst_port;
+};
+
+struct dsthash_ent {
+       /* static / read-only parts in the beginning */
+       struct list_head list;
+       struct dsthash_dst dst;
+
+       /* modified structure members in the end */
+       unsigned long expires;          /* precalculated expiry time */
+       struct {
+               unsigned long prev;     /* last modification */
+               u_int32_t credit;
+               u_int32_t credit_cap, cost;
+       } rateinfo;
+};
+
+struct ipt_hashlimit_htable {
+       struct list_head list;          /* global list of all htables */
+       atomic_t use;
+
+       struct hashlimit_cfg cfg;       /* config */
+
+       /* used internally */
+       spinlock_t lock;                /* lock for list_head */
+       u_int32_t rnd;                  /* random seed for hash */
+       struct timer_list timer;        /* timer for gc */
+       atomic_t count;                 /* number entries in table */
+
+       /* seq_file stuff */
+       struct proc_dir_entry *pde;
+
+       struct list_head hash[0];       /* hashtable itself */
+};
+
+DECLARE_RWLOCK(hashlimit_lock);                /* protects htables list */
+static LIST_HEAD(hashlimit_htables);
+static kmem_cache_t *hashlimit_cachep;
+
+static inline int dst_cmp(const struct dsthash_ent *ent, struct dsthash_dst *b)
+{
+       return (ent->dst.dst_ip == b->dst_ip 
+               && ent->dst.dst_port == b->dst_port
+               && ent->dst.src_port == b->src_port
+               && ent->dst.src_ip == b->src_ip);
+}
+
+static inline u_int32_t
+hash_dst(const struct ipt_hashlimit_htable *ht, const struct dsthash_dst *dst)
+{
+       return (jhash_3words(dst->dst_ip, (dst->dst_port<<16 & dst->src_port), 
+                            dst->src_ip, ht->rnd) % ht->cfg.size);
+}
+
+static inline struct dsthash_ent *
+__dsthash_find(const struct ipt_hashlimit_htable *ht, struct dsthash_dst *dst)
+{
+       struct dsthash_ent *ent;
+       u_int32_t hash = hash_dst(ht, dst);
+       MUST_BE_LOCKED(&ht->lock);
+       ent = LIST_FIND(&ht->hash[hash], dst_cmp, struct dsthash_ent *, dst);
+       return ent;
+}
+
+/* allocate dsthash_ent, initialize dst, put in htable and lock it */
+static struct dsthash_ent *
+__dsthash_alloc_init(struct ipt_hashlimit_htable *ht, struct dsthash_dst *dst)
+{
+       struct dsthash_ent *ent;
+
+       /* initialize hash with random val at the time we allocate
+        * the first hashtable entry */
+       if (!ht->rnd)
+               get_random_bytes(&ht->rnd, 4);
+
+       if (ht->cfg.max &&
+           atomic_read(&ht->count) >= ht->cfg.max) {
+               /* FIXME: do something. question is what.. */
+               if (net_ratelimit())
+                       printk(KERN_WARNING 
+                               "ipt_hashlimit: max count of %u reached\n", 
+                               ht->cfg.max);
+               return NULL;
+       }
+
+       ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
+       if (!ent) {
+               if (net_ratelimit())
+                       printk(KERN_ERR 
+                               "ipt_hashlimit: can't allocate dsthash_ent\n");
+               return NULL;
+       }
+
+       atomic_inc(&ht->count);
+
+       ent->dst.dst_ip = dst->dst_ip;
+       ent->dst.dst_port = dst->dst_port;
+       ent->dst.src_ip = dst->src_ip;
+       ent->dst.src_port = dst->src_port;
+
+       list_add(&ent->list, &ht->hash[hash_dst(ht, dst)]);
+
+       return ent;
+}
+
+static inline void 
+__dsthash_free(struct ipt_hashlimit_htable *ht, struct dsthash_ent *ent)
+{
+       MUST_BE_LOCKED(&ht->lock);
+
+       list_del(&ent->list);
+       kmem_cache_free(hashlimit_cachep, ent);
+       atomic_dec(&ht->count);
+}
+static void htable_gc(unsigned long htlong);
+
+static int htable_create(struct ipt_hashlimit_info *minfo)
+{
+       int i;
+       unsigned int size;
+       struct ipt_hashlimit_htable *hinfo;
+
+       if (minfo->cfg.size)
+               size = minfo->cfg.size;
+       else {
+               size = (((num_physpages << PAGE_SHIFT) / 16384)
+                        / sizeof(struct list_head));
+               if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
+                       size = 8192;
+               if (size < 16)
+                       size = 16;
+       }
+       /* FIXME: don't use vmalloc() here or anywhere else -HW */
+       hinfo = vmalloc(sizeof(struct ipt_hashlimit_htable)
+                       + (sizeof(struct list_head) * size));
+       if (!hinfo) {
+               printk(KERN_ERR "ipt_hashlimit: Unable to create hashtable\n");
+               return -1;
+       }
+       minfo->hinfo = hinfo;
+
+       /* copy match config into hashtable config */
+       memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
+       hinfo->cfg.size = size;
+       if (!hinfo->cfg.max)
+               hinfo->cfg.max = 8 * hinfo->cfg.size;
+       else if (hinfo->cfg.max < hinfo->cfg.size)
+               hinfo->cfg.max = hinfo->cfg.size;
+
+       for (i = 0; i < hinfo->cfg.size; i++)
+               INIT_LIST_HEAD(&hinfo->hash[i]);
+
+       atomic_set(&hinfo->count, 0);
+       atomic_set(&hinfo->use, 1);
+       hinfo->rnd = 0;
+       hinfo->lock = SPIN_LOCK_UNLOCKED;
+       hinfo->pde = create_proc_entry(minfo->name, 0, hashlimit_procdir);
+       if (!hinfo->pde) {
+               vfree(hinfo);
+               return -1;
+       }
+       hinfo->pde->proc_fops = &dl_file_ops;
+       hinfo->pde->data = hinfo;
+
+       init_timer(&hinfo->timer);
+       hinfo->timer.expires = jiffies + MS2JIFFIES(hinfo->cfg.gc_interval);
+       hinfo->timer.data = (unsigned long )hinfo;
+       hinfo->timer.function = htable_gc;
+       add_timer(&hinfo->timer);
+
+       WRITE_LOCK(&hashlimit_lock);
+       list_add(&hinfo->list, &hashlimit_htables);
+       WRITE_UNLOCK(&hashlimit_lock);
+
+       return 0;
+}
+
+static int select_all(struct ipt_hashlimit_htable *ht, struct dsthash_ent *he)
+{
+       return 1;
+}
+
+static int select_gc(struct ipt_hashlimit_htable *ht, struct dsthash_ent *he)
+{
+       return (jiffies >= he->expires);
+}
+
+static void htable_selective_cleanup(struct ipt_hashlimit_htable *ht,
+                               int (*select)(struct ipt_hashlimit_htable *ht, 
+                                             struct dsthash_ent *he))
+{
+       int i;
+
+       IP_NF_ASSERT(ht->cfg.size && ht->cfg.max);
+
+       /* lock hash table and iterate over it */
+       LOCK_BH(&ht->lock);
+       for (i = 0; i < ht->cfg.size; i++) {
+               struct dsthash_ent *dh, *n;
+               list_for_each_entry_safe(dh, n, &ht->hash[i], list) {
+                       if ((*select)(ht, dh))
+                               __dsthash_free(ht, dh);
+               }
+       }
+       UNLOCK_BH(&ht->lock);
+}
+
+/* hash table garbage collector, run by timer */
+static void htable_gc(unsigned long htlong)
+{
+       struct ipt_hashlimit_htable *ht = (struct ipt_hashlimit_htable *)htlong;
+
+       htable_selective_cleanup(ht, select_gc);
+
+       /* re-add the timer accordingly */
+       ht->timer.expires = jiffies + MS2JIFFIES(ht->cfg.gc_interval);
+       add_timer(&ht->timer);
+}
+
+static void htable_destroy(struct ipt_hashlimit_htable *hinfo)
+{
+       /* remove timer, if it is pending */
+       if (timer_pending(&hinfo->timer))
+               del_timer(&hinfo->timer);
+
+       /* remove proc entry */
+       remove_proc_entry(hinfo->pde->name, hashlimit_procdir);
+
+       htable_selective_cleanup(hinfo, select_all);
+       vfree(hinfo);
+}
+
+static struct ipt_hashlimit_htable *htable_find_get(char *name)
+{
+       struct ipt_hashlimit_htable *hinfo;
+
+       READ_LOCK(&hashlimit_lock);
+       list_for_each_entry(hinfo, &hashlimit_htables, list) {
+               if (!strcmp(name, hinfo->pde->name)) {
+                       atomic_inc(&hinfo->use);
+                       READ_UNLOCK(&hashlimit_lock);
+                       return hinfo;
+               }
+       }
+       READ_UNLOCK(&hashlimit_lock);
+
+       return NULL;
+}
+
+static void htable_put(struct ipt_hashlimit_htable *hinfo)
+{
+       if (atomic_dec_and_test(&hinfo->use)) {
+               WRITE_LOCK(&hashlimit_lock);
+               list_del(&hinfo->list);
+               WRITE_UNLOCK(&hashlimit_lock);
+               htable_destroy(hinfo);
+       }
+}
+
+
+/* The algorithm used is the Simple Token Bucket Filter (TBF)
+ * see net/sched/sch_tbf.c in the linux source tree
+ */
+
+/* Rusty: This is my (non-mathematically-inclined) understanding of
+   this algorithm.  The `average rate' in jiffies becomes your initial
+   amount of credit `credit' and the most credit you can ever have
+   `credit_cap'.  The `peak rate' becomes the cost of passing the
+   test, `cost'.
+
+   `prev' tracks the last packet hit: you gain one credit per jiffy.
+   If you get credit balance more than this, the extra credit is
+   discarded.  Every time the match passes, you lose `cost' credits;
+   if you don't have that many, the test fails.
+
+   See Alexey's formal explanation in net/sched/sch_tbf.c.
+
+   To get the maximum range, we multiply by this factor (ie. you get N
+   credits per jiffy).  We want to allow a rate as low as 1 per day
+   (slowest userspace tool allows), which means
+   CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
+*/
+#define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
+
+/* Repeated shift and or gives us all 1s, final shift and add 1 gives
+ * us the power of 2 below the theoretical max, so GCC simply does a
+ * shift. */
+#define _POW2_BELOW2(x) ((x)|((x)>>1))
+#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
+#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
+#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
+#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
+#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
+
+#define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
+
+/* Precision saver. */
+static inline u_int32_t
+user2credits(u_int32_t user)
+{
+       /* If multiplying would overflow... */
+       if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
+               /* Divide first. */
+               return (user / IPT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
+
+       return (user * HZ * CREDITS_PER_JIFFY) / IPT_HASHLIMIT_SCALE;
+}
+
+static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
+{
+       dh->rateinfo.credit += (now - xchg(&dh->rateinfo.prev, now)) 
+                                       * CREDITS_PER_JIFFY;
+       if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
+               dh->rateinfo.credit = dh->rateinfo.credit_cap;
+}
+
+static inline int get_ports(const struct sk_buff *skb, int offset, 
+                           u16 ports[2])
+{
+       union {
+               struct tcphdr th;
+               struct udphdr uh;
+               sctp_sctphdr_t sctph;
+       } hdr_u, *ptr_u;
+
+       /* Must not be a fragment. */
+       if (offset)
+               return 1;
+
+       /* Must be big enough to read ports (both UDP and TCP have
+          them at the start). */
+       ptr_u = skb_header_pointer(skb, skb->nh.iph->ihl*4, 8, &hdr_u); 
+       if (!ptr_u)
+               return 1;
+
+       switch (skb->nh.iph->protocol) {
+               case IPPROTO_TCP:
+                       ports[0] = ptr_u->th.source;
+                       ports[1] = ptr_u->th.dest;
+                       break;
+               case IPPROTO_UDP:
+                       ports[0] = ptr_u->uh.source;
+                       ports[1] = ptr_u->uh.dest;
+                       break;
+               case IPPROTO_SCTP:
+                       ports[0] = ptr_u->sctph.source;
+                       ports[1] = ptr_u->sctph.dest;
+                       break;
+               default:
+                       /* all other protocols don't supprot per-port hash
+                        * buckets */
+                       ports[0] = ports[1] = 0;
+                       break;
+       }
+
+       return 0;
+}
+
+
+static int
+hashlimit_match(const struct sk_buff *skb,
+               const struct net_device *in,
+               const struct net_device *out,
+               const void *matchinfo,
+               int offset,
+               int *hotdrop)
+{
+       struct ipt_hashlimit_info *r = 
+               ((struct ipt_hashlimit_info *)matchinfo)->u.master;
+       struct ipt_hashlimit_htable *hinfo = r->hinfo;
+       unsigned long now = jiffies;
+       struct dsthash_ent *dh;
+       struct dsthash_dst dst;
+
+       /* build 'dst' according to hinfo->cfg and current packet */
+       memset(&dst, 0, sizeof(dst));
+       if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DIP)
+               dst.dst_ip = skb->nh.iph->daddr;
+       if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SIP)
+               dst.src_ip = skb->nh.iph->saddr;
+       if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DPT
+           ||hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SPT) {
+               u_int16_t ports[2];
+               if (get_ports(skb, offset, ports)) {
+                       /* We've been asked to examine this packet, and we
+                         can't.  Hence, no choice but to drop. */
+                       *hotdrop = 1;
+                       return 0;
+               }
+               if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SPT)
+                       dst.src_port = ports[0];
+               if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DPT)
+                       dst.dst_port = ports[1];
+       } 
+
+       LOCK_BH(&hinfo->lock);
+       dh = __dsthash_find(hinfo, &dst);
+       if (!dh) {
+               dh = __dsthash_alloc_init(hinfo, &dst);
+
+               if (!dh) {
+                       /* enomem... don't match == DROP */
+                       if (net_ratelimit())
+                               printk(KERN_ERR "%s: ENOMEM\n", __FUNCTION__);
+                       UNLOCK_BH(&hinfo->lock);
+                       return 0;
+               }
+
+               dh->expires = jiffies + MS2JIFFIES(hinfo->cfg.expire);
+
+               dh->rateinfo.prev = jiffies;
+               dh->rateinfo.credit = user2credits(hinfo->cfg.avg * 
+                                                       hinfo->cfg.burst);
+               dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg * 
+                                                       hinfo->cfg.burst);
+               dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
+
+               UNLOCK_BH(&hinfo->lock);
+               return 1;
+       }
+
+       /* update expiration timeout */
+       dh->expires = now + MS2JIFFIES(hinfo->cfg.expire);
+
+       rateinfo_recalc(dh, now);
+       if (dh->rateinfo.credit >= dh->rateinfo.cost) {
+               /* We're underlimit. */
+               dh->rateinfo.credit -= dh->rateinfo.cost;
+               UNLOCK_BH(&hinfo->lock);
+               return 1;
+       }
+
+               UNLOCK_BH(&hinfo->lock);
+
+       /* default case: we're overlimit, thus don't match */
+       return 0;
+}
+
+static int
+hashlimit_checkentry(const char *tablename,
+                    const struct ipt_ip *ip,
+                    void *matchinfo,
+                    unsigned int matchsize,
+                    unsigned int hook_mask)
+{
+       struct ipt_hashlimit_info *r = matchinfo;
+
+       if (matchsize != IPT_ALIGN(sizeof(struct ipt_hashlimit_info)))
+               return 0;
+
+       /* Check for overflow. */
+       if (r->cfg.burst == 0
+           || user2credits(r->cfg.avg * r->cfg.burst) < 
+                                       user2credits(r->cfg.avg)) {
+               printk(KERN_ERR "ipt_hashlimit: Overflow, try lower: %u/%u\n",
+                      r->cfg.avg, r->cfg.burst);
+               return 0;
+       }
+
+       if (r->cfg.mode == 0 
+           || r->cfg.mode > (IPT_HASHLIMIT_HASH_DPT
+                         |IPT_HASHLIMIT_HASH_DIP
+                         |IPT_HASHLIMIT_HASH_SIP
+                         |IPT_HASHLIMIT_HASH_SPT))
+               return 0;
+
+       if (!r->cfg.gc_interval)
+               return 0;
+       
+       if (!r->cfg.expire)
+               return 0;
+
+       r->hinfo = htable_find_get(r->name);
+       if (!r->hinfo && (htable_create(r) != 0)) {
+               return 0;
+       }
+
+       /* Ugly hack: For SMP, we only want to use one set */
+       r->u.master = r;
+
+       return 1;
+}
+
+static void
+hashlimit_destroy(void *matchinfo, unsigned int matchsize)
+{
+       struct ipt_hashlimit_info *r = (struct ipt_hashlimit_info *) matchinfo;
+
+       htable_put(r->hinfo);
+}
+
+static struct ipt_match ipt_hashlimit = { 
+       .name = "hashlimit", 
+       .match = hashlimit_match, 
+       .checkentry = hashlimit_checkentry, 
+       .destroy = hashlimit_destroy,
+       .me = THIS_MODULE 
+};
+
+/* PROC stuff */
+
+static void *dl_seq_start(struct seq_file *s, loff_t *pos)
+{
+       struct proc_dir_entry *pde = s->private;
+       struct ipt_hashlimit_htable *htable = pde->data;
+       unsigned int *bucket;
+
+       LOCK_BH(&htable->lock);
+       if (*pos >= htable->cfg.size)
+               return NULL;
+
+       bucket = kmalloc(sizeof(unsigned int), GFP_KERNEL);
+       if (!bucket)
+               return ERR_PTR(-ENOMEM);
+
+       *bucket = *pos;
+       return bucket;
+}
+
+static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
+{
+       struct proc_dir_entry *pde = s->private;
+       struct ipt_hashlimit_htable *htable = pde->data;
+       unsigned int *bucket = (unsigned int *)v;
+
+       *pos = ++(*bucket);
+       if (*pos >= htable->cfg.size) {
+               kfree(v);
+               return NULL;
+       }
+       return bucket;
+}
+
+static void dl_seq_stop(struct seq_file *s, void *v)
+{
+       struct proc_dir_entry *pde = s->private;
+       struct ipt_hashlimit_htable *htable = pde->data;
+       unsigned int *bucket = (unsigned int *)v;
+
+       kfree(bucket);
+
+       UNLOCK_BH(&htable->lock);
+}
+
+static inline int dl_seq_real_show(struct dsthash_ent *ent, struct seq_file *s)
+{
+       /* recalculate to show accurate numbers */
+       rateinfo_recalc(ent, jiffies);
+
+       return seq_printf(s, "%ld %u.%u.%u.%u:%u->%u.%u.%u.%u:%u %u %u %u\n",
+                       (ent->expires - jiffies)/HZ,
+                       NIPQUAD(ent->dst.src_ip), ntohs(ent->dst.src_port),
+                       NIPQUAD(ent->dst.dst_ip), ntohs(ent->dst.dst_port),
+                       ent->rateinfo.credit, ent->rateinfo.credit_cap,
+                       ent->rateinfo.cost);
+}
+
+static int dl_seq_show(struct seq_file *s, void *v)
+{
+       struct proc_dir_entry *pde = s->private;
+       struct ipt_hashlimit_htable *htable = pde->data;
+       unsigned int *bucket = (unsigned int *)v;
+
+       if (LIST_FIND_W(&htable->hash[*bucket], dl_seq_real_show,
+                     struct dsthash_ent *, s)) {
+               /* buffer was filled and unable to print that tuple */
+               return 1;
+       }
+       return 0;
+}
+
+static struct seq_operations dl_seq_ops = {
+       .start = dl_seq_start,
+       .next  = dl_seq_next,
+       .stop  = dl_seq_stop,
+       .show  = dl_seq_show
+};
+
+static int dl_proc_open(struct inode *inode, struct file *file)
+{
+       int ret = seq_open(file, &dl_seq_ops);
+
+       if (!ret) {
+               struct seq_file *sf = file->private_data;
+               sf->private = PDE(inode);
+       }
+       return ret;
+}
+
+static struct file_operations dl_file_ops = {
+       .owner   = THIS_MODULE,
+       .open    = dl_proc_open,
+       .read    = seq_read,
+       .llseek  = seq_lseek,
+       .release = seq_release
+};
+
+static int init_or_fini(int fini)
+{
+       int ret = 0;
+
+       if (fini)
+               goto cleanup;
+
+       if (ipt_register_match(&ipt_hashlimit)) {
+               ret = -EINVAL;
+               goto cleanup_nothing;
+       }
+
+       /* FIXME: do we really want HWCACHE_ALIGN since our objects are
+        * quite small ? */
+       hashlimit_cachep = kmem_cache_create("ipt_hashlimit",
+                                           sizeof(struct dsthash_ent), 0,
+                                           SLAB_HWCACHE_ALIGN, NULL, NULL);
+       if (!hashlimit_cachep) {
+               printk(KERN_ERR "Unable to create ipt_hashlimit slab cache\n");
+               ret = -ENOMEM;
+               goto cleanup_unreg_match;
+       }
+
+       hashlimit_procdir = proc_mkdir("ipt_hashlimit", proc_net);
+       if (!hashlimit_procdir) {
+               printk(KERN_ERR "Unable to create proc dir entry\n");
+               ret = -ENOMEM;
+               goto cleanup_free_slab;
+       }
+
+       return ret;
+
+cleanup:
+       remove_proc_entry("ipt_hashlimit", proc_net);
+cleanup_free_slab:
+       kmem_cache_destroy(hashlimit_cachep);
+cleanup_unreg_match:
+       ipt_unregister_match(&ipt_hashlimit);
+cleanup_nothing:
+       return ret;
+       
+}
+
+static int __init init(void)
+{
+       return init_or_fini(0);
+}
+
+static void __exit fini(void)
+{
+       init_or_fini(1);
+}
+
+module_init(init);
+module_exit(fini);