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 * based on ipt_limit.c by:
11 * Jérôme de Vivie <devivie@info.enserb.u-bordeaux.fr>
12 * Hervé Eychenne <eychenne@info.enserb.u-bordeaux.fr>
13 * Rusty Russell <rusty@rustcorp.com.au>
15 * The general idea is to create a hash table for every dstip and have a
16 * seperate limit counter per tuple. This way you can do something like 'limit
17 * the number of syn packets for each of my internal addresses.
19 * Ideally this would just be implemented as a general 'hash' match, which would
20 * allow us to attach any iptables target to it's hash buckets. But this is
21 * not possible in the current iptables architecture. As always, pkttables for
24 #include <linux/module.h>
25 #include <linux/skbuff.h>
26 #include <linux/spinlock.h>
27 #include <linux/random.h>
28 #include <linux/jhash.h>
29 #include <linux/slab.h>
30 #include <linux/vmalloc.h>
31 #include <linux/proc_fs.h>
32 #include <linux/seq_file.h>
33 #include <linux/list.h>
35 #include <linux/netfilter_ipv4/ip_tables.h>
36 #include <linux/netfilter_ipv4/ipt_hashlimit.h>
38 /* FIXME: this is just for IP_NF_ASSERRT */
39 #include <linux/netfilter_ipv4/ip_conntrack.h>
40 #include <linux/mutex.h>
42 MODULE_LICENSE("GPL");
43 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
44 MODULE_DESCRIPTION("iptables match for limiting per hash-bucket");
46 /* need to declare this at the top */
47 static struct proc_dir_entry *hashlimit_procdir;
48 static struct file_operations dl_file_ops;
55 /* ports have to be consecutive !!! */
61 /* static / read-only parts in the beginning */
62 struct hlist_node node;
63 struct dsthash_dst dst;
65 /* modified structure members in the end */
66 unsigned long expires; /* precalculated expiry time */
68 unsigned long prev; /* last modification */
70 u_int32_t credit_cap, cost;
74 struct ipt_hashlimit_htable {
75 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 struct timer_list timer; /* timer for gc */
85 atomic_t count; /* number entries in table */
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 kmem_cache_t *hashlimit_cachep __read_mostly;
98 static inline int dst_cmp(const struct dsthash_ent *ent, struct dsthash_dst *b)
100 return (ent->dst.dst_ip == b->dst_ip
101 && ent->dst.dst_port == b->dst_port
102 && ent->dst.src_port == b->src_port
103 && ent->dst.src_ip == b->src_ip);
106 static inline u_int32_t
107 hash_dst(const struct ipt_hashlimit_htable *ht, const struct dsthash_dst *dst)
109 return (jhash_3words((__force u32)dst->dst_ip,
110 ((__force u32)dst->dst_port<<16 |
111 (__force u32)dst->src_port),
112 (__force u32)dst->src_ip, ht->rnd) % ht->cfg.size);
115 static inline struct dsthash_ent *
116 __dsthash_find(const struct ipt_hashlimit_htable *ht, struct dsthash_dst *dst)
118 struct dsthash_ent *ent;
119 struct hlist_node *pos;
120 u_int32_t hash = hash_dst(ht, dst);
122 if (!hlist_empty(&ht->hash[hash]))
123 hlist_for_each_entry(ent, pos, &ht->hash[hash], node) {
124 if (dst_cmp(ent, dst)) {
132 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
133 static struct dsthash_ent *
134 __dsthash_alloc_init(struct ipt_hashlimit_htable *ht, struct dsthash_dst *dst)
136 struct dsthash_ent *ent;
138 /* initialize hash with random val at the time we allocate
139 * the first hashtable entry */
140 if (!ht->rnd_initialized) {
141 get_random_bytes(&ht->rnd, 4);
142 ht->rnd_initialized = 1;
146 atomic_read(&ht->count) >= ht->cfg.max) {
147 /* FIXME: do something. question is what.. */
150 "ipt_hashlimit: max count of %u reached\n",
155 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
159 "ipt_hashlimit: can't allocate dsthash_ent\n");
163 atomic_inc(&ht->count);
165 ent->dst.dst_ip = dst->dst_ip;
166 ent->dst.dst_port = dst->dst_port;
167 ent->dst.src_ip = dst->src_ip;
168 ent->dst.src_port = dst->src_port;
170 hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
176 __dsthash_free(struct ipt_hashlimit_htable *ht, struct dsthash_ent *ent)
178 hlist_del(&ent->node);
179 kmem_cache_free(hashlimit_cachep, ent);
180 atomic_dec(&ht->count);
182 static void htable_gc(unsigned long htlong);
184 static int htable_create(struct ipt_hashlimit_info *minfo)
188 struct ipt_hashlimit_htable *hinfo;
191 size = minfo->cfg.size;
193 size = (((num_physpages << PAGE_SHIFT) / 16384)
194 / sizeof(struct list_head));
195 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
200 /* FIXME: don't use vmalloc() here or anywhere else -HW */
201 hinfo = vmalloc(sizeof(struct ipt_hashlimit_htable)
202 + (sizeof(struct list_head) * size));
204 printk(KERN_ERR "ipt_hashlimit: Unable to create hashtable\n");
207 minfo->hinfo = hinfo;
209 /* copy match config into hashtable config */
210 memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
211 hinfo->cfg.size = size;
213 hinfo->cfg.max = 8 * hinfo->cfg.size;
214 else if (hinfo->cfg.max < hinfo->cfg.size)
215 hinfo->cfg.max = hinfo->cfg.size;
217 for (i = 0; i < hinfo->cfg.size; i++)
218 INIT_HLIST_HEAD(&hinfo->hash[i]);
220 atomic_set(&hinfo->count, 0);
221 atomic_set(&hinfo->use, 1);
222 hinfo->rnd_initialized = 0;
223 spin_lock_init(&hinfo->lock);
224 hinfo->pde = create_proc_entry(minfo->name, 0, hashlimit_procdir);
229 hinfo->pde->proc_fops = &dl_file_ops;
230 hinfo->pde->data = hinfo;
232 init_timer(&hinfo->timer);
233 hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
234 hinfo->timer.data = (unsigned long )hinfo;
235 hinfo->timer.function = htable_gc;
236 add_timer(&hinfo->timer);
238 spin_lock_bh(&hashlimit_lock);
239 hlist_add_head(&hinfo->node, &hashlimit_htables);
240 spin_unlock_bh(&hashlimit_lock);
245 static int select_all(struct ipt_hashlimit_htable *ht, struct dsthash_ent *he)
250 static int select_gc(struct ipt_hashlimit_htable *ht, struct dsthash_ent *he)
252 return (jiffies >= he->expires);
255 static void htable_selective_cleanup(struct ipt_hashlimit_htable *ht,
256 int (*select)(struct ipt_hashlimit_htable *ht,
257 struct dsthash_ent *he))
261 IP_NF_ASSERT(ht->cfg.size && ht->cfg.max);
263 /* lock hash table and iterate over it */
264 spin_lock_bh(&ht->lock);
265 for (i = 0; i < ht->cfg.size; i++) {
266 struct dsthash_ent *dh;
267 struct hlist_node *pos, *n;
268 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
269 if ((*select)(ht, dh))
270 __dsthash_free(ht, dh);
273 spin_unlock_bh(&ht->lock);
276 /* hash table garbage collector, run by timer */
277 static void htable_gc(unsigned long htlong)
279 struct ipt_hashlimit_htable *ht = (struct ipt_hashlimit_htable *)htlong;
281 htable_selective_cleanup(ht, select_gc);
283 /* re-add the timer accordingly */
284 ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
285 add_timer(&ht->timer);
288 static void htable_destroy(struct ipt_hashlimit_htable *hinfo)
290 /* remove timer, if it is pending */
291 if (timer_pending(&hinfo->timer))
292 del_timer(&hinfo->timer);
294 /* remove proc entry */
295 remove_proc_entry(hinfo->pde->name, hashlimit_procdir);
297 htable_selective_cleanup(hinfo, select_all);
301 static struct ipt_hashlimit_htable *htable_find_get(char *name)
303 struct ipt_hashlimit_htable *hinfo;
304 struct hlist_node *pos;
306 spin_lock_bh(&hashlimit_lock);
307 hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) {
308 if (!strcmp(name, hinfo->pde->name)) {
309 atomic_inc(&hinfo->use);
310 spin_unlock_bh(&hashlimit_lock);
314 spin_unlock_bh(&hashlimit_lock);
319 static void htable_put(struct ipt_hashlimit_htable *hinfo)
321 if (atomic_dec_and_test(&hinfo->use)) {
322 spin_lock_bh(&hashlimit_lock);
323 hlist_del(&hinfo->node);
324 spin_unlock_bh(&hashlimit_lock);
325 htable_destroy(hinfo);
330 /* The algorithm used is the Simple Token Bucket Filter (TBF)
331 * see net/sched/sch_tbf.c in the linux source tree
334 /* Rusty: This is my (non-mathematically-inclined) understanding of
335 this algorithm. The `average rate' in jiffies becomes your initial
336 amount of credit `credit' and the most credit you can ever have
337 `credit_cap'. The `peak rate' becomes the cost of passing the
340 `prev' tracks the last packet hit: you gain one credit per jiffy.
341 If you get credit balance more than this, the extra credit is
342 discarded. Every time the match passes, you lose `cost' credits;
343 if you don't have that many, the test fails.
345 See Alexey's formal explanation in net/sched/sch_tbf.c.
347 To get the maximum range, we multiply by this factor (ie. you get N
348 credits per jiffy). We want to allow a rate as low as 1 per day
349 (slowest userspace tool allows), which means
350 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
352 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
354 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
355 * us the power of 2 below the theoretical max, so GCC simply does a
357 #define _POW2_BELOW2(x) ((x)|((x)>>1))
358 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
359 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
360 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
361 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
362 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
364 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
366 /* Precision saver. */
367 static inline u_int32_t
368 user2credits(u_int32_t user)
370 /* If multiplying would overflow... */
371 if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
373 return (user / IPT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
375 return (user * HZ * CREDITS_PER_JIFFY) / IPT_HASHLIMIT_SCALE;
378 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
380 dh->rateinfo.credit += (now - xchg(&dh->rateinfo.prev, now))
382 if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
383 dh->rateinfo.credit = dh->rateinfo.credit_cap;
387 hashlimit_match(const struct sk_buff *skb,
388 const struct net_device *in,
389 const struct net_device *out,
390 const struct xt_match *match,
391 const void *matchinfo,
393 unsigned int protoff,
396 struct ipt_hashlimit_info *r =
397 ((struct ipt_hashlimit_info *)matchinfo)->u.master;
398 struct ipt_hashlimit_htable *hinfo = r->hinfo;
399 unsigned long now = jiffies;
400 struct dsthash_ent *dh;
401 struct dsthash_dst dst;
403 /* build 'dst' according to hinfo->cfg and current packet */
404 memset(&dst, 0, sizeof(dst));
405 if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DIP)
406 dst.dst_ip = skb->nh.iph->daddr;
407 if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SIP)
408 dst.src_ip = skb->nh.iph->saddr;
409 if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DPT
410 ||hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SPT) {
411 __be16 _ports[2], *ports;
413 switch (skb->nh.iph->protocol) {
418 ports = skb_header_pointer(skb, skb->nh.iph->ihl*4,
419 sizeof(_ports), &_ports);
422 _ports[0] = _ports[1] = 0;
427 /* We've been asked to examine this packet, and we
428 can't. Hence, no choice but to drop. */
432 if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_SPT)
433 dst.src_port = ports[0];
434 if (hinfo->cfg.mode & IPT_HASHLIMIT_HASH_DPT)
435 dst.dst_port = ports[1];
438 spin_lock_bh(&hinfo->lock);
439 dh = __dsthash_find(hinfo, &dst);
441 dh = __dsthash_alloc_init(hinfo, &dst);
444 /* enomem... don't match == DROP */
446 printk(KERN_ERR "%s: ENOMEM\n", __FUNCTION__);
447 spin_unlock_bh(&hinfo->lock);
451 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
453 dh->rateinfo.prev = jiffies;
454 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
456 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
458 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
460 /* update expiration timeout */
461 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
462 rateinfo_recalc(dh, now);
465 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
466 /* We're underlimit. */
467 dh->rateinfo.credit -= dh->rateinfo.cost;
468 spin_unlock_bh(&hinfo->lock);
472 spin_unlock_bh(&hinfo->lock);
474 /* default case: we're overlimit, thus don't match */
479 hashlimit_checkentry(const char *tablename,
481 const struct xt_match *match,
483 unsigned int hook_mask)
485 struct ipt_hashlimit_info *r = matchinfo;
487 /* Check for overflow. */
488 if (r->cfg.burst == 0
489 || user2credits(r->cfg.avg * r->cfg.burst) <
490 user2credits(r->cfg.avg)) {
491 printk(KERN_ERR "ipt_hashlimit: Overflow, try lower: %u/%u\n",
492 r->cfg.avg, r->cfg.burst);
497 || r->cfg.mode > (IPT_HASHLIMIT_HASH_DPT
498 |IPT_HASHLIMIT_HASH_DIP
499 |IPT_HASHLIMIT_HASH_SIP
500 |IPT_HASHLIMIT_HASH_SPT))
503 if (!r->cfg.gc_interval)
509 if (r->name[sizeof(r->name) - 1] != '\0')
512 /* This is the best we've got: We cannot release and re-grab lock,
513 * since checkentry() is called before ip_tables.c grabs ipt_mutex.
514 * We also cannot grab the hashtable spinlock, since htable_create will
515 * call vmalloc, and that can sleep. And we cannot just re-search
516 * the list of htable's in htable_create(), since then we would
517 * create duplicate proc files. -HW */
518 mutex_lock(&hlimit_mutex);
519 r->hinfo = htable_find_get(r->name);
520 if (!r->hinfo && (htable_create(r) != 0)) {
521 mutex_unlock(&hlimit_mutex);
524 mutex_unlock(&hlimit_mutex);
526 /* Ugly hack: For SMP, we only want to use one set */
533 hashlimit_destroy(const struct xt_match *match, void *matchinfo)
535 struct ipt_hashlimit_info *r = matchinfo;
537 htable_put(r->hinfo);
541 struct compat_ipt_hashlimit_info {
543 struct hashlimit_cfg cfg;
545 compat_uptr_t master;
548 static void compat_from_user(void *dst, void *src)
550 int off = offsetof(struct compat_ipt_hashlimit_info, hinfo);
552 memcpy(dst, src, off);
553 memset(dst + off, 0, sizeof(struct compat_ipt_hashlimit_info) - off);
556 static int compat_to_user(void __user *dst, void *src)
558 int off = offsetof(struct compat_ipt_hashlimit_info, hinfo);
560 return copy_to_user(dst, src, off) ? -EFAULT : 0;
564 static struct ipt_match ipt_hashlimit = {
566 .match = hashlimit_match,
567 .matchsize = sizeof(struct ipt_hashlimit_info),
569 .compatsize = sizeof(struct compat_ipt_hashlimit_info),
570 .compat_from_user = compat_from_user,
571 .compat_to_user = compat_to_user,
573 .checkentry = hashlimit_checkentry,
574 .destroy = hashlimit_destroy,
580 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
582 struct proc_dir_entry *pde = s->private;
583 struct ipt_hashlimit_htable *htable = pde->data;
584 unsigned int *bucket;
586 spin_lock_bh(&htable->lock);
587 if (*pos >= htable->cfg.size)
590 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
592 return ERR_PTR(-ENOMEM);
598 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
600 struct proc_dir_entry *pde = s->private;
601 struct ipt_hashlimit_htable *htable = pde->data;
602 unsigned int *bucket = (unsigned int *)v;
605 if (*pos >= htable->cfg.size) {
612 static void dl_seq_stop(struct seq_file *s, void *v)
614 struct proc_dir_entry *pde = s->private;
615 struct ipt_hashlimit_htable *htable = pde->data;
616 unsigned int *bucket = (unsigned int *)v;
620 spin_unlock_bh(&htable->lock);
623 static inline int dl_seq_real_show(struct dsthash_ent *ent, struct seq_file *s)
625 /* recalculate to show accurate numbers */
626 rateinfo_recalc(ent, jiffies);
628 return seq_printf(s, "%ld %u.%u.%u.%u:%u->%u.%u.%u.%u:%u %u %u %u\n",
629 (long)(ent->expires - jiffies)/HZ,
630 NIPQUAD(ent->dst.src_ip), ntohs(ent->dst.src_port),
631 NIPQUAD(ent->dst.dst_ip), ntohs(ent->dst.dst_port),
632 ent->rateinfo.credit, ent->rateinfo.credit_cap,
636 static int dl_seq_show(struct seq_file *s, void *v)
638 struct proc_dir_entry *pde = s->private;
639 struct ipt_hashlimit_htable *htable = pde->data;
640 unsigned int *bucket = (unsigned int *)v;
641 struct dsthash_ent *ent;
642 struct hlist_node *pos;
644 if (!hlist_empty(&htable->hash[*bucket]))
645 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node) {
646 if (dl_seq_real_show(ent, s)) {
647 /* buffer was filled and unable to print that tuple */
655 static struct seq_operations dl_seq_ops = {
656 .start = dl_seq_start,
662 static int dl_proc_open(struct inode *inode, struct file *file)
664 int ret = seq_open(file, &dl_seq_ops);
667 struct seq_file *sf = file->private_data;
668 sf->private = PDE(inode);
673 static struct file_operations dl_file_ops = {
674 .owner = THIS_MODULE,
675 .open = dl_proc_open,
678 .release = seq_release
681 static int init_or_fini(int fini)
688 if (ipt_register_match(&ipt_hashlimit)) {
690 goto cleanup_nothing;
693 hashlimit_cachep = kmem_cache_create("ipt_hashlimit",
694 sizeof(struct dsthash_ent), 0,
696 if (!hashlimit_cachep) {
697 printk(KERN_ERR "Unable to create ipt_hashlimit slab cache\n");
699 goto cleanup_unreg_match;
702 hashlimit_procdir = proc_mkdir("ipt_hashlimit", proc_net);
703 if (!hashlimit_procdir) {
704 printk(KERN_ERR "Unable to create proc dir entry\n");
706 goto cleanup_free_slab;
712 remove_proc_entry("ipt_hashlimit", proc_net);
714 kmem_cache_destroy(hashlimit_cachep);
716 ipt_unregister_match(&ipt_hashlimit);
722 static int __init ipt_hashlimit_init(void)
724 return init_or_fini(0);
727 static void __exit ipt_hashlimit_fini(void)
732 module_init(ipt_hashlimit_init);
733 module_exit(ipt_hashlimit_fini);