Merge tag 'io_uring-5.11-2021-01-16' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / net / netfilter / xt_hashlimit.c
blob9c5cfd74a0ee488e9e996c8892cf9e08ddcdc1c3
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
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>
23 #include <linux/mm.h>
24 #include <linux/in.h>
25 #include <linux/ip.h>
26 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
27 #include <linux/ipv6.h>
28 #include <net/ipv6.h>
29 #endif
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;
71 /* hash table crap */
72 struct dsthash_dst {
73 union {
74 struct {
75 __be32 src;
76 __be32 dst;
77 } ip;
78 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
79 struct {
80 __be32 src[4];
81 __be32 dst[4];
82 } ip6;
83 #endif
85 __be16 src_port;
86 __be16 dst_port;
89 struct dsthash_ent {
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 */
95 spinlock_t lock;
96 unsigned long expires; /* precalculated expiry time */
97 struct {
98 unsigned long prev; /* last modification */
99 union {
100 struct {
101 u_int64_t credit;
102 u_int64_t credit_cap;
103 u_int64_t cost;
105 struct {
106 u_int32_t interval, prev_window;
107 u_int64_t current_rate;
108 u_int64_t rate;
109 int64_t burst;
112 } rateinfo;
113 struct rcu_head rcu;
116 struct xt_hashlimit_htable {
117 struct hlist_node node; /* global list of all htables */
118 refcount_t use;
119 u_int8_t family;
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;
130 /* seq_file stuff */
131 struct proc_dir_entry *pde;
132 const char *name;
133 struct net *net;
135 struct hlist_head hash[]; /* hashtable itself */
138 static int
139 cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
141 if (revision == 1) {
142 struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
144 to->mode = cfg->mode;
145 to->avg = cfg->avg;
146 to->burst = cfg->burst;
147 to->size = cfg->size;
148 to->max = cfg->max;
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;
157 to->avg = cfg->avg;
158 to->burst = cfg->burst;
159 to->size = cfg->size;
160 to->max = cfg->max;
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));
167 } else {
168 return -EINVAL;
171 return 0;
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));
183 static u_int32_t
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),
188 ht->rnd);
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);
209 return ent;
212 return NULL;
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);
228 if (ent != NULL) {
229 spin_unlock(&ht->lock);
230 *race = true;
231 return ent;
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);
244 ent = NULL;
245 } else
246 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
247 if (ent) {
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)]);
253 ht->count++;
255 spin_unlock(&ht->lock);
256 return ent;
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);
266 static inline void
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);
271 ht->count--;
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,
278 int revision)
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();
285 int ret;
287 if (cfg->size) {
288 size = cfg->size;
289 } else {
290 size = (nr_pages << PAGE_SHIFT) / 16384 /
291 sizeof(struct hlist_head);
292 if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
293 size = 8192;
294 if (size < 16)
295 size = 16;
297 /* FIXME: don't use vmalloc() here or anywhere else -HW */
298 hinfo = vmalloc(struct_size(hinfo, hash, size));
299 if (hinfo == NULL)
300 return -ENOMEM;
301 *out_hinfo = hinfo;
303 /* copy match config into hashtable config */
304 ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
305 if (ret) {
306 vfree(hinfo);
307 return ret;
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);
320 hinfo->count = 0;
321 hinfo->family = family;
322 hinfo->rnd_initialized = false;
323 hinfo->name = kstrdup(name, GFP_KERNEL);
324 if (!hinfo->name) {
325 vfree(hinfo);
326 return -ENOMEM;
328 spin_lock_init(&hinfo->lock);
330 switch (revision) {
331 case 1:
332 ops = &dl_seq_ops_v1;
333 break;
334 case 2:
335 ops = &dl_seq_ops_v2;
336 break;
337 default:
338 ops = &dl_seq_ops;
341 hinfo->pde = proc_create_seq_data(name, 0,
342 (family == NFPROTO_IPV4) ?
343 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
344 ops, hinfo);
345 if (hinfo->pde == NULL) {
346 kfree(hinfo->name);
347 vfree(hinfo);
348 return -ENOMEM;
350 hinfo->net = net;
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);
358 return 0;
361 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, bool select_all)
363 unsigned int i;
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);
375 cond_resched();
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;
398 else
399 parent = hashlimit_net->ip6t_hashlimit;
401 if (parent != NULL)
402 remove_proc_entry(hinfo->name, parent);
405 static struct xt_hashlimit_htable *htable_find_get(struct net *net,
406 const char *name,
407 u_int8_t family)
409 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
410 struct xt_hashlimit_htable *hinfo;
412 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
413 if (!strcmp(name, hinfo->name) &&
414 hinfo->family == family) {
415 refcount_inc(&hinfo->use);
416 return hinfo;
419 return NULL;
422 static void htable_put(struct xt_hashlimit_htable *hinfo)
424 if (refcount_dec_and_mutex_lock(&hinfo->use, &hashlimit_mutex)) {
425 hlist_del(&hinfo->node);
426 htable_remove_proc_entry(hinfo);
427 mutex_unlock(&hashlimit_mutex);
429 cancel_delayed_work_sync(&hinfo->gc_work);
430 htable_selective_cleanup(hinfo, true);
431 kfree(hinfo->name);
432 vfree(hinfo);
436 /* The algorithm used is the Simple Token Bucket Filter (TBF)
437 * see net/sched/sch_tbf.c in the linux source tree
440 /* Rusty: This is my (non-mathematically-inclined) understanding of
441 this algorithm. The `average rate' in jiffies becomes your initial
442 amount of credit `credit' and the most credit you can ever have
443 `credit_cap'. The `peak rate' becomes the cost of passing the
444 test, `cost'.
446 `prev' tracks the last packet hit: you gain one credit per jiffy.
447 If you get credit balance more than this, the extra credit is
448 discarded. Every time the match passes, you lose `cost' credits;
449 if you don't have that many, the test fails.
451 See Alexey's formal explanation in net/sched/sch_tbf.c.
453 To get the maximum range, we multiply by this factor (ie. you get N
454 credits per jiffy). We want to allow a rate as low as 1 per day
455 (slowest userspace tool allows), which means
456 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
458 #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
459 #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
461 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
462 * us the power of 2 below the theoretical max, so GCC simply does a
463 * shift. */
464 #define _POW2_BELOW2(x) ((x)|((x)>>1))
465 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
466 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
467 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
468 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
469 #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
470 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
471 #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
473 #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
474 #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
476 /* in byte mode, the lowest possible rate is one packet/second.
477 * credit_cap is used as a counter that tells us how many times we can
478 * refill the "credits available" counter when it becomes empty.
480 #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
481 #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
483 static u32 xt_hashlimit_len_to_chunks(u32 len)
485 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
488 /* Precision saver. */
489 static u64 user2credits(u64 user, int revision)
491 u64 scale = (revision == 1) ?
492 XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
493 u64 cpj = (revision == 1) ?
494 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
496 /* Avoid overflow: divide the constant operands first */
497 if (scale >= HZ * cpj)
498 return div64_u64(user, div64_u64(scale, HZ * cpj));
500 return user * div64_u64(HZ * cpj, scale);
503 static u32 user2credits_byte(u32 user)
505 u64 us = user;
506 us *= HZ * CREDITS_PER_JIFFY_BYTES;
507 return (u32) (us >> 32);
510 static u64 user2rate(u64 user)
512 if (user != 0) {
513 return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
514 } else {
515 pr_info_ratelimited("invalid rate from userspace: %llu\n",
516 user);
517 return 0;
521 static u64 user2rate_bytes(u32 user)
523 u64 r;
525 r = user ? U32_MAX / user : U32_MAX;
526 return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
529 static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
530 u32 mode, int revision)
532 unsigned long delta = now - dh->rateinfo.prev;
533 u64 cap, cpj;
535 if (delta == 0)
536 return;
538 if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
539 u64 interval = dh->rateinfo.interval * HZ;
541 if (delta < interval)
542 return;
544 dh->rateinfo.prev = now;
545 dh->rateinfo.prev_window =
546 ((dh->rateinfo.current_rate * interval) >
547 (delta * dh->rateinfo.rate));
548 dh->rateinfo.current_rate = 0;
550 return;
553 dh->rateinfo.prev = now;
555 if (mode & XT_HASHLIMIT_BYTES) {
556 u64 tmp = dh->rateinfo.credit;
557 dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
558 cap = CREDITS_PER_JIFFY_BYTES * HZ;
559 if (tmp >= dh->rateinfo.credit) {/* overflow */
560 dh->rateinfo.credit = cap;
561 return;
563 } else {
564 cpj = (revision == 1) ?
565 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
566 dh->rateinfo.credit += delta * cpj;
567 cap = dh->rateinfo.credit_cap;
569 if (dh->rateinfo.credit > cap)
570 dh->rateinfo.credit = cap;
573 static void rateinfo_init(struct dsthash_ent *dh,
574 struct xt_hashlimit_htable *hinfo, int revision)
576 dh->rateinfo.prev = jiffies;
577 if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
578 dh->rateinfo.prev_window = 0;
579 dh->rateinfo.current_rate = 0;
580 if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
581 dh->rateinfo.rate =
582 user2rate_bytes((u32)hinfo->cfg.avg);
583 if (hinfo->cfg.burst)
584 dh->rateinfo.burst =
585 hinfo->cfg.burst * dh->rateinfo.rate;
586 else
587 dh->rateinfo.burst = dh->rateinfo.rate;
588 } else {
589 dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
590 dh->rateinfo.burst =
591 hinfo->cfg.burst + dh->rateinfo.rate;
593 dh->rateinfo.interval = hinfo->cfg.interval;
594 } else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
595 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
596 dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
597 dh->rateinfo.credit_cap = hinfo->cfg.burst;
598 } else {
599 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
600 hinfo->cfg.burst, revision);
601 dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
602 dh->rateinfo.credit_cap = dh->rateinfo.credit;
606 static inline __be32 maskl(__be32 a, unsigned int l)
608 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
611 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
612 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
614 switch (p) {
615 case 0 ... 31:
616 i[0] = maskl(i[0], p);
617 i[1] = i[2] = i[3] = 0;
618 break;
619 case 32 ... 63:
620 i[1] = maskl(i[1], p - 32);
621 i[2] = i[3] = 0;
622 break;
623 case 64 ... 95:
624 i[2] = maskl(i[2], p - 64);
625 i[3] = 0;
626 break;
627 case 96 ... 127:
628 i[3] = maskl(i[3], p - 96);
629 break;
630 case 128:
631 break;
634 #endif
636 static int
637 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
638 struct dsthash_dst *dst,
639 const struct sk_buff *skb, unsigned int protoff)
641 __be16 _ports[2], *ports;
642 u8 nexthdr;
643 int poff;
645 memset(dst, 0, sizeof(*dst));
647 switch (hinfo->family) {
648 case NFPROTO_IPV4:
649 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
650 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
651 hinfo->cfg.dstmask);
652 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
653 dst->ip.src = maskl(ip_hdr(skb)->saddr,
654 hinfo->cfg.srcmask);
656 if (!(hinfo->cfg.mode &
657 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
658 return 0;
659 nexthdr = ip_hdr(skb)->protocol;
660 break;
661 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
662 case NFPROTO_IPV6:
664 __be16 frag_off;
666 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
667 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
668 sizeof(dst->ip6.dst));
669 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
671 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
672 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
673 sizeof(dst->ip6.src));
674 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
677 if (!(hinfo->cfg.mode &
678 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
679 return 0;
680 nexthdr = ipv6_hdr(skb)->nexthdr;
681 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
682 if ((int)protoff < 0)
683 return -1;
684 break;
686 #endif
687 default:
688 BUG();
689 return 0;
692 poff = proto_ports_offset(nexthdr);
693 if (poff >= 0) {
694 ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
695 &_ports);
696 } else {
697 _ports[0] = _ports[1] = 0;
698 ports = _ports;
700 if (!ports)
701 return -1;
702 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
703 dst->src_port = ports[0];
704 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
705 dst->dst_port = ports[1];
706 return 0;
709 static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
711 u64 tmp = xt_hashlimit_len_to_chunks(len);
712 tmp = tmp * dh->rateinfo.cost;
714 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
715 tmp = CREDITS_PER_JIFFY_BYTES * HZ;
717 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
718 dh->rateinfo.credit_cap--;
719 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
721 return (u32) tmp;
724 static bool
725 hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
726 struct xt_hashlimit_htable *hinfo,
727 const struct hashlimit_cfg3 *cfg, int revision)
729 unsigned long now = jiffies;
730 struct dsthash_ent *dh;
731 struct dsthash_dst dst;
732 bool race = false;
733 u64 cost;
735 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
736 goto hotdrop;
738 local_bh_disable();
739 dh = dsthash_find(hinfo, &dst);
740 if (dh == NULL) {
741 dh = dsthash_alloc_init(hinfo, &dst, &race);
742 if (dh == NULL) {
743 local_bh_enable();
744 goto hotdrop;
745 } else if (race) {
746 /* Already got an entry, update expiration timeout */
747 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
748 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
749 } else {
750 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
751 rateinfo_init(dh, hinfo, revision);
753 } else {
754 /* update expiration timeout */
755 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
756 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
759 if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
760 cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
761 dh->rateinfo.current_rate += cost;
763 if (!dh->rateinfo.prev_window &&
764 (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
765 spin_unlock(&dh->lock);
766 local_bh_enable();
767 return !(cfg->mode & XT_HASHLIMIT_INVERT);
768 } else {
769 goto overlimit;
773 if (cfg->mode & XT_HASHLIMIT_BYTES)
774 cost = hashlimit_byte_cost(skb->len, dh);
775 else
776 cost = dh->rateinfo.cost;
778 if (dh->rateinfo.credit >= cost) {
779 /* below the limit */
780 dh->rateinfo.credit -= cost;
781 spin_unlock(&dh->lock);
782 local_bh_enable();
783 return !(cfg->mode & XT_HASHLIMIT_INVERT);
786 overlimit:
787 spin_unlock(&dh->lock);
788 local_bh_enable();
789 /* default match is underlimit - so over the limit, we need to invert */
790 return cfg->mode & XT_HASHLIMIT_INVERT;
792 hotdrop:
793 par->hotdrop = true;
794 return false;
797 static bool
798 hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
800 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
801 struct xt_hashlimit_htable *hinfo = info->hinfo;
802 struct hashlimit_cfg3 cfg = {};
803 int ret;
805 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
806 if (ret)
807 return ret;
809 return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
812 static bool
813 hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
815 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
816 struct xt_hashlimit_htable *hinfo = info->hinfo;
817 struct hashlimit_cfg3 cfg = {};
818 int ret;
820 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
821 if (ret)
822 return ret;
824 return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
827 static bool
828 hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
830 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
831 struct xt_hashlimit_htable *hinfo = info->hinfo;
833 return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
836 #define HASHLIMIT_MAX_SIZE 1048576
838 static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
839 struct xt_hashlimit_htable **hinfo,
840 struct hashlimit_cfg3 *cfg,
841 const char *name, int revision)
843 struct net *net = par->net;
844 int ret;
846 if (cfg->gc_interval == 0 || cfg->expire == 0)
847 return -EINVAL;
848 if (cfg->size > HASHLIMIT_MAX_SIZE) {
849 cfg->size = HASHLIMIT_MAX_SIZE;
850 pr_info_ratelimited("size too large, truncated to %u\n", cfg->size);
852 if (cfg->max > HASHLIMIT_MAX_SIZE) {
853 cfg->max = HASHLIMIT_MAX_SIZE;
854 pr_info_ratelimited("max too large, truncated to %u\n", cfg->max);
856 if (par->family == NFPROTO_IPV4) {
857 if (cfg->srcmask > 32 || cfg->dstmask > 32)
858 return -EINVAL;
859 } else {
860 if (cfg->srcmask > 128 || cfg->dstmask > 128)
861 return -EINVAL;
864 if (cfg->mode & ~XT_HASHLIMIT_ALL) {
865 pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
866 cfg->mode);
867 return -EINVAL;
870 /* Check for overflow. */
871 if (revision >= 3 && cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
872 if (cfg->avg == 0 || cfg->avg > U32_MAX) {
873 pr_info_ratelimited("invalid rate\n");
874 return -ERANGE;
877 if (cfg->interval == 0) {
878 pr_info_ratelimited("invalid interval\n");
879 return -EINVAL;
881 } else if (cfg->mode & XT_HASHLIMIT_BYTES) {
882 if (user2credits_byte(cfg->avg) == 0) {
883 pr_info_ratelimited("overflow, rate too high: %llu\n",
884 cfg->avg);
885 return -EINVAL;
887 } else if (cfg->burst == 0 ||
888 user2credits(cfg->avg * cfg->burst, revision) <
889 user2credits(cfg->avg, revision)) {
890 pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
891 cfg->avg, cfg->burst);
892 return -ERANGE;
895 mutex_lock(&hashlimit_mutex);
896 *hinfo = htable_find_get(net, name, par->family);
897 if (*hinfo == NULL) {
898 ret = htable_create(net, cfg, name, par->family,
899 hinfo, revision);
900 if (ret < 0) {
901 mutex_unlock(&hashlimit_mutex);
902 return ret;
905 mutex_unlock(&hashlimit_mutex);
907 return 0;
910 static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
912 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
913 struct hashlimit_cfg3 cfg = {};
914 int ret;
916 ret = xt_check_proc_name(info->name, sizeof(info->name));
917 if (ret)
918 return ret;
920 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
921 if (ret)
922 return ret;
924 return hashlimit_mt_check_common(par, &info->hinfo,
925 &cfg, info->name, 1);
928 static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
930 struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
931 struct hashlimit_cfg3 cfg = {};
932 int ret;
934 ret = xt_check_proc_name(info->name, sizeof(info->name));
935 if (ret)
936 return ret;
938 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
939 if (ret)
940 return ret;
942 return hashlimit_mt_check_common(par, &info->hinfo,
943 &cfg, info->name, 2);
946 static int hashlimit_mt_check(const struct xt_mtchk_param *par)
948 struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
949 int ret;
951 ret = xt_check_proc_name(info->name, sizeof(info->name));
952 if (ret)
953 return ret;
955 return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
956 info->name, 3);
959 static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
961 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
963 htable_put(info->hinfo);
966 static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
968 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
970 htable_put(info->hinfo);
973 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
975 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
977 htable_put(info->hinfo);
980 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
982 .name = "hashlimit",
983 .revision = 1,
984 .family = NFPROTO_IPV4,
985 .match = hashlimit_mt_v1,
986 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
987 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
988 .checkentry = hashlimit_mt_check_v1,
989 .destroy = hashlimit_mt_destroy_v1,
990 .me = THIS_MODULE,
993 .name = "hashlimit",
994 .revision = 2,
995 .family = NFPROTO_IPV4,
996 .match = hashlimit_mt_v2,
997 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
998 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
999 .checkentry = hashlimit_mt_check_v2,
1000 .destroy = hashlimit_mt_destroy_v2,
1001 .me = THIS_MODULE,
1004 .name = "hashlimit",
1005 .revision = 3,
1006 .family = NFPROTO_IPV4,
1007 .match = hashlimit_mt,
1008 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1009 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1010 .checkentry = hashlimit_mt_check,
1011 .destroy = hashlimit_mt_destroy,
1012 .me = THIS_MODULE,
1014 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1016 .name = "hashlimit",
1017 .revision = 1,
1018 .family = NFPROTO_IPV6,
1019 .match = hashlimit_mt_v1,
1020 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
1021 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
1022 .checkentry = hashlimit_mt_check_v1,
1023 .destroy = hashlimit_mt_destroy_v1,
1024 .me = THIS_MODULE,
1027 .name = "hashlimit",
1028 .revision = 2,
1029 .family = NFPROTO_IPV6,
1030 .match = hashlimit_mt_v2,
1031 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1032 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1033 .checkentry = hashlimit_mt_check_v2,
1034 .destroy = hashlimit_mt_destroy_v2,
1035 .me = THIS_MODULE,
1038 .name = "hashlimit",
1039 .revision = 3,
1040 .family = NFPROTO_IPV6,
1041 .match = hashlimit_mt,
1042 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1043 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1044 .checkentry = hashlimit_mt_check,
1045 .destroy = hashlimit_mt_destroy,
1046 .me = THIS_MODULE,
1048 #endif
1051 /* PROC stuff */
1052 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1053 __acquires(htable->lock)
1055 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1056 unsigned int *bucket;
1058 spin_lock_bh(&htable->lock);
1059 if (*pos >= htable->cfg.size)
1060 return NULL;
1062 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1063 if (!bucket)
1064 return ERR_PTR(-ENOMEM);
1066 *bucket = *pos;
1067 return bucket;
1070 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1072 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1073 unsigned int *bucket = v;
1075 *pos = ++(*bucket);
1076 if (*pos >= htable->cfg.size) {
1077 kfree(v);
1078 return NULL;
1080 return bucket;
1083 static void dl_seq_stop(struct seq_file *s, void *v)
1084 __releases(htable->lock)
1086 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1087 unsigned int *bucket = v;
1089 if (!IS_ERR(bucket))
1090 kfree(bucket);
1091 spin_unlock_bh(&htable->lock);
1094 static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1095 struct seq_file *s)
1097 switch (family) {
1098 case NFPROTO_IPV4:
1099 seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1100 (long)(ent->expires - jiffies)/HZ,
1101 &ent->dst.ip.src,
1102 ntohs(ent->dst.src_port),
1103 &ent->dst.ip.dst,
1104 ntohs(ent->dst.dst_port),
1105 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1106 ent->rateinfo.cost);
1107 break;
1108 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1109 case NFPROTO_IPV6:
1110 seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1111 (long)(ent->expires - jiffies)/HZ,
1112 &ent->dst.ip6.src,
1113 ntohs(ent->dst.src_port),
1114 &ent->dst.ip6.dst,
1115 ntohs(ent->dst.dst_port),
1116 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1117 ent->rateinfo.cost);
1118 break;
1119 #endif
1120 default:
1121 BUG();
1125 static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1126 struct seq_file *s)
1128 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1130 spin_lock(&ent->lock);
1131 /* recalculate to show accurate numbers */
1132 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1134 dl_seq_print(ent, family, s);
1136 spin_unlock(&ent->lock);
1137 return seq_has_overflowed(s);
1140 static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1141 struct seq_file *s)
1143 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1145 spin_lock(&ent->lock);
1146 /* recalculate to show accurate numbers */
1147 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1149 dl_seq_print(ent, family, s);
1151 spin_unlock(&ent->lock);
1152 return seq_has_overflowed(s);
1155 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1156 struct seq_file *s)
1158 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1160 spin_lock(&ent->lock);
1161 /* recalculate to show accurate numbers */
1162 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1164 dl_seq_print(ent, family, s);
1166 spin_unlock(&ent->lock);
1167 return seq_has_overflowed(s);
1170 static int dl_seq_show_v2(struct seq_file *s, void *v)
1172 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1173 unsigned int *bucket = (unsigned int *)v;
1174 struct dsthash_ent *ent;
1176 if (!hlist_empty(&htable->hash[*bucket])) {
1177 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1178 if (dl_seq_real_show_v2(ent, htable->family, s))
1179 return -1;
1181 return 0;
1184 static int dl_seq_show_v1(struct seq_file *s, void *v)
1186 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1187 unsigned int *bucket = v;
1188 struct dsthash_ent *ent;
1190 if (!hlist_empty(&htable->hash[*bucket])) {
1191 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1192 if (dl_seq_real_show_v1(ent, htable->family, s))
1193 return -1;
1195 return 0;
1198 static int dl_seq_show(struct seq_file *s, void *v)
1200 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1201 unsigned int *bucket = v;
1202 struct dsthash_ent *ent;
1204 if (!hlist_empty(&htable->hash[*bucket])) {
1205 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1206 if (dl_seq_real_show(ent, htable->family, s))
1207 return -1;
1209 return 0;
1212 static const struct seq_operations dl_seq_ops_v1 = {
1213 .start = dl_seq_start,
1214 .next = dl_seq_next,
1215 .stop = dl_seq_stop,
1216 .show = dl_seq_show_v1
1219 static const struct seq_operations dl_seq_ops_v2 = {
1220 .start = dl_seq_start,
1221 .next = dl_seq_next,
1222 .stop = dl_seq_stop,
1223 .show = dl_seq_show_v2
1226 static const struct seq_operations dl_seq_ops = {
1227 .start = dl_seq_start,
1228 .next = dl_seq_next,
1229 .stop = dl_seq_stop,
1230 .show = dl_seq_show
1233 static int __net_init hashlimit_proc_net_init(struct net *net)
1235 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1237 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1238 if (!hashlimit_net->ipt_hashlimit)
1239 return -ENOMEM;
1240 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1241 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1242 if (!hashlimit_net->ip6t_hashlimit) {
1243 remove_proc_entry("ipt_hashlimit", net->proc_net);
1244 return -ENOMEM;
1246 #endif
1247 return 0;
1250 static void __net_exit hashlimit_proc_net_exit(struct net *net)
1252 struct xt_hashlimit_htable *hinfo;
1253 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1255 /* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1256 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1257 * entries is empty before trying to remove it.
1259 mutex_lock(&hashlimit_mutex);
1260 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1261 htable_remove_proc_entry(hinfo);
1262 hashlimit_net->ipt_hashlimit = NULL;
1263 hashlimit_net->ip6t_hashlimit = NULL;
1264 mutex_unlock(&hashlimit_mutex);
1266 remove_proc_entry("ipt_hashlimit", net->proc_net);
1267 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1268 remove_proc_entry("ip6t_hashlimit", net->proc_net);
1269 #endif
1272 static int __net_init hashlimit_net_init(struct net *net)
1274 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1276 INIT_HLIST_HEAD(&hashlimit_net->htables);
1277 return hashlimit_proc_net_init(net);
1280 static void __net_exit hashlimit_net_exit(struct net *net)
1282 hashlimit_proc_net_exit(net);
1285 static struct pernet_operations hashlimit_net_ops = {
1286 .init = hashlimit_net_init,
1287 .exit = hashlimit_net_exit,
1288 .id = &hashlimit_net_id,
1289 .size = sizeof(struct hashlimit_net),
1292 static int __init hashlimit_mt_init(void)
1294 int err;
1296 err = register_pernet_subsys(&hashlimit_net_ops);
1297 if (err < 0)
1298 return err;
1299 err = xt_register_matches(hashlimit_mt_reg,
1300 ARRAY_SIZE(hashlimit_mt_reg));
1301 if (err < 0)
1302 goto err1;
1304 err = -ENOMEM;
1305 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1306 sizeof(struct dsthash_ent), 0, 0,
1307 NULL);
1308 if (!hashlimit_cachep) {
1309 pr_warn("unable to create slab cache\n");
1310 goto err2;
1312 return 0;
1314 err2:
1315 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1316 err1:
1317 unregister_pernet_subsys(&hashlimit_net_ops);
1318 return err;
1322 static void __exit hashlimit_mt_exit(void)
1324 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1325 unregister_pernet_subsys(&hashlimit_net_ops);
1327 rcu_barrier();
1328 kmem_cache_destroy(hashlimit_cachep);
1331 module_init(hashlimit_mt_init);
1332 module_exit(hashlimit_mt_exit);