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 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, struct dsthash_dst *dst)
112 struct dsthash_ent *ent;
113 struct hlist_node *pos;
114 u_int32_t hash = hash_dst(ht, dst);
116 if (!hlist_empty(&ht->hash[hash])) {
117 hlist_for_each_entry(ent, pos, &ht->hash[hash], node)
118 if (dst_cmp(ent, dst))
124 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
125 static struct dsthash_ent *
126 dsthash_alloc_init(struct xt_hashlimit_htable *ht, struct dsthash_dst *dst)
128 struct dsthash_ent *ent;
130 /* initialize hash with random val at the time we allocate
131 * the first hashtable entry */
132 if (!ht->rnd_initialized) {
133 get_random_bytes(&ht->rnd, 4);
134 ht->rnd_initialized = 1;
137 if (ht->cfg.max && ht->count >= ht->cfg.max) {
138 /* FIXME: do something. question is what.. */
141 "xt_hashlimit: max count of %u reached\n",
146 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
150 "xt_hashlimit: can't allocate dsthash_ent\n");
153 memcpy(&ent->dst, dst, sizeof(ent->dst));
155 hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
161 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
163 hlist_del(&ent->node);
164 kmem_cache_free(hashlimit_cachep, ent);
167 static void htable_gc(unsigned long htlong);
169 static int htable_create(struct xt_hashlimit_info *minfo, int family)
171 struct xt_hashlimit_htable *hinfo;
176 size = minfo->cfg.size;
178 size = ((num_physpages << PAGE_SHIFT) / 16384) /
179 sizeof(struct list_head);
180 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
185 /* FIXME: don't use vmalloc() here or anywhere else -HW */
186 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
187 sizeof(struct list_head) * size);
189 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
192 minfo->hinfo = hinfo;
194 /* copy match config into hashtable config */
195 memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
196 hinfo->cfg.size = size;
198 hinfo->cfg.max = 8 * hinfo->cfg.size;
199 else if (hinfo->cfg.max < hinfo->cfg.size)
200 hinfo->cfg.max = hinfo->cfg.size;
202 for (i = 0; i < hinfo->cfg.size; i++)
203 INIT_HLIST_HEAD(&hinfo->hash[i]);
205 atomic_set(&hinfo->use, 1);
207 hinfo->family = family;
208 hinfo->rnd_initialized = 0;
209 spin_lock_init(&hinfo->lock);
210 hinfo->pde = create_proc_entry(minfo->name, 0,
211 family == AF_INET ? hashlimit_procdir4 :
217 hinfo->pde->proc_fops = &dl_file_ops;
218 hinfo->pde->data = hinfo;
220 setup_timer(&hinfo->timer, htable_gc, (unsigned long )hinfo);
221 hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
222 add_timer(&hinfo->timer);
224 spin_lock_bh(&hashlimit_lock);
225 hlist_add_head(&hinfo->node, &hashlimit_htables);
226 spin_unlock_bh(&hashlimit_lock);
231 static bool select_all(struct xt_hashlimit_htable *ht, struct dsthash_ent *he)
236 static bool select_gc(struct xt_hashlimit_htable *ht, struct dsthash_ent *he)
238 return (jiffies >= he->expires);
241 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
242 bool (*select)(struct xt_hashlimit_htable *ht,
243 struct dsthash_ent *he))
247 /* lock hash table and iterate over it */
248 spin_lock_bh(&ht->lock);
249 for (i = 0; i < ht->cfg.size; i++) {
250 struct dsthash_ent *dh;
251 struct hlist_node *pos, *n;
252 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
253 if ((*select)(ht, dh))
254 dsthash_free(ht, dh);
257 spin_unlock_bh(&ht->lock);
260 /* hash table garbage collector, run by timer */
261 static void htable_gc(unsigned long htlong)
263 struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
265 htable_selective_cleanup(ht, select_gc);
267 /* re-add the timer accordingly */
268 ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
269 add_timer(&ht->timer);
272 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
274 /* remove timer, if it is pending */
275 if (timer_pending(&hinfo->timer))
276 del_timer(&hinfo->timer);
278 /* remove proc entry */
279 remove_proc_entry(hinfo->pde->name,
280 hinfo->family == AF_INET ? hashlimit_procdir4 :
282 htable_selective_cleanup(hinfo, select_all);
286 static struct xt_hashlimit_htable *htable_find_get(char *name, int family)
288 struct xt_hashlimit_htable *hinfo;
289 struct hlist_node *pos;
291 spin_lock_bh(&hashlimit_lock);
292 hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) {
293 if (!strcmp(name, hinfo->pde->name) &&
294 hinfo->family == family) {
295 atomic_inc(&hinfo->use);
296 spin_unlock_bh(&hashlimit_lock);
300 spin_unlock_bh(&hashlimit_lock);
304 static void htable_put(struct xt_hashlimit_htable *hinfo)
306 if (atomic_dec_and_test(&hinfo->use)) {
307 spin_lock_bh(&hashlimit_lock);
308 hlist_del(&hinfo->node);
309 spin_unlock_bh(&hashlimit_lock);
310 htable_destroy(hinfo);
314 /* The algorithm used is the Simple Token Bucket Filter (TBF)
315 * see net/sched/sch_tbf.c in the linux source tree
318 /* Rusty: This is my (non-mathematically-inclined) understanding of
319 this algorithm. The `average rate' in jiffies becomes your initial
320 amount of credit `credit' and the most credit you can ever have
321 `credit_cap'. The `peak rate' becomes the cost of passing the
324 `prev' tracks the last packet hit: you gain one credit per jiffy.
325 If you get credit balance more than this, the extra credit is
326 discarded. Every time the match passes, you lose `cost' credits;
327 if you don't have that many, the test fails.
329 See Alexey's formal explanation in net/sched/sch_tbf.c.
331 To get the maximum range, we multiply by this factor (ie. you get N
332 credits per jiffy). We want to allow a rate as low as 1 per day
333 (slowest userspace tool allows), which means
334 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
336 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
338 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
339 * us the power of 2 below the theoretical max, so GCC simply does a
341 #define _POW2_BELOW2(x) ((x)|((x)>>1))
342 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
343 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
344 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
345 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
346 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
348 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
350 /* Precision saver. */
351 static inline u_int32_t
352 user2credits(u_int32_t user)
354 /* If multiplying would overflow... */
355 if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
357 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
359 return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
362 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
364 dh->rateinfo.credit += (now - dh->rateinfo.prev) * CREDITS_PER_JIFFY;
365 if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
366 dh->rateinfo.credit = dh->rateinfo.credit_cap;
367 dh->rateinfo.prev = now;
371 hashlimit_init_dst(struct xt_hashlimit_htable *hinfo, struct dsthash_dst *dst,
372 const struct sk_buff *skb, unsigned int protoff)
374 __be16 _ports[2], *ports;
377 memset(dst, 0, sizeof(*dst));
379 switch (hinfo->family) {
381 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
382 dst->addr.ip.dst = ip_hdr(skb)->daddr;
383 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
384 dst->addr.ip.src = ip_hdr(skb)->saddr;
386 if (!(hinfo->cfg.mode &
387 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
389 nexthdr = ip_hdr(skb)->protocol;
391 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
393 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
394 memcpy(&dst->addr.ip6.dst, &ipv6_hdr(skb)->daddr,
395 sizeof(dst->addr.ip6.dst));
396 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
397 memcpy(&dst->addr.ip6.src, &ipv6_hdr(skb)->saddr,
398 sizeof(dst->addr.ip6.src));
400 if (!(hinfo->cfg.mode &
401 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
403 nexthdr = ipv6_find_hdr(skb, &protoff, -1, NULL);
416 case IPPROTO_UDPLITE:
419 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
423 _ports[0] = _ports[1] = 0;
429 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
430 dst->src_port = ports[0];
431 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
432 dst->dst_port = ports[1];
437 hashlimit_match(const struct sk_buff *skb,
438 const struct net_device *in,
439 const struct net_device *out,
440 const struct xt_match *match,
441 const void *matchinfo,
443 unsigned int protoff,
446 struct xt_hashlimit_info *r =
447 ((struct xt_hashlimit_info *)matchinfo)->u.master;
448 struct xt_hashlimit_htable *hinfo = r->hinfo;
449 unsigned long now = jiffies;
450 struct dsthash_ent *dh;
451 struct dsthash_dst dst;
453 if (hashlimit_init_dst(hinfo, &dst, skb, protoff) < 0)
456 spin_lock_bh(&hinfo->lock);
457 dh = dsthash_find(hinfo, &dst);
459 dh = dsthash_alloc_init(hinfo, &dst);
461 spin_unlock_bh(&hinfo->lock);
465 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
466 dh->rateinfo.prev = jiffies;
467 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
469 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
471 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
473 /* update expiration timeout */
474 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
475 rateinfo_recalc(dh, now);
478 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
479 /* We're underlimit. */
480 dh->rateinfo.credit -= dh->rateinfo.cost;
481 spin_unlock_bh(&hinfo->lock);
485 spin_unlock_bh(&hinfo->lock);
487 /* default case: we're overlimit, thus don't match */
496 hashlimit_checkentry(const char *tablename,
498 const struct xt_match *match,
500 unsigned int hook_mask)
502 struct xt_hashlimit_info *r = matchinfo;
504 /* Check for overflow. */
505 if (r->cfg.burst == 0 ||
506 user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
507 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
508 r->cfg.avg, r->cfg.burst);
511 if (r->cfg.mode == 0 ||
512 r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
513 XT_HASHLIMIT_HASH_DIP |
514 XT_HASHLIMIT_HASH_SIP |
515 XT_HASHLIMIT_HASH_SPT))
517 if (!r->cfg.gc_interval)
521 if (r->name[sizeof(r->name) - 1] != '\0')
524 /* This is the best we've got: We cannot release and re-grab lock,
525 * since checkentry() is called before x_tables.c grabs xt_mutex.
526 * We also cannot grab the hashtable spinlock, since htable_create will
527 * call vmalloc, and that can sleep. And we cannot just re-search
528 * the list of htable's in htable_create(), since then we would
529 * create duplicate proc files. -HW */
530 mutex_lock(&hlimit_mutex);
531 r->hinfo = htable_find_get(r->name, match->family);
532 if (!r->hinfo && htable_create(r, match->family) != 0) {
533 mutex_unlock(&hlimit_mutex);
536 mutex_unlock(&hlimit_mutex);
538 /* Ugly hack: For SMP, we only want to use one set */
544 hashlimit_destroy(const struct xt_match *match, void *matchinfo)
546 struct xt_hashlimit_info *r = matchinfo;
548 htable_put(r->hinfo);
552 struct compat_xt_hashlimit_info {
554 struct hashlimit_cfg cfg;
556 compat_uptr_t master;
559 static void compat_from_user(void *dst, void *src)
561 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
563 memcpy(dst, src, off);
564 memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
567 static int compat_to_user(void __user *dst, void *src)
569 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
571 return copy_to_user(dst, src, off) ? -EFAULT : 0;
575 static struct xt_match xt_hashlimit[] = {
579 .match = hashlimit_match,
580 .matchsize = sizeof(struct xt_hashlimit_info),
582 .compatsize = sizeof(struct compat_xt_hashlimit_info),
583 .compat_from_user = compat_from_user,
584 .compat_to_user = compat_to_user,
586 .checkentry = hashlimit_checkentry,
587 .destroy = hashlimit_destroy,
593 .match = hashlimit_match,
594 .matchsize = sizeof(struct xt_hashlimit_info),
596 .compatsize = sizeof(struct compat_xt_hashlimit_info),
597 .compat_from_user = compat_from_user,
598 .compat_to_user = compat_to_user,
600 .checkentry = hashlimit_checkentry,
601 .destroy = hashlimit_destroy,
607 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
609 struct proc_dir_entry *pde = s->private;
610 struct xt_hashlimit_htable *htable = pde->data;
611 unsigned int *bucket;
613 spin_lock_bh(&htable->lock);
614 if (*pos >= htable->cfg.size)
617 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
619 return ERR_PTR(-ENOMEM);
625 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
627 struct proc_dir_entry *pde = s->private;
628 struct xt_hashlimit_htable *htable = pde->data;
629 unsigned int *bucket = (unsigned int *)v;
632 if (*pos >= htable->cfg.size) {
639 static void dl_seq_stop(struct seq_file *s, void *v)
641 struct proc_dir_entry *pde = s->private;
642 struct xt_hashlimit_htable *htable = pde->data;
643 unsigned int *bucket = (unsigned int *)v;
646 spin_unlock_bh(&htable->lock);
649 static int dl_seq_real_show(struct dsthash_ent *ent, int family,
652 /* recalculate to show accurate numbers */
653 rateinfo_recalc(ent, jiffies);
657 return seq_printf(s, "%ld %u.%u.%u.%u:%u->"
658 "%u.%u.%u.%u:%u %u %u %u\n",
659 (long)(ent->expires - jiffies)/HZ,
660 NIPQUAD(ent->dst.addr.ip.src),
661 ntohs(ent->dst.src_port),
662 NIPQUAD(ent->dst.addr.ip.dst),
663 ntohs(ent->dst.dst_port),
664 ent->rateinfo.credit, ent->rateinfo.credit_cap,
667 return seq_printf(s, "%ld " NIP6_FMT ":%u->"
668 NIP6_FMT ":%u %u %u %u\n",
669 (long)(ent->expires - jiffies)/HZ,
670 NIP6(*(struct in6_addr *)&ent->dst.addr.ip6.src),
671 ntohs(ent->dst.src_port),
672 NIP6(*(struct in6_addr *)&ent->dst.addr.ip6.dst),
673 ntohs(ent->dst.dst_port),
674 ent->rateinfo.credit, ent->rateinfo.credit_cap,
682 static int dl_seq_show(struct seq_file *s, void *v)
684 struct proc_dir_entry *pde = s->private;
685 struct xt_hashlimit_htable *htable = pde->data;
686 unsigned int *bucket = (unsigned int *)v;
687 struct dsthash_ent *ent;
688 struct hlist_node *pos;
690 if (!hlist_empty(&htable->hash[*bucket])) {
691 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
692 if (dl_seq_real_show(ent, htable->family, s))
698 static struct seq_operations dl_seq_ops = {
699 .start = dl_seq_start,
705 static int dl_proc_open(struct inode *inode, struct file *file)
707 int ret = seq_open(file, &dl_seq_ops);
710 struct seq_file *sf = file->private_data;
711 sf->private = PDE(inode);
716 static const struct file_operations dl_file_ops = {
717 .owner = THIS_MODULE,
718 .open = dl_proc_open,
721 .release = seq_release
724 static int __init xt_hashlimit_init(void)
728 err = xt_register_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
733 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
734 sizeof(struct dsthash_ent), 0, 0,
736 if (!hashlimit_cachep) {
737 printk(KERN_ERR "xt_hashlimit: unable to create slab cache\n");
740 hashlimit_procdir4 = proc_mkdir("ipt_hashlimit", proc_net);
741 if (!hashlimit_procdir4) {
742 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
746 hashlimit_procdir6 = proc_mkdir("ip6t_hashlimit", proc_net);
747 if (!hashlimit_procdir6) {
748 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
754 remove_proc_entry("ipt_hashlimit", proc_net);
756 kmem_cache_destroy(hashlimit_cachep);
758 xt_unregister_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
764 static void __exit xt_hashlimit_fini(void)
766 remove_proc_entry("ipt_hashlimit", proc_net);
767 remove_proc_entry("ip6t_hashlimit", proc_net);
768 kmem_cache_destroy(hashlimit_cachep);
769 xt_unregister_matches(xt_hashlimit, ARRAY_SIZE(xt_hashlimit));
772 module_init(xt_hashlimit_init);
773 module_exit(xt_hashlimit_fini);