1 // SPDX-License-Identifier: GPL-2.0-only
3 * xt_hashlimit - Netfilter module to limit the number of packets per time
4 * separately for each hashbucket (sourceip/sourceport/dstip/dstport)
6 * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
7 * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
8 * Copyright © CC Computer Consultants GmbH, 2007 - 2008
10 * Development of this code was funded by Astaro AG, http://www.astaro.com/
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/module.h>
14 #include <linux/spinlock.h>
15 #include <linux/random.h>
16 #include <linux/jhash.h>
17 #include <linux/slab.h>
18 #include <linux/vmalloc.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/list.h>
22 #include <linux/skbuff.h>
26 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
27 #include <linux/ipv6.h>
31 #include <net/net_namespace.h>
32 #include <net/netns/generic.h>
34 #include <linux/netfilter/x_tables.h>
35 #include <linux/netfilter_ipv4/ip_tables.h>
36 #include <linux/netfilter_ipv6/ip6_tables.h>
37 #include <linux/mutex.h>
38 #include <linux/kernel.h>
39 #include <linux/refcount.h>
40 #include <uapi/linux/netfilter/xt_hashlimit.h>
42 #define XT_HASHLIMIT_ALL (XT_HASHLIMIT_HASH_DIP | XT_HASHLIMIT_HASH_DPT | \
43 XT_HASHLIMIT_HASH_SIP | XT_HASHLIMIT_HASH_SPT | \
44 XT_HASHLIMIT_INVERT | XT_HASHLIMIT_BYTES |\
45 XT_HASHLIMIT_RATE_MATCH)
47 MODULE_LICENSE("GPL");
48 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
49 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
50 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
51 MODULE_ALIAS("ipt_hashlimit");
52 MODULE_ALIAS("ip6t_hashlimit");
54 struct hashlimit_net {
55 struct hlist_head htables;
56 struct proc_dir_entry *ipt_hashlimit;
57 struct proc_dir_entry *ip6t_hashlimit;
60 static unsigned int hashlimit_net_id;
61 static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
63 return net_generic(net, hashlimit_net_id);
66 /* need to declare this at the top */
67 static const struct seq_operations dl_seq_ops_v2;
68 static const struct seq_operations dl_seq_ops_v1;
69 static const struct seq_operations dl_seq_ops;
78 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
90 /* static / read-only parts in the beginning */
91 struct hlist_node node;
92 struct dsthash_dst dst;
94 /* modified structure members in the end */
96 unsigned long expires; /* precalculated expiry time */
98 unsigned long prev; /* last modification */
102 u_int64_t credit_cap;
106 u_int32_t interval, prev_window;
107 u_int64_t current_rate;
116 struct xt_hashlimit_htable {
117 struct hlist_node node; /* global list of all htables */
120 bool rnd_initialized;
122 struct hashlimit_cfg3 cfg; /* config */
124 /* used internally */
125 spinlock_t lock; /* lock for list_head */
126 u_int32_t rnd; /* random seed for hash */
127 unsigned int count; /* number entries in table */
128 struct delayed_work gc_work;
131 struct proc_dir_entry *pde;
135 struct hlist_head hash[0]; /* hashtable itself */
139 cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
142 struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
144 to->mode = cfg->mode;
146 to->burst = cfg->burst;
147 to->size = cfg->size;
149 to->gc_interval = cfg->gc_interval;
150 to->expire = cfg->expire;
151 to->srcmask = cfg->srcmask;
152 to->dstmask = cfg->dstmask;
153 } else if (revision == 2) {
154 struct hashlimit_cfg2 *cfg = (struct hashlimit_cfg2 *)from;
156 to->mode = cfg->mode;
158 to->burst = cfg->burst;
159 to->size = cfg->size;
161 to->gc_interval = cfg->gc_interval;
162 to->expire = cfg->expire;
163 to->srcmask = cfg->srcmask;
164 to->dstmask = cfg->dstmask;
165 } else if (revision == 3) {
166 memcpy(to, from, sizeof(struct hashlimit_cfg3));
174 static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */
175 static struct kmem_cache *hashlimit_cachep __read_mostly;
177 static inline bool dst_cmp(const struct dsthash_ent *ent,
178 const struct dsthash_dst *b)
180 return !memcmp(&ent->dst, b, sizeof(ent->dst));
184 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
186 u_int32_t hash = jhash2((const u32 *)dst,
187 sizeof(*dst)/sizeof(u32),
190 * Instead of returning hash % ht->cfg.size (implying a divide)
191 * we return the high 32 bits of the (hash * ht->cfg.size) that will
192 * give results between [0 and cfg.size-1] and same hash distribution,
193 * but using a multiply, less expensive than a divide
195 return reciprocal_scale(hash, ht->cfg.size);
198 static struct dsthash_ent *
199 dsthash_find(const struct xt_hashlimit_htable *ht,
200 const struct dsthash_dst *dst)
202 struct dsthash_ent *ent;
203 u_int32_t hash = hash_dst(ht, dst);
205 if (!hlist_empty(&ht->hash[hash])) {
206 hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
207 if (dst_cmp(ent, dst)) {
208 spin_lock(&ent->lock);
215 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
216 static struct dsthash_ent *
217 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
218 const struct dsthash_dst *dst, bool *race)
220 struct dsthash_ent *ent;
222 spin_lock(&ht->lock);
224 /* Two or more packets may race to create the same entry in the
225 * hashtable, double check if this packet lost race.
227 ent = dsthash_find(ht, dst);
229 spin_unlock(&ht->lock);
234 /* initialize hash with random val at the time we allocate
235 * the first hashtable entry */
236 if (unlikely(!ht->rnd_initialized)) {
237 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
238 ht->rnd_initialized = true;
241 if (ht->cfg.max && ht->count >= ht->cfg.max) {
242 /* FIXME: do something. question is what.. */
243 net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
246 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
248 memcpy(&ent->dst, dst, sizeof(ent->dst));
249 spin_lock_init(&ent->lock);
251 spin_lock(&ent->lock);
252 hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
255 spin_unlock(&ht->lock);
259 static void dsthash_free_rcu(struct rcu_head *head)
261 struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
263 kmem_cache_free(hashlimit_cachep, ent);
267 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
269 hlist_del_rcu(&ent->node);
270 call_rcu(&ent->rcu, dsthash_free_rcu);
273 static void htable_gc(struct work_struct *work);
275 static int htable_create(struct net *net, struct hashlimit_cfg3 *cfg,
276 const char *name, u_int8_t family,
277 struct xt_hashlimit_htable **out_hinfo,
280 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
281 struct xt_hashlimit_htable *hinfo;
282 const struct seq_operations *ops;
283 unsigned int size, i;
284 unsigned long nr_pages = totalram_pages();
290 size = (nr_pages << PAGE_SHIFT) / 16384 /
291 sizeof(struct hlist_head);
292 if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
297 /* FIXME: don't use vmalloc() here or anywhere else -HW */
298 hinfo = vmalloc(struct_size(hinfo, hash, size));
303 /* copy match config into hashtable config */
304 ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
310 hinfo->cfg.size = size;
311 if (hinfo->cfg.max == 0)
312 hinfo->cfg.max = 8 * hinfo->cfg.size;
313 else if (hinfo->cfg.max < hinfo->cfg.size)
314 hinfo->cfg.max = hinfo->cfg.size;
316 for (i = 0; i < hinfo->cfg.size; i++)
317 INIT_HLIST_HEAD(&hinfo->hash[i]);
319 refcount_set(&hinfo->use, 1);
321 hinfo->family = family;
322 hinfo->rnd_initialized = false;
323 hinfo->name = kstrdup(name, GFP_KERNEL);
328 spin_lock_init(&hinfo->lock);
332 ops = &dl_seq_ops_v1;
335 ops = &dl_seq_ops_v2;
341 hinfo->pde = proc_create_seq_data(name, 0,
342 (family == NFPROTO_IPV4) ?
343 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
345 if (hinfo->pde == NULL) {
352 INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
353 queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
354 msecs_to_jiffies(hinfo->cfg.gc_interval));
356 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
361 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, bool select_all)
365 for (i = 0; i < ht->cfg.size; i++) {
366 struct dsthash_ent *dh;
367 struct hlist_node *n;
369 spin_lock_bh(&ht->lock);
370 hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
371 if (time_after_eq(jiffies, dh->expires) || select_all)
372 dsthash_free(ht, dh);
374 spin_unlock_bh(&ht->lock);
379 static void htable_gc(struct work_struct *work)
381 struct xt_hashlimit_htable *ht;
383 ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
385 htable_selective_cleanup(ht, false);
387 queue_delayed_work(system_power_efficient_wq,
388 &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
391 static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
393 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
394 struct proc_dir_entry *parent;
396 if (hinfo->family == NFPROTO_IPV4)
397 parent = hashlimit_net->ipt_hashlimit;
399 parent = hashlimit_net->ip6t_hashlimit;
402 remove_proc_entry(hinfo->name, parent);
405 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
407 cancel_delayed_work_sync(&hinfo->gc_work);
408 htable_remove_proc_entry(hinfo);
409 htable_selective_cleanup(hinfo, true);
414 static struct xt_hashlimit_htable *htable_find_get(struct net *net,
418 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
419 struct xt_hashlimit_htable *hinfo;
421 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
422 if (!strcmp(name, hinfo->name) &&
423 hinfo->family == family) {
424 refcount_inc(&hinfo->use);
431 static void htable_put(struct xt_hashlimit_htable *hinfo)
433 if (refcount_dec_and_mutex_lock(&hinfo->use, &hashlimit_mutex)) {
434 hlist_del(&hinfo->node);
435 mutex_unlock(&hashlimit_mutex);
436 htable_destroy(hinfo);
440 /* The algorithm used is the Simple Token Bucket Filter (TBF)
441 * see net/sched/sch_tbf.c in the linux source tree
444 /* Rusty: This is my (non-mathematically-inclined) understanding of
445 this algorithm. The `average rate' in jiffies becomes your initial
446 amount of credit `credit' and the most credit you can ever have
447 `credit_cap'. The `peak rate' becomes the cost of passing the
450 `prev' tracks the last packet hit: you gain one credit per jiffy.
451 If you get credit balance more than this, the extra credit is
452 discarded. Every time the match passes, you lose `cost' credits;
453 if you don't have that many, the test fails.
455 See Alexey's formal explanation in net/sched/sch_tbf.c.
457 To get the maximum range, we multiply by this factor (ie. you get N
458 credits per jiffy). We want to allow a rate as low as 1 per day
459 (slowest userspace tool allows), which means
460 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
462 #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
463 #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
465 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
466 * us the power of 2 below the theoretical max, so GCC simply does a
468 #define _POW2_BELOW2(x) ((x)|((x)>>1))
469 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
470 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
471 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
472 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
473 #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
474 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
475 #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
477 #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
478 #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
480 /* in byte mode, the lowest possible rate is one packet/second.
481 * credit_cap is used as a counter that tells us how many times we can
482 * refill the "credits available" counter when it becomes empty.
484 #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
485 #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
487 static u32 xt_hashlimit_len_to_chunks(u32 len)
489 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
492 /* Precision saver. */
493 static u64 user2credits(u64 user, int revision)
495 u64 scale = (revision == 1) ?
496 XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
497 u64 cpj = (revision == 1) ?
498 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
500 /* Avoid overflow: divide the constant operands first */
501 if (scale >= HZ * cpj)
502 return div64_u64(user, div64_u64(scale, HZ * cpj));
504 return user * div64_u64(HZ * cpj, scale);
507 static u32 user2credits_byte(u32 user)
510 us *= HZ * CREDITS_PER_JIFFY_BYTES;
511 return (u32) (us >> 32);
514 static u64 user2rate(u64 user)
517 return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
519 pr_info_ratelimited("invalid rate from userspace: %llu\n",
525 static u64 user2rate_bytes(u32 user)
529 r = user ? U32_MAX / user : U32_MAX;
530 return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
533 static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
534 u32 mode, int revision)
536 unsigned long delta = now - dh->rateinfo.prev;
542 if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
543 u64 interval = dh->rateinfo.interval * HZ;
545 if (delta < interval)
548 dh->rateinfo.prev = now;
549 dh->rateinfo.prev_window =
550 ((dh->rateinfo.current_rate * interval) >
551 (delta * dh->rateinfo.rate));
552 dh->rateinfo.current_rate = 0;
557 dh->rateinfo.prev = now;
559 if (mode & XT_HASHLIMIT_BYTES) {
560 u64 tmp = dh->rateinfo.credit;
561 dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
562 cap = CREDITS_PER_JIFFY_BYTES * HZ;
563 if (tmp >= dh->rateinfo.credit) {/* overflow */
564 dh->rateinfo.credit = cap;
568 cpj = (revision == 1) ?
569 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
570 dh->rateinfo.credit += delta * cpj;
571 cap = dh->rateinfo.credit_cap;
573 if (dh->rateinfo.credit > cap)
574 dh->rateinfo.credit = cap;
577 static void rateinfo_init(struct dsthash_ent *dh,
578 struct xt_hashlimit_htable *hinfo, int revision)
580 dh->rateinfo.prev = jiffies;
581 if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
582 dh->rateinfo.prev_window = 0;
583 dh->rateinfo.current_rate = 0;
584 if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
586 user2rate_bytes((u32)hinfo->cfg.avg);
587 if (hinfo->cfg.burst)
589 hinfo->cfg.burst * dh->rateinfo.rate;
591 dh->rateinfo.burst = dh->rateinfo.rate;
593 dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
595 hinfo->cfg.burst + dh->rateinfo.rate;
597 dh->rateinfo.interval = hinfo->cfg.interval;
598 } else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
599 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
600 dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
601 dh->rateinfo.credit_cap = hinfo->cfg.burst;
603 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
604 hinfo->cfg.burst, revision);
605 dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
606 dh->rateinfo.credit_cap = dh->rateinfo.credit;
610 static inline __be32 maskl(__be32 a, unsigned int l)
612 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
615 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
616 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
620 i[0] = maskl(i[0], p);
621 i[1] = i[2] = i[3] = 0;
624 i[1] = maskl(i[1], p - 32);
628 i[2] = maskl(i[2], p - 64);
632 i[3] = maskl(i[3], p - 96);
641 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
642 struct dsthash_dst *dst,
643 const struct sk_buff *skb, unsigned int protoff)
645 __be16 _ports[2], *ports;
649 memset(dst, 0, sizeof(*dst));
651 switch (hinfo->family) {
653 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
654 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
656 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
657 dst->ip.src = maskl(ip_hdr(skb)->saddr,
660 if (!(hinfo->cfg.mode &
661 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
663 nexthdr = ip_hdr(skb)->protocol;
665 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
670 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
671 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
672 sizeof(dst->ip6.dst));
673 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
675 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
676 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
677 sizeof(dst->ip6.src));
678 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
681 if (!(hinfo->cfg.mode &
682 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
684 nexthdr = ipv6_hdr(skb)->nexthdr;
685 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
686 if ((int)protoff < 0)
696 poff = proto_ports_offset(nexthdr);
698 ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
701 _ports[0] = _ports[1] = 0;
706 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
707 dst->src_port = ports[0];
708 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
709 dst->dst_port = ports[1];
713 static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
715 u64 tmp = xt_hashlimit_len_to_chunks(len);
716 tmp = tmp * dh->rateinfo.cost;
718 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
719 tmp = CREDITS_PER_JIFFY_BYTES * HZ;
721 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
722 dh->rateinfo.credit_cap--;
723 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
729 hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
730 struct xt_hashlimit_htable *hinfo,
731 const struct hashlimit_cfg3 *cfg, int revision)
733 unsigned long now = jiffies;
734 struct dsthash_ent *dh;
735 struct dsthash_dst dst;
739 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
743 dh = dsthash_find(hinfo, &dst);
745 dh = dsthash_alloc_init(hinfo, &dst, &race);
750 /* Already got an entry, update expiration timeout */
751 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
752 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
754 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
755 rateinfo_init(dh, hinfo, revision);
758 /* update expiration timeout */
759 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
760 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
763 if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
764 cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
765 dh->rateinfo.current_rate += cost;
767 if (!dh->rateinfo.prev_window &&
768 (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
769 spin_unlock(&dh->lock);
771 return !(cfg->mode & XT_HASHLIMIT_INVERT);
777 if (cfg->mode & XT_HASHLIMIT_BYTES)
778 cost = hashlimit_byte_cost(skb->len, dh);
780 cost = dh->rateinfo.cost;
782 if (dh->rateinfo.credit >= cost) {
783 /* below the limit */
784 dh->rateinfo.credit -= cost;
785 spin_unlock(&dh->lock);
787 return !(cfg->mode & XT_HASHLIMIT_INVERT);
791 spin_unlock(&dh->lock);
793 /* default match is underlimit - so over the limit, we need to invert */
794 return cfg->mode & XT_HASHLIMIT_INVERT;
802 hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
804 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
805 struct xt_hashlimit_htable *hinfo = info->hinfo;
806 struct hashlimit_cfg3 cfg = {};
809 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
813 return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
817 hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
819 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
820 struct xt_hashlimit_htable *hinfo = info->hinfo;
821 struct hashlimit_cfg3 cfg = {};
824 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
828 return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
832 hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
834 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
835 struct xt_hashlimit_htable *hinfo = info->hinfo;
837 return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
840 #define HASHLIMIT_MAX_SIZE 1048576
842 static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
843 struct xt_hashlimit_htable **hinfo,
844 struct hashlimit_cfg3 *cfg,
845 const char *name, int revision)
847 struct net *net = par->net;
850 if (cfg->gc_interval == 0 || cfg->expire == 0)
852 if (cfg->size > HASHLIMIT_MAX_SIZE) {
853 cfg->size = HASHLIMIT_MAX_SIZE;
854 pr_info_ratelimited("size too large, truncated to %u\n", cfg->size);
856 if (cfg->max > HASHLIMIT_MAX_SIZE) {
857 cfg->max = HASHLIMIT_MAX_SIZE;
858 pr_info_ratelimited("max too large, truncated to %u\n", cfg->max);
860 if (par->family == NFPROTO_IPV4) {
861 if (cfg->srcmask > 32 || cfg->dstmask > 32)
864 if (cfg->srcmask > 128 || cfg->dstmask > 128)
868 if (cfg->mode & ~XT_HASHLIMIT_ALL) {
869 pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
874 /* Check for overflow. */
875 if (revision >= 3 && cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
876 if (cfg->avg == 0 || cfg->avg > U32_MAX) {
877 pr_info_ratelimited("invalid rate\n");
881 if (cfg->interval == 0) {
882 pr_info_ratelimited("invalid interval\n");
885 } else if (cfg->mode & XT_HASHLIMIT_BYTES) {
886 if (user2credits_byte(cfg->avg) == 0) {
887 pr_info_ratelimited("overflow, rate too high: %llu\n",
891 } else if (cfg->burst == 0 ||
892 user2credits(cfg->avg * cfg->burst, revision) <
893 user2credits(cfg->avg, revision)) {
894 pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
895 cfg->avg, cfg->burst);
899 mutex_lock(&hashlimit_mutex);
900 *hinfo = htable_find_get(net, name, par->family);
901 if (*hinfo == NULL) {
902 ret = htable_create(net, cfg, name, par->family,
905 mutex_unlock(&hashlimit_mutex);
909 mutex_unlock(&hashlimit_mutex);
914 static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
916 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
917 struct hashlimit_cfg3 cfg = {};
920 ret = xt_check_proc_name(info->name, sizeof(info->name));
924 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
928 return hashlimit_mt_check_common(par, &info->hinfo,
929 &cfg, info->name, 1);
932 static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
934 struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
935 struct hashlimit_cfg3 cfg = {};
938 ret = xt_check_proc_name(info->name, sizeof(info->name));
942 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
946 return hashlimit_mt_check_common(par, &info->hinfo,
947 &cfg, info->name, 2);
950 static int hashlimit_mt_check(const struct xt_mtchk_param *par)
952 struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
955 ret = xt_check_proc_name(info->name, sizeof(info->name));
959 return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
963 static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
965 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
967 htable_put(info->hinfo);
970 static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
972 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
974 htable_put(info->hinfo);
977 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
979 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
981 htable_put(info->hinfo);
984 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
988 .family = NFPROTO_IPV4,
989 .match = hashlimit_mt_v1,
990 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
991 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
992 .checkentry = hashlimit_mt_check_v1,
993 .destroy = hashlimit_mt_destroy_v1,
999 .family = NFPROTO_IPV4,
1000 .match = hashlimit_mt_v2,
1001 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1002 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1003 .checkentry = hashlimit_mt_check_v2,
1004 .destroy = hashlimit_mt_destroy_v2,
1008 .name = "hashlimit",
1010 .family = NFPROTO_IPV4,
1011 .match = hashlimit_mt,
1012 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1013 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1014 .checkentry = hashlimit_mt_check,
1015 .destroy = hashlimit_mt_destroy,
1018 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1020 .name = "hashlimit",
1022 .family = NFPROTO_IPV6,
1023 .match = hashlimit_mt_v1,
1024 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
1025 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
1026 .checkentry = hashlimit_mt_check_v1,
1027 .destroy = hashlimit_mt_destroy_v1,
1031 .name = "hashlimit",
1033 .family = NFPROTO_IPV6,
1034 .match = hashlimit_mt_v2,
1035 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1036 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1037 .checkentry = hashlimit_mt_check_v2,
1038 .destroy = hashlimit_mt_destroy_v2,
1042 .name = "hashlimit",
1044 .family = NFPROTO_IPV6,
1045 .match = hashlimit_mt,
1046 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1047 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1048 .checkentry = hashlimit_mt_check,
1049 .destroy = hashlimit_mt_destroy,
1056 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1057 __acquires(htable->lock)
1059 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1060 unsigned int *bucket;
1062 spin_lock_bh(&htable->lock);
1063 if (*pos >= htable->cfg.size)
1066 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1068 return ERR_PTR(-ENOMEM);
1074 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1076 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1077 unsigned int *bucket = v;
1080 if (*pos >= htable->cfg.size) {
1087 static void dl_seq_stop(struct seq_file *s, void *v)
1088 __releases(htable->lock)
1090 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1091 unsigned int *bucket = v;
1093 if (!IS_ERR(bucket))
1095 spin_unlock_bh(&htable->lock);
1098 static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1103 seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1104 (long)(ent->expires - jiffies)/HZ,
1106 ntohs(ent->dst.src_port),
1108 ntohs(ent->dst.dst_port),
1109 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1110 ent->rateinfo.cost);
1112 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1114 seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1115 (long)(ent->expires - jiffies)/HZ,
1117 ntohs(ent->dst.src_port),
1119 ntohs(ent->dst.dst_port),
1120 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1121 ent->rateinfo.cost);
1129 static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1132 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1134 spin_lock(&ent->lock);
1135 /* recalculate to show accurate numbers */
1136 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1138 dl_seq_print(ent, family, s);
1140 spin_unlock(&ent->lock);
1141 return seq_has_overflowed(s);
1144 static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1147 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1149 spin_lock(&ent->lock);
1150 /* recalculate to show accurate numbers */
1151 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1153 dl_seq_print(ent, family, s);
1155 spin_unlock(&ent->lock);
1156 return seq_has_overflowed(s);
1159 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1162 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1164 spin_lock(&ent->lock);
1165 /* recalculate to show accurate numbers */
1166 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1168 dl_seq_print(ent, family, s);
1170 spin_unlock(&ent->lock);
1171 return seq_has_overflowed(s);
1174 static int dl_seq_show_v2(struct seq_file *s, void *v)
1176 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1177 unsigned int *bucket = (unsigned int *)v;
1178 struct dsthash_ent *ent;
1180 if (!hlist_empty(&htable->hash[*bucket])) {
1181 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1182 if (dl_seq_real_show_v2(ent, htable->family, s))
1188 static int dl_seq_show_v1(struct seq_file *s, void *v)
1190 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1191 unsigned int *bucket = v;
1192 struct dsthash_ent *ent;
1194 if (!hlist_empty(&htable->hash[*bucket])) {
1195 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1196 if (dl_seq_real_show_v1(ent, htable->family, s))
1202 static int dl_seq_show(struct seq_file *s, void *v)
1204 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1205 unsigned int *bucket = v;
1206 struct dsthash_ent *ent;
1208 if (!hlist_empty(&htable->hash[*bucket])) {
1209 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1210 if (dl_seq_real_show(ent, htable->family, s))
1216 static const struct seq_operations dl_seq_ops_v1 = {
1217 .start = dl_seq_start,
1218 .next = dl_seq_next,
1219 .stop = dl_seq_stop,
1220 .show = dl_seq_show_v1
1223 static const struct seq_operations dl_seq_ops_v2 = {
1224 .start = dl_seq_start,
1225 .next = dl_seq_next,
1226 .stop = dl_seq_stop,
1227 .show = dl_seq_show_v2
1230 static const struct seq_operations dl_seq_ops = {
1231 .start = dl_seq_start,
1232 .next = dl_seq_next,
1233 .stop = dl_seq_stop,
1237 static int __net_init hashlimit_proc_net_init(struct net *net)
1239 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1241 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1242 if (!hashlimit_net->ipt_hashlimit)
1244 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1245 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1246 if (!hashlimit_net->ip6t_hashlimit) {
1247 remove_proc_entry("ipt_hashlimit", net->proc_net);
1254 static void __net_exit hashlimit_proc_net_exit(struct net *net)
1256 struct xt_hashlimit_htable *hinfo;
1257 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1259 /* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1260 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1261 * entries is empty before trying to remove it.
1263 mutex_lock(&hashlimit_mutex);
1264 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1265 htable_remove_proc_entry(hinfo);
1266 hashlimit_net->ipt_hashlimit = NULL;
1267 hashlimit_net->ip6t_hashlimit = NULL;
1268 mutex_unlock(&hashlimit_mutex);
1270 remove_proc_entry("ipt_hashlimit", net->proc_net);
1271 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1272 remove_proc_entry("ip6t_hashlimit", net->proc_net);
1276 static int __net_init hashlimit_net_init(struct net *net)
1278 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1280 INIT_HLIST_HEAD(&hashlimit_net->htables);
1281 return hashlimit_proc_net_init(net);
1284 static void __net_exit hashlimit_net_exit(struct net *net)
1286 hashlimit_proc_net_exit(net);
1289 static struct pernet_operations hashlimit_net_ops = {
1290 .init = hashlimit_net_init,
1291 .exit = hashlimit_net_exit,
1292 .id = &hashlimit_net_id,
1293 .size = sizeof(struct hashlimit_net),
1296 static int __init hashlimit_mt_init(void)
1300 err = register_pernet_subsys(&hashlimit_net_ops);
1303 err = xt_register_matches(hashlimit_mt_reg,
1304 ARRAY_SIZE(hashlimit_mt_reg));
1309 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1310 sizeof(struct dsthash_ent), 0, 0,
1312 if (!hashlimit_cachep) {
1313 pr_warn("unable to create slab cache\n");
1319 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1321 unregister_pernet_subsys(&hashlimit_net_ops);
1326 static void __exit hashlimit_mt_exit(void)
1328 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1329 unregister_pernet_subsys(&hashlimit_net_ops);
1332 kmem_cache_destroy(hashlimit_cachep);
1335 module_init(hashlimit_mt_init);
1336 module_exit(hashlimit_mt_exit);