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>
24 #include <net/net_namespace.h>
26 #include <linux/netfilter/x_tables.h>
27 #include <linux/netfilter_ipv4/ip_tables.h>
28 #include <linux/netfilter_ipv6/ip6_tables.h>
29 #include <linux/netfilter/xt_hashlimit.h>
30 #include <linux/mutex.h>
32 MODULE_LICENSE("GPL");
33 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
34 MODULE_DESCRIPTION("iptables match for limiting per hash-bucket");
35 MODULE_ALIAS("ipt_hashlimit");
36 MODULE_ALIAS("ip6t_hashlimit");
38 /* need to declare this at the top */
39 static struct proc_dir_entry *hashlimit_procdir4;
40 static struct proc_dir_entry *hashlimit_procdir6;
41 static const struct file_operations dl_file_ops;
60 /* static / read-only parts in the beginning */
61 struct hlist_node node;
62 struct dsthash_dst dst;
64 /* modified structure members in the end */
65 unsigned long expires; /* precalculated expiry time */
67 unsigned long prev; /* last modification */
69 u_int32_t credit_cap, cost;
73 struct xt_hashlimit_htable {
74 struct hlist_node node; /* global list of all htables */
78 struct hashlimit_cfg cfg; /* config */
81 spinlock_t lock; /* lock for list_head */
82 u_int32_t rnd; /* random seed for hash */
84 unsigned int count; /* number entries in table */
85 struct timer_list timer; /* timer for gc */
88 struct proc_dir_entry *pde;
90 struct hlist_head hash[0]; /* hashtable itself */
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 struct kmem_cache *hashlimit_cachep __read_mostly;
98 static inline bool dst_cmp(const struct dsthash_ent *ent,
99 const struct dsthash_dst *b)
101 return !memcmp(&ent->dst, b, sizeof(ent->dst));
105 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
107 return jhash(dst, sizeof(*dst), ht->rnd) % ht->cfg.size;
110 static struct dsthash_ent *
111 dsthash_find(const struct xt_hashlimit_htable *ht,
112 const struct dsthash_dst *dst)
114 struct dsthash_ent *ent;
115 struct hlist_node *pos;
116 u_int32_t hash = hash_dst(ht, dst);
118 if (!hlist_empty(&ht->hash[hash])) {
119 hlist_for_each_entry(ent, pos, &ht->hash[hash], node)
120 if (dst_cmp(ent, dst))
126 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
127 static struct dsthash_ent *
128 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
129 const struct dsthash_dst *dst)
131 struct dsthash_ent *ent;
133 /* initialize hash with random val at the time we allocate
134 * the first hashtable entry */
135 if (!ht->rnd_initialized) {
136 get_random_bytes(&ht->rnd, 4);
137 ht->rnd_initialized = 1;
140 if (ht->cfg.max && ht->count >= ht->cfg.max) {
141 /* FIXME: do something. question is what.. */
144 "xt_hashlimit: max count of %u reached\n",
149 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
153 "xt_hashlimit: can't allocate dsthash_ent\n");
156 memcpy(&ent->dst, dst, sizeof(ent->dst));
158 hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
164 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
166 hlist_del(&ent->node);
167 kmem_cache_free(hashlimit_cachep, ent);
170 static void htable_gc(unsigned long htlong);
172 static int htable_create(struct xt_hashlimit_info *minfo, int family)
174 struct xt_hashlimit_htable *hinfo;
179 size = minfo->cfg.size;
181 size = ((num_physpages << PAGE_SHIFT) / 16384) /
182 sizeof(struct list_head);
183 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
188 /* FIXME: don't use vmalloc() here or anywhere else -HW */
189 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
190 sizeof(struct list_head) * size);
192 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
195 minfo->hinfo = hinfo;
197 /* copy match config into hashtable config */
198 memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
199 hinfo->cfg.size = size;
201 hinfo->cfg.max = 8 * hinfo->cfg.size;
202 else if (hinfo->cfg.max < hinfo->cfg.size)
203 hinfo->cfg.max = hinfo->cfg.size;
205 for (i = 0; i < hinfo->cfg.size; i++)
206 INIT_HLIST_HEAD(&hinfo->hash[i]);
208 atomic_set(&hinfo->use, 1);
210 hinfo->family = family;
211 hinfo->rnd_initialized = 0;
212 spin_lock_init(&hinfo->lock);
213 hinfo->pde = create_proc_entry(minfo->name, 0,
214 family == AF_INET ? hashlimit_procdir4 :
220 hinfo->pde->proc_fops = &dl_file_ops;
221 hinfo->pde->data = hinfo;
223 setup_timer(&hinfo->timer, htable_gc, (unsigned long )hinfo);
224 hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
225 add_timer(&hinfo->timer);
227 spin_lock_bh(&hashlimit_lock);
228 hlist_add_head(&hinfo->node, &hashlimit_htables);
229 spin_unlock_bh(&hashlimit_lock);
234 static bool select_all(const struct xt_hashlimit_htable *ht,
235 const struct dsthash_ent *he)
240 static bool select_gc(const struct xt_hashlimit_htable *ht,
241 const struct dsthash_ent *he)
243 return jiffies >= he->expires;
246 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
247 bool (*select)(const struct xt_hashlimit_htable *ht,
248 const struct dsthash_ent *he))
252 /* lock hash table and iterate over it */
253 spin_lock_bh(&ht->lock);
254 for (i = 0; i < ht->cfg.size; i++) {
255 struct dsthash_ent *dh;
256 struct hlist_node *pos, *n;
257 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
258 if ((*select)(ht, dh))
259 dsthash_free(ht, dh);
262 spin_unlock_bh(&ht->lock);
265 /* hash table garbage collector, run by timer */
266 static void htable_gc(unsigned long htlong)
268 struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
270 htable_selective_cleanup(ht, select_gc);
272 /* re-add the timer accordingly */
273 ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
274 add_timer(&ht->timer);
277 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
279 /* remove timer, if it is pending */
280 if (timer_pending(&hinfo->timer))
281 del_timer(&hinfo->timer);
283 /* remove proc entry */
284 remove_proc_entry(hinfo->pde->name,
285 hinfo->family == AF_INET ? hashlimit_procdir4 :
287 htable_selective_cleanup(hinfo, select_all);
291 static struct xt_hashlimit_htable *htable_find_get(const char *name,
294 struct xt_hashlimit_htable *hinfo;
295 struct hlist_node *pos;
297 spin_lock_bh(&hashlimit_lock);
298 hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) {
299 if (!strcmp(name, hinfo->pde->name) &&
300 hinfo->family == family) {
301 atomic_inc(&hinfo->use);
302 spin_unlock_bh(&hashlimit_lock);
306 spin_unlock_bh(&hashlimit_lock);
310 static void htable_put(struct xt_hashlimit_htable *hinfo)
312 if (atomic_dec_and_test(&hinfo->use)) {
313 spin_lock_bh(&hashlimit_lock);
314 hlist_del(&hinfo->node);
315 spin_unlock_bh(&hashlimit_lock);
316 htable_destroy(hinfo);
320 /* The algorithm used is the Simple Token Bucket Filter (TBF)
321 * see net/sched/sch_tbf.c in the linux source tree
324 /* Rusty: This is my (non-mathematically-inclined) understanding of
325 this algorithm. The `average rate' in jiffies becomes your initial
326 amount of credit `credit' and the most credit you can ever have
327 `credit_cap'. The `peak rate' becomes the cost of passing the
330 `prev' tracks the last packet hit: you gain one credit per jiffy.
331 If you get credit balance more than this, the extra credit is
332 discarded. Every time the match passes, you lose `cost' credits;
333 if you don't have that many, the test fails.
335 See Alexey's formal explanation in net/sched/sch_tbf.c.
337 To get the maximum range, we multiply by this factor (ie. you get N
338 credits per jiffy). We want to allow a rate as low as 1 per day
339 (slowest userspace tool allows), which means
340 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
342 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
344 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
345 * us the power of 2 below the theoretical max, so GCC simply does a
347 #define _POW2_BELOW2(x) ((x)|((x)>>1))
348 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
349 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
350 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
351 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
352 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
354 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
356 /* Precision saver. */
357 static inline u_int32_t
358 user2credits(u_int32_t user)
360 /* If multiplying would overflow... */
361 if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
363 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
365 return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
368 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
370 dh->rateinfo.credit += (now - dh->rateinfo.prev) * CREDITS_PER_JIFFY;
371 if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
372 dh->rateinfo.credit = dh->rateinfo.credit_cap;
373 dh->rateinfo.prev = now;
377 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
378 struct dsthash_dst *dst,
379 const struct sk_buff *skb, unsigned int protoff)
381 __be16 _ports[2], *ports;
384 memset(dst, 0, sizeof(*dst));
386 switch (hinfo->family) {
388 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
389 dst->addr.ip.dst = ip_hdr(skb)->daddr;
390 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
391 dst->addr.ip.src = ip_hdr(skb)->saddr;
393 if (!(hinfo->cfg.mode &
394 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
396 nexthdr = ip_hdr(skb)->protocol;
398 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
400 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
401 memcpy(&dst->addr.ip6.dst, &ipv6_hdr(skb)->daddr,
402 sizeof(dst->addr.ip6.dst));
403 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
404 memcpy(&dst->addr.ip6.src, &ipv6_hdr(skb)->saddr,
405 sizeof(dst->addr.ip6.src));
407 if (!(hinfo->cfg.mode &
408 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
410 nexthdr = ipv6_find_hdr(skb, &protoff, -1, NULL);
423 case IPPROTO_UDPLITE:
426 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
430 _ports[0] = _ports[1] = 0;
436 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
437 dst->src_port = ports[0];
438 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
439 dst->dst_port = ports[1];
444 hashlimit_match(const struct sk_buff *skb,
445 const struct net_device *in,
446 const struct net_device *out,
447 const struct xt_match *match,
448 const void *matchinfo,
450 unsigned int protoff,
453 const struct xt_hashlimit_info *r =
454 ((const struct xt_hashlimit_info *)matchinfo)->u.master;
455 struct xt_hashlimit_htable *hinfo = r->hinfo;
456 unsigned long now = jiffies;
457 struct dsthash_ent *dh;
458 struct dsthash_dst dst;
460 if (hashlimit_init_dst(hinfo, &dst, skb, protoff) < 0)
463 spin_lock_bh(&hinfo->lock);
464 dh = dsthash_find(hinfo, &dst);
466 dh = dsthash_alloc_init(hinfo, &dst);
468 spin_unlock_bh(&hinfo->lock);
472 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
473 dh->rateinfo.prev = jiffies;
474 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
476 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
478 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
480 /* update expiration timeout */
481 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
482 rateinfo_recalc(dh, now);
485 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
486 /* We're underlimit. */
487 dh->rateinfo.credit -= dh->rateinfo.cost;
488 spin_unlock_bh(&hinfo->lock);
492 spin_unlock_bh(&hinfo->lock);
494 /* default case: we're overlimit, thus don't match */
503 hashlimit_checkentry(const char *tablename,
505 const struct xt_match *match,
507 unsigned int hook_mask)
509 struct xt_hashlimit_info *r = matchinfo;
511 /* Check for overflow. */
512 if (r->cfg.burst == 0 ||
513 user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
514 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
515 r->cfg.avg, r->cfg.burst);
518 if (r->cfg.mode == 0 ||
519 r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
520 XT_HASHLIMIT_HASH_DIP |
521 XT_HASHLIMIT_HASH_SIP |
522 XT_HASHLIMIT_HASH_SPT))
524 if (!r->cfg.gc_interval)
528 if (r->name[sizeof(r->name) - 1] != '\0')
531 /* This is the best we've got: We cannot release and re-grab lock,
532 * since checkentry() is called before x_tables.c grabs xt_mutex.
533 * We also cannot grab the hashtable spinlock, since htable_create will
534 * call vmalloc, and that can sleep. And we cannot just re-search
535 * the list of htable's in htable_create(), since then we would
536 * create duplicate proc files. -HW */
537 mutex_lock(&hlimit_mutex);
538 r->hinfo = htable_find_get(r->name, match->family);
539 if (!r->hinfo && htable_create(r, match->family) != 0) {
540 mutex_unlock(&hlimit_mutex);
543 mutex_unlock(&hlimit_mutex);
545 /* Ugly hack: For SMP, we only want to use one set */
551 hashlimit_destroy(const struct xt_match *match, void *matchinfo)
553 const struct xt_hashlimit_info *r = matchinfo;
555 htable_put(r->hinfo);
559 struct compat_xt_hashlimit_info {
561 struct hashlimit_cfg cfg;
563 compat_uptr_t master;
566 static void compat_from_user(void *dst, void *src)
568 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
570 memcpy(dst, src, off);
571 memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
574 static int compat_to_user(void __user *dst, void *src)
576 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
578 return copy_to_user(dst, src, off) ? -EFAULT : 0;
582 static struct xt_match xt_hashlimit[] __read_mostly = {
586 .match = hashlimit_match,
587 .matchsize = sizeof(struct xt_hashlimit_info),
589 .compatsize = sizeof(struct compat_xt_hashlimit_info),
590 .compat_from_user = compat_from_user,
591 .compat_to_user = compat_to_user,
593 .checkentry = hashlimit_checkentry,
594 .destroy = hashlimit_destroy,
600 .match = hashlimit_match,
601 .matchsize = sizeof(struct xt_hashlimit_info),
603 .compatsize = sizeof(struct compat_xt_hashlimit_info),
604 .compat_from_user = compat_from_user,
605 .compat_to_user = compat_to_user,
607 .checkentry = hashlimit_checkentry,
608 .destroy = hashlimit_destroy,
614 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
616 struct proc_dir_entry *pde = s->private;
617 struct xt_hashlimit_htable *htable = pde->data;
618 unsigned int *bucket;
620 spin_lock_bh(&htable->lock);
621 if (*pos >= htable->cfg.size)
624 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
626 return ERR_PTR(-ENOMEM);
632 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
634 struct proc_dir_entry *pde = s->private;
635 struct xt_hashlimit_htable *htable = pde->data;
636 unsigned int *bucket = (unsigned int *)v;
639 if (*pos >= htable->cfg.size) {
646 static void dl_seq_stop(struct seq_file *s, void *v)
648 struct proc_dir_entry *pde = s->private;
649 struct xt_hashlimit_htable *htable = pde->data;
650 unsigned int *bucket = (unsigned int *)v;
653 spin_unlock_bh(&htable->lock);
656 static int dl_seq_real_show(struct dsthash_ent *ent, int family,
659 /* recalculate to show accurate numbers */
660 rateinfo_recalc(ent, jiffies);
664 return seq_printf(s, "%ld %u.%u.%u.%u:%u->"
665 "%u.%u.%u.%u:%u %u %u %u\n",
666 (long)(ent->expires - jiffies)/HZ,
667 NIPQUAD(ent->dst.addr.ip.src),
668 ntohs(ent->dst.src_port),
669 NIPQUAD(ent->dst.addr.ip.dst),
670 ntohs(ent->dst.dst_port),
671 ent->rateinfo.credit, ent->rateinfo.credit_cap,
674 return seq_printf(s, "%ld " NIP6_FMT ":%u->"
675 NIP6_FMT ":%u %u %u %u\n",
676 (long)(ent->expires - jiffies)/HZ,
677 NIP6(*(struct in6_addr *)&ent->dst.addr.ip6.src),
678 ntohs(ent->dst.src_port),
679 NIP6(*(struct in6_addr *)&ent->dst.addr.ip6.dst),
680 ntohs(ent->dst.dst_port),
681 ent->rateinfo.credit, ent->rateinfo.credit_cap,
689 static int dl_seq_show(struct seq_file *s, void *v)
691 struct proc_dir_entry *pde = s->private;
692 struct xt_hashlimit_htable *htable = pde->data;
693 unsigned int *bucket = (unsigned int *)v;
694 struct dsthash_ent *ent;
695 struct hlist_node *pos;
697 if (!hlist_empty(&htable->hash[*bucket])) {
698 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
699 if (dl_seq_real_show(ent, htable->family, s))
705 static const struct seq_operations dl_seq_ops = {
706 .start = dl_seq_start,
712 static int dl_proc_open(struct inode *inode, struct file *file)
714 int ret = seq_open(file, &dl_seq_ops);
717 struct seq_file *sf = file->private_data;
718 sf->private = PDE(inode);
723 static const struct file_operations dl_file_ops = {
724 .owner = THIS_MODULE,
725 .open = dl_proc_open,
728 .release = seq_release
731 static int __init xt_hashlimit_init(void)
735 err = xt_register_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
740 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
741 sizeof(struct dsthash_ent), 0, 0,
743 if (!hashlimit_cachep) {
744 printk(KERN_ERR "xt_hashlimit: unable to create slab cache\n");
747 hashlimit_procdir4 = proc_mkdir("ipt_hashlimit", init_net.proc_net);
748 if (!hashlimit_procdir4) {
749 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
753 hashlimit_procdir6 = proc_mkdir("ip6t_hashlimit", init_net.proc_net);
754 if (!hashlimit_procdir6) {
755 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
761 remove_proc_entry("ipt_hashlimit", init_net.proc_net);
763 kmem_cache_destroy(hashlimit_cachep);
765 xt_unregister_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
771 static void __exit xt_hashlimit_fini(void)
773 remove_proc_entry("ipt_hashlimit", init_net.proc_net);
774 remove_proc_entry("ip6t_hashlimit", init_net.proc_net);
775 kmem_cache_destroy(hashlimit_cachep);
776 xt_unregister_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
779 module_init(xt_hashlimit_init);
780 module_exit(xt_hashlimit_fini);