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/netfilter/xt_hashlimit.h>
38 #include <linux/mutex.h>
39 #include <linux/kernel.h>
41 MODULE_LICENSE("GPL");
42 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
43 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
44 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
45 MODULE_ALIAS("ipt_hashlimit");
46 MODULE_ALIAS("ip6t_hashlimit");
48 struct hashlimit_net {
49 struct hlist_head htables;
50 struct proc_dir_entry *ipt_hashlimit;
51 struct proc_dir_entry *ip6t_hashlimit;
54 static unsigned int hashlimit_net_id;
55 static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
57 return net_generic(net, hashlimit_net_id);
60 /* need to declare this at the top */
61 static const struct seq_operations dl_seq_ops_v2;
62 static const struct seq_operations dl_seq_ops_v1;
63 static const struct seq_operations dl_seq_ops;
72 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
84 /* static / read-only parts in the beginning */
85 struct hlist_node node;
86 struct dsthash_dst dst;
88 /* modified structure members in the end */
90 unsigned long expires; /* precalculated expiry time */
92 unsigned long prev; /* last modification */
100 u_int32_t interval, prev_window;
101 u_int64_t current_rate;
110 struct xt_hashlimit_htable {
111 struct hlist_node node; /* global list of all htables */
114 bool rnd_initialized;
116 struct hashlimit_cfg3 cfg; /* config */
118 /* used internally */
119 spinlock_t lock; /* lock for list_head */
120 u_int32_t rnd; /* random seed for hash */
121 unsigned int count; /* number entries in table */
122 struct delayed_work gc_work;
125 struct proc_dir_entry *pde;
129 struct hlist_head hash[0]; /* hashtable itself */
133 cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
136 struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
138 to->mode = cfg->mode;
140 to->burst = cfg->burst;
141 to->size = cfg->size;
143 to->gc_interval = cfg->gc_interval;
144 to->expire = cfg->expire;
145 to->srcmask = cfg->srcmask;
146 to->dstmask = cfg->dstmask;
147 } else if (revision == 2) {
148 struct hashlimit_cfg2 *cfg = (struct hashlimit_cfg2 *)from;
150 to->mode = cfg->mode;
152 to->burst = cfg->burst;
153 to->size = cfg->size;
155 to->gc_interval = cfg->gc_interval;
156 to->expire = cfg->expire;
157 to->srcmask = cfg->srcmask;
158 to->dstmask = cfg->dstmask;
159 } else if (revision == 3) {
160 memcpy(to, from, sizeof(struct hashlimit_cfg3));
168 static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */
169 static struct kmem_cache *hashlimit_cachep __read_mostly;
171 static inline bool dst_cmp(const struct dsthash_ent *ent,
172 const struct dsthash_dst *b)
174 return !memcmp(&ent->dst, b, sizeof(ent->dst));
178 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
180 u_int32_t hash = jhash2((const u32 *)dst,
181 sizeof(*dst)/sizeof(u32),
184 * Instead of returning hash % ht->cfg.size (implying a divide)
185 * we return the high 32 bits of the (hash * ht->cfg.size) that will
186 * give results between [0 and cfg.size-1] and same hash distribution,
187 * but using a multiply, less expensive than a divide
189 return reciprocal_scale(hash, ht->cfg.size);
192 static struct dsthash_ent *
193 dsthash_find(const struct xt_hashlimit_htable *ht,
194 const struct dsthash_dst *dst)
196 struct dsthash_ent *ent;
197 u_int32_t hash = hash_dst(ht, dst);
199 if (!hlist_empty(&ht->hash[hash])) {
200 hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
201 if (dst_cmp(ent, dst)) {
202 spin_lock(&ent->lock);
209 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
210 static struct dsthash_ent *
211 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
212 const struct dsthash_dst *dst, bool *race)
214 struct dsthash_ent *ent;
216 spin_lock(&ht->lock);
218 /* Two or more packets may race to create the same entry in the
219 * hashtable, double check if this packet lost race.
221 ent = dsthash_find(ht, dst);
223 spin_unlock(&ht->lock);
228 /* initialize hash with random val at the time we allocate
229 * the first hashtable entry */
230 if (unlikely(!ht->rnd_initialized)) {
231 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
232 ht->rnd_initialized = true;
235 if (ht->cfg.max && ht->count >= ht->cfg.max) {
236 /* FIXME: do something. question is what.. */
237 net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
240 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
242 memcpy(&ent->dst, dst, sizeof(ent->dst));
243 spin_lock_init(&ent->lock);
245 spin_lock(&ent->lock);
246 hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
249 spin_unlock(&ht->lock);
253 static void dsthash_free_rcu(struct rcu_head *head)
255 struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
257 kmem_cache_free(hashlimit_cachep, ent);
261 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
263 hlist_del_rcu(&ent->node);
264 call_rcu(&ent->rcu, dsthash_free_rcu);
267 static void htable_gc(struct work_struct *work);
269 static int htable_create(struct net *net, struct hashlimit_cfg3 *cfg,
270 const char *name, u_int8_t family,
271 struct xt_hashlimit_htable **out_hinfo,
274 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
275 struct xt_hashlimit_htable *hinfo;
276 const struct seq_operations *ops;
277 unsigned int size, i;
278 unsigned long nr_pages = totalram_pages();
284 size = (nr_pages << PAGE_SHIFT) / 16384 /
285 sizeof(struct hlist_head);
286 if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
291 /* FIXME: don't use vmalloc() here or anywhere else -HW */
292 hinfo = vmalloc(struct_size(hinfo, hash, size));
297 /* copy match config into hashtable config */
298 ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
304 hinfo->cfg.size = size;
305 if (hinfo->cfg.max == 0)
306 hinfo->cfg.max = 8 * hinfo->cfg.size;
307 else if (hinfo->cfg.max < hinfo->cfg.size)
308 hinfo->cfg.max = hinfo->cfg.size;
310 for (i = 0; i < hinfo->cfg.size; i++)
311 INIT_HLIST_HEAD(&hinfo->hash[i]);
315 hinfo->family = family;
316 hinfo->rnd_initialized = false;
317 hinfo->name = kstrdup(name, GFP_KERNEL);
322 spin_lock_init(&hinfo->lock);
326 ops = &dl_seq_ops_v1;
329 ops = &dl_seq_ops_v2;
335 hinfo->pde = proc_create_seq_data(name, 0,
336 (family == NFPROTO_IPV4) ?
337 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
339 if (hinfo->pde == NULL) {
346 INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
347 queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
348 msecs_to_jiffies(hinfo->cfg.gc_interval));
350 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
355 static bool select_all(const struct xt_hashlimit_htable *ht,
356 const struct dsthash_ent *he)
361 static bool select_gc(const struct xt_hashlimit_htable *ht,
362 const struct dsthash_ent *he)
364 return time_after_eq(jiffies, he->expires);
367 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
368 bool (*select)(const struct xt_hashlimit_htable *ht,
369 const struct dsthash_ent *he))
373 for (i = 0; i < ht->cfg.size; i++) {
374 struct dsthash_ent *dh;
375 struct hlist_node *n;
377 spin_lock_bh(&ht->lock);
378 hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
379 if ((*select)(ht, dh))
380 dsthash_free(ht, dh);
382 spin_unlock_bh(&ht->lock);
387 static void htable_gc(struct work_struct *work)
389 struct xt_hashlimit_htable *ht;
391 ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
393 htable_selective_cleanup(ht, select_gc);
395 queue_delayed_work(system_power_efficient_wq,
396 &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
399 static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
401 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
402 struct proc_dir_entry *parent;
404 if (hinfo->family == NFPROTO_IPV4)
405 parent = hashlimit_net->ipt_hashlimit;
407 parent = hashlimit_net->ip6t_hashlimit;
410 remove_proc_entry(hinfo->name, parent);
413 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
415 cancel_delayed_work_sync(&hinfo->gc_work);
416 htable_remove_proc_entry(hinfo);
417 htable_selective_cleanup(hinfo, select_all);
422 static struct xt_hashlimit_htable *htable_find_get(struct net *net,
426 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
427 struct xt_hashlimit_htable *hinfo;
429 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
430 if (!strcmp(name, hinfo->name) &&
431 hinfo->family == family) {
439 static void htable_put(struct xt_hashlimit_htable *hinfo)
441 mutex_lock(&hashlimit_mutex);
442 if (--hinfo->use == 0) {
443 hlist_del(&hinfo->node);
444 htable_destroy(hinfo);
446 mutex_unlock(&hashlimit_mutex);
449 /* The algorithm used is the Simple Token Bucket Filter (TBF)
450 * see net/sched/sch_tbf.c in the linux source tree
453 /* Rusty: This is my (non-mathematically-inclined) understanding of
454 this algorithm. The `average rate' in jiffies becomes your initial
455 amount of credit `credit' and the most credit you can ever have
456 `credit_cap'. The `peak rate' becomes the cost of passing the
459 `prev' tracks the last packet hit: you gain one credit per jiffy.
460 If you get credit balance more than this, the extra credit is
461 discarded. Every time the match passes, you lose `cost' credits;
462 if you don't have that many, the test fails.
464 See Alexey's formal explanation in net/sched/sch_tbf.c.
466 To get the maximum range, we multiply by this factor (ie. you get N
467 credits per jiffy). We want to allow a rate as low as 1 per day
468 (slowest userspace tool allows), which means
469 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
471 #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
472 #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
474 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
475 * us the power of 2 below the theoretical max, so GCC simply does a
477 #define _POW2_BELOW2(x) ((x)|((x)>>1))
478 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
479 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
480 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
481 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
482 #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
483 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
484 #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
486 #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
487 #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
489 /* in byte mode, the lowest possible rate is one packet/second.
490 * credit_cap is used as a counter that tells us how many times we can
491 * refill the "credits available" counter when it becomes empty.
493 #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
494 #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
496 static u32 xt_hashlimit_len_to_chunks(u32 len)
498 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
501 /* Precision saver. */
502 static u64 user2credits(u64 user, int revision)
504 u64 scale = (revision == 1) ?
505 XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
506 u64 cpj = (revision == 1) ?
507 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
509 /* Avoid overflow: divide the constant operands first */
510 if (scale >= HZ * cpj)
511 return div64_u64(user, div64_u64(scale, HZ * cpj));
513 return user * div64_u64(HZ * cpj, scale);
516 static u32 user2credits_byte(u32 user)
519 us *= HZ * CREDITS_PER_JIFFY_BYTES;
520 return (u32) (us >> 32);
523 static u64 user2rate(u64 user)
526 return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
528 pr_info_ratelimited("invalid rate from userspace: %llu\n",
534 static u64 user2rate_bytes(u32 user)
538 r = user ? U32_MAX / user : U32_MAX;
539 return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
542 static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
543 u32 mode, int revision)
545 unsigned long delta = now - dh->rateinfo.prev;
551 if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
552 u64 interval = dh->rateinfo.interval * HZ;
554 if (delta < interval)
557 dh->rateinfo.prev = now;
558 dh->rateinfo.prev_window =
559 ((dh->rateinfo.current_rate * interval) >
560 (delta * dh->rateinfo.rate));
561 dh->rateinfo.current_rate = 0;
566 dh->rateinfo.prev = now;
568 if (mode & XT_HASHLIMIT_BYTES) {
569 u64 tmp = dh->rateinfo.credit;
570 dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
571 cap = CREDITS_PER_JIFFY_BYTES * HZ;
572 if (tmp >= dh->rateinfo.credit) {/* overflow */
573 dh->rateinfo.credit = cap;
577 cpj = (revision == 1) ?
578 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
579 dh->rateinfo.credit += delta * cpj;
580 cap = dh->rateinfo.credit_cap;
582 if (dh->rateinfo.credit > cap)
583 dh->rateinfo.credit = cap;
586 static void rateinfo_init(struct dsthash_ent *dh,
587 struct xt_hashlimit_htable *hinfo, int revision)
589 dh->rateinfo.prev = jiffies;
590 if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
591 dh->rateinfo.prev_window = 0;
592 dh->rateinfo.current_rate = 0;
593 if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
595 user2rate_bytes((u32)hinfo->cfg.avg);
596 if (hinfo->cfg.burst)
598 hinfo->cfg.burst * dh->rateinfo.rate;
600 dh->rateinfo.burst = dh->rateinfo.rate;
602 dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
604 hinfo->cfg.burst + dh->rateinfo.rate;
606 dh->rateinfo.interval = hinfo->cfg.interval;
607 } else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
608 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
609 dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
610 dh->rateinfo.credit_cap = hinfo->cfg.burst;
612 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
613 hinfo->cfg.burst, revision);
614 dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
615 dh->rateinfo.credit_cap = dh->rateinfo.credit;
619 static inline __be32 maskl(__be32 a, unsigned int l)
621 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
624 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
625 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
629 i[0] = maskl(i[0], p);
630 i[1] = i[2] = i[3] = 0;
633 i[1] = maskl(i[1], p - 32);
637 i[2] = maskl(i[2], p - 64);
641 i[3] = maskl(i[3], p - 96);
650 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
651 struct dsthash_dst *dst,
652 const struct sk_buff *skb, unsigned int protoff)
654 __be16 _ports[2], *ports;
658 memset(dst, 0, sizeof(*dst));
660 switch (hinfo->family) {
662 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
663 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
665 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
666 dst->ip.src = maskl(ip_hdr(skb)->saddr,
669 if (!(hinfo->cfg.mode &
670 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
672 nexthdr = ip_hdr(skb)->protocol;
674 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
679 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
680 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
681 sizeof(dst->ip6.dst));
682 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
684 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
685 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
686 sizeof(dst->ip6.src));
687 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
690 if (!(hinfo->cfg.mode &
691 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
693 nexthdr = ipv6_hdr(skb)->nexthdr;
694 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
695 if ((int)protoff < 0)
705 poff = proto_ports_offset(nexthdr);
707 ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
710 _ports[0] = _ports[1] = 0;
715 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
716 dst->src_port = ports[0];
717 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
718 dst->dst_port = ports[1];
722 static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
724 u64 tmp = xt_hashlimit_len_to_chunks(len);
725 tmp = tmp * dh->rateinfo.cost;
727 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
728 tmp = CREDITS_PER_JIFFY_BYTES * HZ;
730 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
731 dh->rateinfo.credit_cap--;
732 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
738 hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
739 struct xt_hashlimit_htable *hinfo,
740 const struct hashlimit_cfg3 *cfg, int revision)
742 unsigned long now = jiffies;
743 struct dsthash_ent *dh;
744 struct dsthash_dst dst;
748 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
752 dh = dsthash_find(hinfo, &dst);
754 dh = dsthash_alloc_init(hinfo, &dst, &race);
759 /* Already got an entry, update expiration timeout */
760 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
761 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
763 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
764 rateinfo_init(dh, hinfo, revision);
767 /* update expiration timeout */
768 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
769 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
772 if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
773 cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
774 dh->rateinfo.current_rate += cost;
776 if (!dh->rateinfo.prev_window &&
777 (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
778 spin_unlock(&dh->lock);
780 return !(cfg->mode & XT_HASHLIMIT_INVERT);
786 if (cfg->mode & XT_HASHLIMIT_BYTES)
787 cost = hashlimit_byte_cost(skb->len, dh);
789 cost = dh->rateinfo.cost;
791 if (dh->rateinfo.credit >= cost) {
792 /* below the limit */
793 dh->rateinfo.credit -= cost;
794 spin_unlock(&dh->lock);
796 return !(cfg->mode & XT_HASHLIMIT_INVERT);
800 spin_unlock(&dh->lock);
802 /* default match is underlimit - so over the limit, we need to invert */
803 return cfg->mode & XT_HASHLIMIT_INVERT;
811 hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
813 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
814 struct xt_hashlimit_htable *hinfo = info->hinfo;
815 struct hashlimit_cfg3 cfg = {};
818 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
822 return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
826 hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
828 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
829 struct xt_hashlimit_htable *hinfo = info->hinfo;
830 struct hashlimit_cfg3 cfg = {};
833 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
837 return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
841 hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
843 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
844 struct xt_hashlimit_htable *hinfo = info->hinfo;
846 return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
849 static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
850 struct xt_hashlimit_htable **hinfo,
851 struct hashlimit_cfg3 *cfg,
852 const char *name, int revision)
854 struct net *net = par->net;
857 if (cfg->gc_interval == 0 || cfg->expire == 0)
859 if (par->family == NFPROTO_IPV4) {
860 if (cfg->srcmask > 32 || cfg->dstmask > 32)
863 if (cfg->srcmask > 128 || cfg->dstmask > 128)
867 if (cfg->mode & ~XT_HASHLIMIT_ALL) {
868 pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
873 /* Check for overflow. */
874 if (revision >= 3 && cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
875 if (cfg->avg == 0 || cfg->avg > U32_MAX) {
876 pr_info_ratelimited("invalid rate\n");
880 if (cfg->interval == 0) {
881 pr_info_ratelimited("invalid interval\n");
884 } else if (cfg->mode & XT_HASHLIMIT_BYTES) {
885 if (user2credits_byte(cfg->avg) == 0) {
886 pr_info_ratelimited("overflow, rate too high: %llu\n",
890 } else if (cfg->burst == 0 ||
891 user2credits(cfg->avg * cfg->burst, revision) <
892 user2credits(cfg->avg, revision)) {
893 pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
894 cfg->avg, cfg->burst);
898 mutex_lock(&hashlimit_mutex);
899 *hinfo = htable_find_get(net, name, par->family);
900 if (*hinfo == NULL) {
901 ret = htable_create(net, cfg, name, par->family,
904 mutex_unlock(&hashlimit_mutex);
908 mutex_unlock(&hashlimit_mutex);
913 static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
915 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
916 struct hashlimit_cfg3 cfg = {};
919 ret = xt_check_proc_name(info->name, sizeof(info->name));
923 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
927 return hashlimit_mt_check_common(par, &info->hinfo,
928 &cfg, info->name, 1);
931 static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
933 struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
934 struct hashlimit_cfg3 cfg = {};
937 ret = xt_check_proc_name(info->name, sizeof(info->name));
941 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
945 return hashlimit_mt_check_common(par, &info->hinfo,
946 &cfg, info->name, 2);
949 static int hashlimit_mt_check(const struct xt_mtchk_param *par)
951 struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
954 ret = xt_check_proc_name(info->name, sizeof(info->name));
958 return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
962 static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
964 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
966 htable_put(info->hinfo);
969 static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
971 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
973 htable_put(info->hinfo);
976 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
978 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
980 htable_put(info->hinfo);
983 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
987 .family = NFPROTO_IPV4,
988 .match = hashlimit_mt_v1,
989 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
990 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
991 .checkentry = hashlimit_mt_check_v1,
992 .destroy = hashlimit_mt_destroy_v1,
998 .family = NFPROTO_IPV4,
999 .match = hashlimit_mt_v2,
1000 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1001 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1002 .checkentry = hashlimit_mt_check_v2,
1003 .destroy = hashlimit_mt_destroy_v2,
1007 .name = "hashlimit",
1009 .family = NFPROTO_IPV4,
1010 .match = hashlimit_mt,
1011 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1012 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1013 .checkentry = hashlimit_mt_check,
1014 .destroy = hashlimit_mt_destroy,
1017 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1019 .name = "hashlimit",
1021 .family = NFPROTO_IPV6,
1022 .match = hashlimit_mt_v1,
1023 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
1024 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
1025 .checkentry = hashlimit_mt_check_v1,
1026 .destroy = hashlimit_mt_destroy_v1,
1030 .name = "hashlimit",
1032 .family = NFPROTO_IPV6,
1033 .match = hashlimit_mt_v2,
1034 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1035 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1036 .checkentry = hashlimit_mt_check_v2,
1037 .destroy = hashlimit_mt_destroy_v2,
1041 .name = "hashlimit",
1043 .family = NFPROTO_IPV6,
1044 .match = hashlimit_mt,
1045 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1046 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1047 .checkentry = hashlimit_mt_check,
1048 .destroy = hashlimit_mt_destroy,
1055 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1056 __acquires(htable->lock)
1058 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1059 unsigned int *bucket;
1061 spin_lock_bh(&htable->lock);
1062 if (*pos >= htable->cfg.size)
1065 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1067 return ERR_PTR(-ENOMEM);
1073 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1075 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1076 unsigned int *bucket = v;
1079 if (*pos >= htable->cfg.size) {
1086 static void dl_seq_stop(struct seq_file *s, void *v)
1087 __releases(htable->lock)
1089 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1090 unsigned int *bucket = v;
1092 if (!IS_ERR(bucket))
1094 spin_unlock_bh(&htable->lock);
1097 static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1102 seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1103 (long)(ent->expires - jiffies)/HZ,
1105 ntohs(ent->dst.src_port),
1107 ntohs(ent->dst.dst_port),
1108 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1109 ent->rateinfo.cost);
1111 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1113 seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1114 (long)(ent->expires - jiffies)/HZ,
1116 ntohs(ent->dst.src_port),
1118 ntohs(ent->dst.dst_port),
1119 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1120 ent->rateinfo.cost);
1128 static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1131 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1133 spin_lock(&ent->lock);
1134 /* recalculate to show accurate numbers */
1135 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1137 dl_seq_print(ent, family, s);
1139 spin_unlock(&ent->lock);
1140 return seq_has_overflowed(s);
1143 static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1146 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1148 spin_lock(&ent->lock);
1149 /* recalculate to show accurate numbers */
1150 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1152 dl_seq_print(ent, family, s);
1154 spin_unlock(&ent->lock);
1155 return seq_has_overflowed(s);
1158 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1161 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1163 spin_lock(&ent->lock);
1164 /* recalculate to show accurate numbers */
1165 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1167 dl_seq_print(ent, family, s);
1169 spin_unlock(&ent->lock);
1170 return seq_has_overflowed(s);
1173 static int dl_seq_show_v2(struct seq_file *s, void *v)
1175 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1176 unsigned int *bucket = (unsigned int *)v;
1177 struct dsthash_ent *ent;
1179 if (!hlist_empty(&htable->hash[*bucket])) {
1180 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1181 if (dl_seq_real_show_v2(ent, htable->family, s))
1187 static int dl_seq_show_v1(struct seq_file *s, void *v)
1189 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1190 unsigned int *bucket = v;
1191 struct dsthash_ent *ent;
1193 if (!hlist_empty(&htable->hash[*bucket])) {
1194 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1195 if (dl_seq_real_show_v1(ent, htable->family, s))
1201 static int dl_seq_show(struct seq_file *s, void *v)
1203 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1204 unsigned int *bucket = v;
1205 struct dsthash_ent *ent;
1207 if (!hlist_empty(&htable->hash[*bucket])) {
1208 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1209 if (dl_seq_real_show(ent, htable->family, s))
1215 static const struct seq_operations dl_seq_ops_v1 = {
1216 .start = dl_seq_start,
1217 .next = dl_seq_next,
1218 .stop = dl_seq_stop,
1219 .show = dl_seq_show_v1
1222 static const struct seq_operations dl_seq_ops_v2 = {
1223 .start = dl_seq_start,
1224 .next = dl_seq_next,
1225 .stop = dl_seq_stop,
1226 .show = dl_seq_show_v2
1229 static const struct seq_operations dl_seq_ops = {
1230 .start = dl_seq_start,
1231 .next = dl_seq_next,
1232 .stop = dl_seq_stop,
1236 static int __net_init hashlimit_proc_net_init(struct net *net)
1238 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1240 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1241 if (!hashlimit_net->ipt_hashlimit)
1243 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1244 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1245 if (!hashlimit_net->ip6t_hashlimit) {
1246 remove_proc_entry("ipt_hashlimit", net->proc_net);
1253 static void __net_exit hashlimit_proc_net_exit(struct net *net)
1255 struct xt_hashlimit_htable *hinfo;
1256 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1258 /* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1259 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1260 * entries is empty before trying to remove it.
1262 mutex_lock(&hashlimit_mutex);
1263 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1264 htable_remove_proc_entry(hinfo);
1265 hashlimit_net->ipt_hashlimit = NULL;
1266 hashlimit_net->ip6t_hashlimit = NULL;
1267 mutex_unlock(&hashlimit_mutex);
1269 remove_proc_entry("ipt_hashlimit", net->proc_net);
1270 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1271 remove_proc_entry("ip6t_hashlimit", net->proc_net);
1275 static int __net_init hashlimit_net_init(struct net *net)
1277 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1279 INIT_HLIST_HEAD(&hashlimit_net->htables);
1280 return hashlimit_proc_net_init(net);
1283 static void __net_exit hashlimit_net_exit(struct net *net)
1285 hashlimit_proc_net_exit(net);
1288 static struct pernet_operations hashlimit_net_ops = {
1289 .init = hashlimit_net_init,
1290 .exit = hashlimit_net_exit,
1291 .id = &hashlimit_net_id,
1292 .size = sizeof(struct hashlimit_net),
1295 static int __init hashlimit_mt_init(void)
1299 err = register_pernet_subsys(&hashlimit_net_ops);
1302 err = xt_register_matches(hashlimit_mt_reg,
1303 ARRAY_SIZE(hashlimit_mt_reg));
1308 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1309 sizeof(struct dsthash_ent), 0, 0,
1311 if (!hashlimit_cachep) {
1312 pr_warn("unable to create slab cache\n");
1318 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1320 unregister_pernet_subsys(&hashlimit_net_ops);
1325 static void __exit hashlimit_mt_exit(void)
1327 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1328 unregister_pernet_subsys(&hashlimit_net_ops);
1331 kmem_cache_destroy(hashlimit_cachep);
1334 module_init(hashlimit_mt_init);
1335 module_exit(hashlimit_mt_exit);