1 /* iptables match extension to limit the number of packets per second
2 * seperately for each hashbucket (sourceip/sourceport/dstip/dstport)
4 * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
6 * $Id: ipt_hashlimit.c 3244 2004-10-20 16:24:29Z laforge@netfilter.org $
8 * Development of this code was funded by Astaro AG, http://www.astaro.com/
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>
23 #include <linux/ipv6.h>
25 #include <linux/netfilter/x_tables.h>
26 #include <linux/netfilter_ipv4/ip_tables.h>
27 #include <linux/netfilter_ipv6/ip6_tables.h>
28 #include <linux/netfilter/xt_hashlimit.h>
29 #include <linux/mutex.h>
31 MODULE_LICENSE("GPL");
32 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
33 MODULE_DESCRIPTION("iptables match for limiting per hash-bucket");
34 MODULE_ALIAS("ipt_hashlimit");
35 MODULE_ALIAS("ip6t_hashlimit");
37 /* need to declare this at the top */
38 static struct proc_dir_entry *hashlimit_procdir4;
39 static struct proc_dir_entry *hashlimit_procdir6;
40 static const struct file_operations dl_file_ops;
59 /* static / read-only parts in the beginning */
60 struct hlist_node node;
61 struct dsthash_dst dst;
63 /* modified structure members in the end */
64 unsigned long expires; /* precalculated expiry time */
66 unsigned long prev; /* last modification */
68 u_int32_t credit_cap, cost;
72 struct xt_hashlimit_htable {
73 struct hlist_node node; /* global list of all htables */
77 struct hashlimit_cfg cfg; /* config */
80 spinlock_t lock; /* lock for list_head */
81 u_int32_t rnd; /* random seed for hash */
83 unsigned int count; /* number entries in table */
84 struct timer_list timer; /* timer for gc */
87 struct proc_dir_entry *pde;
89 struct hlist_head hash[0]; /* hashtable itself */
92 static DEFINE_SPINLOCK(hashlimit_lock); /* protects htables list */
93 static DEFINE_MUTEX(hlimit_mutex); /* additional checkentry protection */
94 static HLIST_HEAD(hashlimit_htables);
95 static struct kmem_cache *hashlimit_cachep __read_mostly;
97 static inline bool dst_cmp(const struct dsthash_ent *ent,
98 const struct dsthash_dst *b)
100 return !memcmp(&ent->dst, b, sizeof(ent->dst));
104 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
106 return jhash(dst, sizeof(*dst), ht->rnd) % ht->cfg.size;
109 static struct dsthash_ent *
110 dsthash_find(const struct xt_hashlimit_htable *ht,
111 const struct dsthash_dst *dst)
113 struct dsthash_ent *ent;
114 struct hlist_node *pos;
115 u_int32_t hash = hash_dst(ht, dst);
117 if (!hlist_empty(&ht->hash[hash])) {
118 hlist_for_each_entry(ent, pos, &ht->hash[hash], node)
119 if (dst_cmp(ent, dst))
125 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
126 static struct dsthash_ent *
127 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
128 const struct dsthash_dst *dst)
130 struct dsthash_ent *ent;
132 /* initialize hash with random val at the time we allocate
133 * the first hashtable entry */
134 if (!ht->rnd_initialized) {
135 get_random_bytes(&ht->rnd, 4);
136 ht->rnd_initialized = 1;
139 if (ht->cfg.max && ht->count >= ht->cfg.max) {
140 /* FIXME: do something. question is what.. */
143 "xt_hashlimit: max count of %u reached\n",
148 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
152 "xt_hashlimit: can't allocate dsthash_ent\n");
155 memcpy(&ent->dst, dst, sizeof(ent->dst));
157 hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
163 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
165 hlist_del(&ent->node);
166 kmem_cache_free(hashlimit_cachep, ent);
169 static void htable_gc(unsigned long htlong);
171 static int htable_create(struct xt_hashlimit_info *minfo, int family)
173 struct xt_hashlimit_htable *hinfo;
178 size = minfo->cfg.size;
180 size = ((num_physpages << PAGE_SHIFT) / 16384) /
181 sizeof(struct list_head);
182 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
187 /* FIXME: don't use vmalloc() here or anywhere else -HW */
188 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
189 sizeof(struct list_head) * size);
191 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
194 minfo->hinfo = hinfo;
196 /* copy match config into hashtable config */
197 memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
198 hinfo->cfg.size = size;
200 hinfo->cfg.max = 8 * hinfo->cfg.size;
201 else if (hinfo->cfg.max < hinfo->cfg.size)
202 hinfo->cfg.max = hinfo->cfg.size;
204 for (i = 0; i < hinfo->cfg.size; i++)
205 INIT_HLIST_HEAD(&hinfo->hash[i]);
207 atomic_set(&hinfo->use, 1);
209 hinfo->family = family;
210 hinfo->rnd_initialized = 0;
211 spin_lock_init(&hinfo->lock);
212 hinfo->pde = create_proc_entry(minfo->name, 0,
213 family == AF_INET ? hashlimit_procdir4 :
219 hinfo->pde->proc_fops = &dl_file_ops;
220 hinfo->pde->data = hinfo;
222 setup_timer(&hinfo->timer, htable_gc, (unsigned long )hinfo);
223 hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
224 add_timer(&hinfo->timer);
226 spin_lock_bh(&hashlimit_lock);
227 hlist_add_head(&hinfo->node, &hashlimit_htables);
228 spin_unlock_bh(&hashlimit_lock);
233 static bool select_all(const struct xt_hashlimit_htable *ht,
234 const struct dsthash_ent *he)
239 static bool select_gc(const struct xt_hashlimit_htable *ht,
240 const struct dsthash_ent *he)
242 return jiffies >= he->expires;
245 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
246 bool (*select)(const struct xt_hashlimit_htable *ht,
247 const struct dsthash_ent *he))
251 /* lock hash table and iterate over it */
252 spin_lock_bh(&ht->lock);
253 for (i = 0; i < ht->cfg.size; i++) {
254 struct dsthash_ent *dh;
255 struct hlist_node *pos, *n;
256 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
257 if ((*select)(ht, dh))
258 dsthash_free(ht, dh);
261 spin_unlock_bh(&ht->lock);
264 /* hash table garbage collector, run by timer */
265 static void htable_gc(unsigned long htlong)
267 struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
269 htable_selective_cleanup(ht, select_gc);
271 /* re-add the timer accordingly */
272 ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
273 add_timer(&ht->timer);
276 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
278 /* remove timer, if it is pending */
279 if (timer_pending(&hinfo->timer))
280 del_timer(&hinfo->timer);
282 /* remove proc entry */
283 remove_proc_entry(hinfo->pde->name,
284 hinfo->family == AF_INET ? hashlimit_procdir4 :
286 htable_selective_cleanup(hinfo, select_all);
290 static struct xt_hashlimit_htable *htable_find_get(const char *name,
293 struct xt_hashlimit_htable *hinfo;
294 struct hlist_node *pos;
296 spin_lock_bh(&hashlimit_lock);
297 hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) {
298 if (!strcmp(name, hinfo->pde->name) &&
299 hinfo->family == family) {
300 atomic_inc(&hinfo->use);
301 spin_unlock_bh(&hashlimit_lock);
305 spin_unlock_bh(&hashlimit_lock);
309 static void htable_put(struct xt_hashlimit_htable *hinfo)
311 if (atomic_dec_and_test(&hinfo->use)) {
312 spin_lock_bh(&hashlimit_lock);
313 hlist_del(&hinfo->node);
314 spin_unlock_bh(&hashlimit_lock);
315 htable_destroy(hinfo);
319 /* The algorithm used is the Simple Token Bucket Filter (TBF)
320 * see net/sched/sch_tbf.c in the linux source tree
323 /* Rusty: This is my (non-mathematically-inclined) understanding of
324 this algorithm. The `average rate' in jiffies becomes your initial
325 amount of credit `credit' and the most credit you can ever have
326 `credit_cap'. The `peak rate' becomes the cost of passing the
329 `prev' tracks the last packet hit: you gain one credit per jiffy.
330 If you get credit balance more than this, the extra credit is
331 discarded. Every time the match passes, you lose `cost' credits;
332 if you don't have that many, the test fails.
334 See Alexey's formal explanation in net/sched/sch_tbf.c.
336 To get the maximum range, we multiply by this factor (ie. you get N
337 credits per jiffy). We want to allow a rate as low as 1 per day
338 (slowest userspace tool allows), which means
339 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
341 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
343 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
344 * us the power of 2 below the theoretical max, so GCC simply does a
346 #define _POW2_BELOW2(x) ((x)|((x)>>1))
347 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
348 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
349 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
350 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
351 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
353 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
355 /* Precision saver. */
356 static inline u_int32_t
357 user2credits(u_int32_t user)
359 /* If multiplying would overflow... */
360 if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
362 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
364 return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
367 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
369 dh->rateinfo.credit += (now - dh->rateinfo.prev) * CREDITS_PER_JIFFY;
370 if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
371 dh->rateinfo.credit = dh->rateinfo.credit_cap;
372 dh->rateinfo.prev = now;
376 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
377 struct dsthash_dst *dst,
378 const struct sk_buff *skb, unsigned int protoff)
380 __be16 _ports[2], *ports;
383 memset(dst, 0, sizeof(*dst));
385 switch (hinfo->family) {
387 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
388 dst->addr.ip.dst = ip_hdr(skb)->daddr;
389 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
390 dst->addr.ip.src = ip_hdr(skb)->saddr;
392 if (!(hinfo->cfg.mode &
393 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
395 nexthdr = ip_hdr(skb)->protocol;
397 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
399 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
400 memcpy(&dst->addr.ip6.dst, &ipv6_hdr(skb)->daddr,
401 sizeof(dst->addr.ip6.dst));
402 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
403 memcpy(&dst->addr.ip6.src, &ipv6_hdr(skb)->saddr,
404 sizeof(dst->addr.ip6.src));
406 if (!(hinfo->cfg.mode &
407 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
409 nexthdr = ipv6_find_hdr(skb, &protoff, -1, NULL);
422 case IPPROTO_UDPLITE:
425 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
429 _ports[0] = _ports[1] = 0;
435 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
436 dst->src_port = ports[0];
437 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
438 dst->dst_port = ports[1];
443 hashlimit_match(const struct sk_buff *skb,
444 const struct net_device *in,
445 const struct net_device *out,
446 const struct xt_match *match,
447 const void *matchinfo,
449 unsigned int protoff,
452 const struct xt_hashlimit_info *r =
453 ((const struct xt_hashlimit_info *)matchinfo)->u.master;
454 struct xt_hashlimit_htable *hinfo = r->hinfo;
455 unsigned long now = jiffies;
456 struct dsthash_ent *dh;
457 struct dsthash_dst dst;
459 if (hashlimit_init_dst(hinfo, &dst, skb, protoff) < 0)
462 spin_lock_bh(&hinfo->lock);
463 dh = dsthash_find(hinfo, &dst);
465 dh = dsthash_alloc_init(hinfo, &dst);
467 spin_unlock_bh(&hinfo->lock);
471 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
472 dh->rateinfo.prev = jiffies;
473 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
475 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
477 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
479 /* update expiration timeout */
480 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
481 rateinfo_recalc(dh, now);
484 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
485 /* We're underlimit. */
486 dh->rateinfo.credit -= dh->rateinfo.cost;
487 spin_unlock_bh(&hinfo->lock);
491 spin_unlock_bh(&hinfo->lock);
493 /* default case: we're overlimit, thus don't match */
502 hashlimit_checkentry(const char *tablename,
504 const struct xt_match *match,
506 unsigned int hook_mask)
508 struct xt_hashlimit_info *r = matchinfo;
510 /* Check for overflow. */
511 if (r->cfg.burst == 0 ||
512 user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
513 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
514 r->cfg.avg, r->cfg.burst);
517 if (r->cfg.mode == 0 ||
518 r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
519 XT_HASHLIMIT_HASH_DIP |
520 XT_HASHLIMIT_HASH_SIP |
521 XT_HASHLIMIT_HASH_SPT))
523 if (!r->cfg.gc_interval)
527 if (r->name[sizeof(r->name) - 1] != '\0')
530 /* This is the best we've got: We cannot release and re-grab lock,
531 * since checkentry() is called before x_tables.c grabs xt_mutex.
532 * We also cannot grab the hashtable spinlock, since htable_create will
533 * call vmalloc, and that can sleep. And we cannot just re-search
534 * the list of htable's in htable_create(), since then we would
535 * create duplicate proc files. -HW */
536 mutex_lock(&hlimit_mutex);
537 r->hinfo = htable_find_get(r->name, match->family);
538 if (!r->hinfo && htable_create(r, match->family) != 0) {
539 mutex_unlock(&hlimit_mutex);
542 mutex_unlock(&hlimit_mutex);
544 /* Ugly hack: For SMP, we only want to use one set */
550 hashlimit_destroy(const struct xt_match *match, void *matchinfo)
552 const struct xt_hashlimit_info *r = matchinfo;
554 htable_put(r->hinfo);
558 struct compat_xt_hashlimit_info {
560 struct hashlimit_cfg cfg;
562 compat_uptr_t master;
565 static void compat_from_user(void *dst, void *src)
567 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
569 memcpy(dst, src, off);
570 memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
573 static int compat_to_user(void __user *dst, void *src)
575 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
577 return copy_to_user(dst, src, off) ? -EFAULT : 0;
581 static struct xt_match xt_hashlimit[] __read_mostly = {
585 .match = hashlimit_match,
586 .matchsize = sizeof(struct xt_hashlimit_info),
588 .compatsize = sizeof(struct compat_xt_hashlimit_info),
589 .compat_from_user = compat_from_user,
590 .compat_to_user = compat_to_user,
592 .checkentry = hashlimit_checkentry,
593 .destroy = hashlimit_destroy,
599 .match = hashlimit_match,
600 .matchsize = sizeof(struct xt_hashlimit_info),
602 .compatsize = sizeof(struct compat_xt_hashlimit_info),
603 .compat_from_user = compat_from_user,
604 .compat_to_user = compat_to_user,
606 .checkentry = hashlimit_checkentry,
607 .destroy = hashlimit_destroy,
613 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
615 struct proc_dir_entry *pde = s->private;
616 struct xt_hashlimit_htable *htable = pde->data;
617 unsigned int *bucket;
619 spin_lock_bh(&htable->lock);
620 if (*pos >= htable->cfg.size)
623 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
625 return ERR_PTR(-ENOMEM);
631 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
633 struct proc_dir_entry *pde = s->private;
634 struct xt_hashlimit_htable *htable = pde->data;
635 unsigned int *bucket = (unsigned int *)v;
638 if (*pos >= htable->cfg.size) {
645 static void dl_seq_stop(struct seq_file *s, void *v)
647 struct proc_dir_entry *pde = s->private;
648 struct xt_hashlimit_htable *htable = pde->data;
649 unsigned int *bucket = (unsigned int *)v;
652 spin_unlock_bh(&htable->lock);
655 static int dl_seq_real_show(struct dsthash_ent *ent, int family,
658 /* recalculate to show accurate numbers */
659 rateinfo_recalc(ent, jiffies);
663 return seq_printf(s, "%ld %u.%u.%u.%u:%u->"
664 "%u.%u.%u.%u:%u %u %u %u\n",
665 (long)(ent->expires - jiffies)/HZ,
666 NIPQUAD(ent->dst.addr.ip.src),
667 ntohs(ent->dst.src_port),
668 NIPQUAD(ent->dst.addr.ip.dst),
669 ntohs(ent->dst.dst_port),
670 ent->rateinfo.credit, ent->rateinfo.credit_cap,
673 return seq_printf(s, "%ld " NIP6_FMT ":%u->"
674 NIP6_FMT ":%u %u %u %u\n",
675 (long)(ent->expires - jiffies)/HZ,
676 NIP6(*(struct in6_addr *)&ent->dst.addr.ip6.src),
677 ntohs(ent->dst.src_port),
678 NIP6(*(struct in6_addr *)&ent->dst.addr.ip6.dst),
679 ntohs(ent->dst.dst_port),
680 ent->rateinfo.credit, ent->rateinfo.credit_cap,
688 static int dl_seq_show(struct seq_file *s, void *v)
690 struct proc_dir_entry *pde = s->private;
691 struct xt_hashlimit_htable *htable = pde->data;
692 unsigned int *bucket = (unsigned int *)v;
693 struct dsthash_ent *ent;
694 struct hlist_node *pos;
696 if (!hlist_empty(&htable->hash[*bucket])) {
697 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
698 if (dl_seq_real_show(ent, htable->family, s))
704 static const struct seq_operations dl_seq_ops = {
705 .start = dl_seq_start,
711 static int dl_proc_open(struct inode *inode, struct file *file)
713 int ret = seq_open(file, &dl_seq_ops);
716 struct seq_file *sf = file->private_data;
717 sf->private = PDE(inode);
722 static const struct file_operations dl_file_ops = {
723 .owner = THIS_MODULE,
724 .open = dl_proc_open,
727 .release = seq_release
730 static int __init xt_hashlimit_init(void)
734 err = xt_register_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
739 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
740 sizeof(struct dsthash_ent), 0, 0,
742 if (!hashlimit_cachep) {
743 printk(KERN_ERR "xt_hashlimit: unable to create slab cache\n");
746 hashlimit_procdir4 = proc_mkdir("ipt_hashlimit", proc_net);
747 if (!hashlimit_procdir4) {
748 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
752 hashlimit_procdir6 = proc_mkdir("ip6t_hashlimit", proc_net);
753 if (!hashlimit_procdir6) {
754 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
760 remove_proc_entry("ipt_hashlimit", proc_net);
762 kmem_cache_destroy(hashlimit_cachep);
764 xt_unregister_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
770 static void __exit xt_hashlimit_fini(void)
772 remove_proc_entry("ipt_hashlimit", proc_net);
773 remove_proc_entry("ip6t_hashlimit", proc_net);
774 kmem_cache_destroy(hashlimit_cachep);
775 xt_unregister_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
778 module_init(xt_hashlimit_init);
779 module_exit(xt_hashlimit_fini);