Loading...
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)
5 *
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
9 *
10 * Development of this code was funded by Astaro AG, http://www.astaro.com/
11 */
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
30
31#include <net/net_namespace.h>
32#include <net/netns/generic.h>
33
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>
41
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)
46
47MODULE_LICENSE("GPL");
48MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
49MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
50MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
51MODULE_ALIAS("ipt_hashlimit");
52MODULE_ALIAS("ip6t_hashlimit");
53
54struct hashlimit_net {
55 struct hlist_head htables;
56 struct proc_dir_entry *ipt_hashlimit;
57 struct proc_dir_entry *ip6t_hashlimit;
58};
59
60static unsigned int hashlimit_net_id;
61static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
62{
63 return net_generic(net, hashlimit_net_id);
64}
65
66/* need to declare this at the top */
67static const struct seq_operations dl_seq_ops_v2;
68static const struct seq_operations dl_seq_ops_v1;
69static const struct seq_operations dl_seq_ops;
70
71/* hash table crap */
72struct 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
84 };
85 __be16 src_port;
86 __be16 dst_port;
87};
88
89struct dsthash_ent {
90 /* static / read-only parts in the beginning */
91 struct hlist_node node;
92 struct dsthash_dst dst;
93
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;
104 };
105 struct {
106 u_int32_t interval, prev_window;
107 u_int64_t current_rate;
108 u_int64_t rate;
109 int64_t burst;
110 };
111 };
112 } rateinfo;
113 struct rcu_head rcu;
114};
115
116struct 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;
121
122 struct hashlimit_cfg3 cfg; /* config */
123
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;
129
130 /* seq_file stuff */
131 struct proc_dir_entry *pde;
132 const char *name;
133 struct net *net;
134
135 struct hlist_head hash[]; /* hashtable itself */
136};
137
138static int
139cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
140{
141 if (revision == 1) {
142 struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
143
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;
155
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;
169 }
170
171 return 0;
172}
173
174static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */
175static struct kmem_cache *hashlimit_cachep __read_mostly;
176
177static inline bool dst_cmp(const struct dsthash_ent *ent,
178 const struct dsthash_dst *b)
179{
180 return !memcmp(&ent->dst, b, sizeof(ent->dst));
181}
182
183static u_int32_t
184hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
185{
186 u_int32_t hash = jhash2((const u32 *)dst,
187 sizeof(*dst)/sizeof(u32),
188 ht->rnd);
189 /*
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
194 */
195 return reciprocal_scale(hash, ht->cfg.size);
196}
197
198static struct dsthash_ent *
199dsthash_find(const struct xt_hashlimit_htable *ht,
200 const struct dsthash_dst *dst)
201{
202 struct dsthash_ent *ent;
203 u_int32_t hash = hash_dst(ht, dst);
204
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;
210 }
211 }
212 return NULL;
213}
214
215/* allocate dsthash_ent, initialize dst, put in htable and lock it */
216static struct dsthash_ent *
217dsthash_alloc_init(struct xt_hashlimit_htable *ht,
218 const struct dsthash_dst *dst, bool *race)
219{
220 struct dsthash_ent *ent;
221
222 spin_lock(&ht->lock);
223
224 /* Two or more packets may race to create the same entry in the
225 * hashtable, double check if this packet lost race.
226 */
227 ent = dsthash_find(ht, dst);
228 if (ent != NULL) {
229 spin_unlock(&ht->lock);
230 *race = true;
231 return ent;
232 }
233
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;
239 }
240
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);
250
251 spin_lock(&ent->lock);
252 hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
253 ht->count++;
254 }
255 spin_unlock(&ht->lock);
256 return ent;
257}
258
259static void dsthash_free_rcu(struct rcu_head *head)
260{
261 struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
262
263 kmem_cache_free(hashlimit_cachep, ent);
264}
265
266static inline void
267dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
268{
269 hlist_del_rcu(&ent->node);
270 call_rcu(&ent->rcu, dsthash_free_rcu);
271 ht->count--;
272}
273static void htable_gc(struct work_struct *work);
274
275static 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)
279{
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;
286
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;
296 }
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;
302
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;
308 }
309
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;
315
316 for (i = 0; i < hinfo->cfg.size; i++)
317 INIT_HLIST_HEAD(&hinfo->hash[i]);
318
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;
327 }
328 spin_lock_init(&hinfo->lock);
329
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;
339 }
340
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;
349 }
350 hinfo->net = net;
351
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));
355
356 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
357
358 return 0;
359}
360
361static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, bool select_all)
362{
363 unsigned int i;
364
365 for (i = 0; i < ht->cfg.size; i++) {
366 struct dsthash_ent *dh;
367 struct hlist_node *n;
368
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);
373 }
374 spin_unlock_bh(&ht->lock);
375 cond_resched();
376 }
377}
378
379static void htable_gc(struct work_struct *work)
380{
381 struct xt_hashlimit_htable *ht;
382
383 ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
384
385 htable_selective_cleanup(ht, false);
386
387 queue_delayed_work(system_power_efficient_wq,
388 &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
389}
390
391static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
392{
393 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
394 struct proc_dir_entry *parent;
395
396 if (hinfo->family == NFPROTO_IPV4)
397 parent = hashlimit_net->ipt_hashlimit;
398 else
399 parent = hashlimit_net->ip6t_hashlimit;
400
401 if (parent != NULL)
402 remove_proc_entry(hinfo->name, parent);
403}
404
405static struct xt_hashlimit_htable *htable_find_get(struct net *net,
406 const char *name,
407 u_int8_t family)
408{
409 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
410 struct xt_hashlimit_htable *hinfo;
411
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;
417 }
418 }
419 return NULL;
420}
421
422static void htable_put(struct xt_hashlimit_htable *hinfo)
423{
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);
428
429 cancel_delayed_work_sync(&hinfo->gc_work);
430 htable_selective_cleanup(hinfo, true);
431 kfree(hinfo->name);
432 vfree(hinfo);
433 }
434}
435
436/* The algorithm used is the Simple Token Bucket Filter (TBF)
437 * see net/sched/sch_tbf.c in the linux source tree
438 */
439
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'.
445
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.
450
451 See Alexey's formal explanation in net/sched/sch_tbf.c.
452
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.
457*/
458#define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
459#define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
460
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)
472
473#define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
474#define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
475
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.
479 */
480#define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
481#define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
482
483static u32 xt_hashlimit_len_to_chunks(u32 len)
484{
485 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
486}
487
488/* Precision saver. */
489static u64 user2credits(u64 user, int revision)
490{
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;
495
496 /* Avoid overflow: divide the constant operands first */
497 if (scale >= HZ * cpj)
498 return div64_u64(user, div64_u64(scale, HZ * cpj));
499
500 return user * div64_u64(HZ * cpj, scale);
501}
502
503static u32 user2credits_byte(u32 user)
504{
505 u64 us = user;
506 us *= HZ * CREDITS_PER_JIFFY_BYTES;
507 return (u32) (us >> 32);
508}
509
510static u64 user2rate(u64 user)
511{
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;
518 }
519}
520
521static u64 user2rate_bytes(u32 user)
522{
523 u64 r;
524
525 r = user ? U32_MAX / user : U32_MAX;
526 return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
527}
528
529static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
530 u32 mode, int revision)
531{
532 unsigned long delta = now - dh->rateinfo.prev;
533 u64 cap, cpj;
534
535 if (delta == 0)
536 return;
537
538 if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
539 u64 interval = dh->rateinfo.interval * HZ;
540
541 if (delta < interval)
542 return;
543
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;
549
550 return;
551 }
552
553 dh->rateinfo.prev = now;
554
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;
562 }
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;
568 }
569 if (dh->rateinfo.credit > cap)
570 dh->rateinfo.credit = cap;
571}
572
573static void rateinfo_init(struct dsthash_ent *dh,
574 struct xt_hashlimit_htable *hinfo, int revision)
575{
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;
592 }
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;
603 }
604}
605
606static inline __be32 maskl(__be32 a, unsigned int l)
607{
608 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
609}
610
611#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
612static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
613{
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;
632 }
633}
634#endif
635
636static int
637hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
638 struct dsthash_dst *dst,
639 const struct sk_buff *skb, unsigned int protoff)
640{
641 __be16 _ports[2], *ports;
642 u8 nexthdr;
643 int poff;
644
645 memset(dst, 0, sizeof(*dst));
646
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);
655
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:
663 {
664 __be16 frag_off;
665
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);
670 }
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);
675 }
676
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;
685 }
686#endif
687 default:
688 BUG();
689 return 0;
690 }
691
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;
699 }
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;
707}
708
709static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
710{
711 u64 tmp = xt_hashlimit_len_to_chunks(len);
712 tmp = tmp * dh->rateinfo.cost;
713
714 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
715 tmp = CREDITS_PER_JIFFY_BYTES * HZ;
716
717 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
718 dh->rateinfo.credit_cap--;
719 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
720 }
721 return (u32) tmp;
722}
723
724static bool
725hashlimit_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)
728{
729 unsigned long now = jiffies;
730 struct dsthash_ent *dh;
731 struct dsthash_dst dst;
732 bool race = false;
733 u64 cost;
734
735 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
736 goto hotdrop;
737
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);
752 }
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);
757 }
758
759 if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
760 cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
761 dh->rateinfo.current_rate += cost;
762
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;
770 }
771 }
772
773 if (cfg->mode & XT_HASHLIMIT_BYTES)
774 cost = hashlimit_byte_cost(skb->len, dh);
775 else
776 cost = dh->rateinfo.cost;
777
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);
784 }
785
786overlimit:
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;
791
792 hotdrop:
793 par->hotdrop = true;
794 return false;
795}
796
797static bool
798hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
799{
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;
804
805 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
806 if (ret)
807 return ret;
808
809 return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
810}
811
812static bool
813hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
814{
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;
819
820 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
821 if (ret)
822 return ret;
823
824 return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
825}
826
827static bool
828hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
829{
830 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
831 struct xt_hashlimit_htable *hinfo = info->hinfo;
832
833 return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
834}
835
836#define HASHLIMIT_MAX_SIZE 1048576
837
838static 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)
842{
843 struct net *net = par->net;
844 int ret;
845
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);
851 }
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);
855 }
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;
862 }
863
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;
868 }
869
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;
875 }
876
877 if (cfg->interval == 0) {
878 pr_info_ratelimited("invalid interval\n");
879 return -EINVAL;
880 }
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;
886 }
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;
893 }
894
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;
903 }
904 }
905 mutex_unlock(&hashlimit_mutex);
906
907 return 0;
908}
909
910static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
911{
912 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
913 struct hashlimit_cfg3 cfg = {};
914 int ret;
915
916 ret = xt_check_proc_name(info->name, sizeof(info->name));
917 if (ret)
918 return ret;
919
920 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
921 if (ret)
922 return ret;
923
924 return hashlimit_mt_check_common(par, &info->hinfo,
925 &cfg, info->name, 1);
926}
927
928static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
929{
930 struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
931 struct hashlimit_cfg3 cfg = {};
932 int ret;
933
934 ret = xt_check_proc_name(info->name, sizeof(info->name));
935 if (ret)
936 return ret;
937
938 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
939 if (ret)
940 return ret;
941
942 return hashlimit_mt_check_common(par, &info->hinfo,
943 &cfg, info->name, 2);
944}
945
946static int hashlimit_mt_check(const struct xt_mtchk_param *par)
947{
948 struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
949 int ret;
950
951 ret = xt_check_proc_name(info->name, sizeof(info->name));
952 if (ret)
953 return ret;
954
955 return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
956 info->name, 3);
957}
958
959static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
960{
961 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
962
963 htable_put(info->hinfo);
964}
965
966static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
967{
968 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
969
970 htable_put(info->hinfo);
971}
972
973static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
974{
975 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
976
977 htable_put(info->hinfo);
978}
979
980static struct xt_match hashlimit_mt_reg[] __read_mostly = {
981 {
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,
991 },
992 {
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,
1002 },
1003 {
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,
1013 },
1014#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1015 {
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,
1025 },
1026 {
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,
1036 },
1037 {
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,
1047 },
1048#endif
1049};
1050
1051/* PROC stuff */
1052static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1053 __acquires(htable->lock)
1054{
1055 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1056 unsigned int *bucket;
1057
1058 spin_lock_bh(&htable->lock);
1059 if (*pos >= htable->cfg.size)
1060 return NULL;
1061
1062 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1063 if (!bucket)
1064 return ERR_PTR(-ENOMEM);
1065
1066 *bucket = *pos;
1067 return bucket;
1068}
1069
1070static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1071{
1072 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1073 unsigned int *bucket = v;
1074
1075 *pos = ++(*bucket);
1076 if (*pos >= htable->cfg.size) {
1077 kfree(v);
1078 return NULL;
1079 }
1080 return bucket;
1081}
1082
1083static void dl_seq_stop(struct seq_file *s, void *v)
1084 __releases(htable->lock)
1085{
1086 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1087 unsigned int *bucket = v;
1088
1089 if (!IS_ERR(bucket))
1090 kfree(bucket);
1091 spin_unlock_bh(&htable->lock);
1092}
1093
1094static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1095 struct seq_file *s)
1096{
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();
1122 }
1123}
1124
1125static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1126 struct seq_file *s)
1127{
1128 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1129
1130 spin_lock(&ent->lock);
1131 /* recalculate to show accurate numbers */
1132 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1133
1134 dl_seq_print(ent, family, s);
1135
1136 spin_unlock(&ent->lock);
1137 return seq_has_overflowed(s);
1138}
1139
1140static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1141 struct seq_file *s)
1142{
1143 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1144
1145 spin_lock(&ent->lock);
1146 /* recalculate to show accurate numbers */
1147 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1148
1149 dl_seq_print(ent, family, s);
1150
1151 spin_unlock(&ent->lock);
1152 return seq_has_overflowed(s);
1153}
1154
1155static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1156 struct seq_file *s)
1157{
1158 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1159
1160 spin_lock(&ent->lock);
1161 /* recalculate to show accurate numbers */
1162 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1163
1164 dl_seq_print(ent, family, s);
1165
1166 spin_unlock(&ent->lock);
1167 return seq_has_overflowed(s);
1168}
1169
1170static int dl_seq_show_v2(struct seq_file *s, void *v)
1171{
1172 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1173 unsigned int *bucket = (unsigned int *)v;
1174 struct dsthash_ent *ent;
1175
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;
1180 }
1181 return 0;
1182}
1183
1184static int dl_seq_show_v1(struct seq_file *s, void *v)
1185{
1186 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1187 unsigned int *bucket = v;
1188 struct dsthash_ent *ent;
1189
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;
1194 }
1195 return 0;
1196}
1197
1198static int dl_seq_show(struct seq_file *s, void *v)
1199{
1200 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1201 unsigned int *bucket = v;
1202 struct dsthash_ent *ent;
1203
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;
1208 }
1209 return 0;
1210}
1211
1212static 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
1217};
1218
1219static 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
1224};
1225
1226static 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
1231};
1232
1233static int __net_init hashlimit_proc_net_init(struct net *net)
1234{
1235 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1236
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;
1245 }
1246#endif
1247 return 0;
1248}
1249
1250static void __net_exit hashlimit_proc_net_exit(struct net *net)
1251{
1252 struct xt_hashlimit_htable *hinfo;
1253 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1254
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.
1258 */
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);
1265
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
1270}
1271
1272static int __net_init hashlimit_net_init(struct net *net)
1273{
1274 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1275
1276 INIT_HLIST_HEAD(&hashlimit_net->htables);
1277 return hashlimit_proc_net_init(net);
1278}
1279
1280static void __net_exit hashlimit_net_exit(struct net *net)
1281{
1282 hashlimit_proc_net_exit(net);
1283}
1284
1285static 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),
1290};
1291
1292static int __init hashlimit_mt_init(void)
1293{
1294 int err;
1295
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;
1303
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;
1311 }
1312 return 0;
1313
1314err2:
1315 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1316err1:
1317 unregister_pernet_subsys(&hashlimit_net_ops);
1318 return err;
1319
1320}
1321
1322static void __exit hashlimit_mt_exit(void)
1323{
1324 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1325 unregister_pernet_subsys(&hashlimit_net_ops);
1326
1327 rcu_barrier();
1328 kmem_cache_destroy(hashlimit_cachep);
1329}
1330
1331module_init(hashlimit_mt_init);
1332module_exit(hashlimit_mt_exit);
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)
5 *
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
9 *
10 * Development of this code was funded by Astaro AG, http://www.astaro.com/
11 */
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
30
31#include <net/net_namespace.h>
32#include <net/netns/generic.h>
33
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 <uapi/linux/netfilter/xt_hashlimit.h>
40
41#define XT_HASHLIMIT_ALL (XT_HASHLIMIT_HASH_DIP | XT_HASHLIMIT_HASH_DPT | \
42 XT_HASHLIMIT_HASH_SIP | XT_HASHLIMIT_HASH_SPT | \
43 XT_HASHLIMIT_INVERT | XT_HASHLIMIT_BYTES |\
44 XT_HASHLIMIT_RATE_MATCH)
45
46MODULE_LICENSE("GPL");
47MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
48MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
49MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
50MODULE_ALIAS("ipt_hashlimit");
51MODULE_ALIAS("ip6t_hashlimit");
52
53struct hashlimit_net {
54 struct hlist_head htables;
55 struct proc_dir_entry *ipt_hashlimit;
56 struct proc_dir_entry *ip6t_hashlimit;
57};
58
59static unsigned int hashlimit_net_id;
60static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
61{
62 return net_generic(net, hashlimit_net_id);
63}
64
65/* need to declare this at the top */
66static const struct seq_operations dl_seq_ops_v2;
67static const struct seq_operations dl_seq_ops_v1;
68static const struct seq_operations dl_seq_ops;
69
70/* hash table crap */
71struct dsthash_dst {
72 union {
73 struct {
74 __be32 src;
75 __be32 dst;
76 } ip;
77#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
78 struct {
79 __be32 src[4];
80 __be32 dst[4];
81 } ip6;
82#endif
83 };
84 __be16 src_port;
85 __be16 dst_port;
86};
87
88struct dsthash_ent {
89 /* static / read-only parts in the beginning */
90 struct hlist_node node;
91 struct dsthash_dst dst;
92
93 /* modified structure members in the end */
94 spinlock_t lock;
95 unsigned long expires; /* precalculated expiry time */
96 struct {
97 unsigned long prev; /* last modification */
98 union {
99 struct {
100 u_int64_t credit;
101 u_int64_t credit_cap;
102 u_int64_t cost;
103 };
104 struct {
105 u_int32_t interval, prev_window;
106 u_int64_t current_rate;
107 u_int64_t rate;
108 int64_t burst;
109 };
110 };
111 } rateinfo;
112 struct rcu_head rcu;
113};
114
115struct xt_hashlimit_htable {
116 struct hlist_node node; /* global list of all htables */
117 int use;
118 u_int8_t family;
119 bool rnd_initialized;
120
121 struct hashlimit_cfg3 cfg; /* config */
122
123 /* used internally */
124 spinlock_t lock; /* lock for list_head */
125 u_int32_t rnd; /* random seed for hash */
126 unsigned int count; /* number entries in table */
127 struct delayed_work gc_work;
128
129 /* seq_file stuff */
130 struct proc_dir_entry *pde;
131 const char *name;
132 struct net *net;
133
134 struct hlist_head hash[0]; /* hashtable itself */
135};
136
137static int
138cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
139{
140 if (revision == 1) {
141 struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
142
143 to->mode = cfg->mode;
144 to->avg = cfg->avg;
145 to->burst = cfg->burst;
146 to->size = cfg->size;
147 to->max = cfg->max;
148 to->gc_interval = cfg->gc_interval;
149 to->expire = cfg->expire;
150 to->srcmask = cfg->srcmask;
151 to->dstmask = cfg->dstmask;
152 } else if (revision == 2) {
153 struct hashlimit_cfg2 *cfg = (struct hashlimit_cfg2 *)from;
154
155 to->mode = cfg->mode;
156 to->avg = cfg->avg;
157 to->burst = cfg->burst;
158 to->size = cfg->size;
159 to->max = cfg->max;
160 to->gc_interval = cfg->gc_interval;
161 to->expire = cfg->expire;
162 to->srcmask = cfg->srcmask;
163 to->dstmask = cfg->dstmask;
164 } else if (revision == 3) {
165 memcpy(to, from, sizeof(struct hashlimit_cfg3));
166 } else {
167 return -EINVAL;
168 }
169
170 return 0;
171}
172
173static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */
174static struct kmem_cache *hashlimit_cachep __read_mostly;
175
176static inline bool dst_cmp(const struct dsthash_ent *ent,
177 const struct dsthash_dst *b)
178{
179 return !memcmp(&ent->dst, b, sizeof(ent->dst));
180}
181
182static u_int32_t
183hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
184{
185 u_int32_t hash = jhash2((const u32 *)dst,
186 sizeof(*dst)/sizeof(u32),
187 ht->rnd);
188 /*
189 * Instead of returning hash % ht->cfg.size (implying a divide)
190 * we return the high 32 bits of the (hash * ht->cfg.size) that will
191 * give results between [0 and cfg.size-1] and same hash distribution,
192 * but using a multiply, less expensive than a divide
193 */
194 return reciprocal_scale(hash, ht->cfg.size);
195}
196
197static struct dsthash_ent *
198dsthash_find(const struct xt_hashlimit_htable *ht,
199 const struct dsthash_dst *dst)
200{
201 struct dsthash_ent *ent;
202 u_int32_t hash = hash_dst(ht, dst);
203
204 if (!hlist_empty(&ht->hash[hash])) {
205 hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
206 if (dst_cmp(ent, dst)) {
207 spin_lock(&ent->lock);
208 return ent;
209 }
210 }
211 return NULL;
212}
213
214/* allocate dsthash_ent, initialize dst, put in htable and lock it */
215static struct dsthash_ent *
216dsthash_alloc_init(struct xt_hashlimit_htable *ht,
217 const struct dsthash_dst *dst, bool *race)
218{
219 struct dsthash_ent *ent;
220
221 spin_lock(&ht->lock);
222
223 /* Two or more packets may race to create the same entry in the
224 * hashtable, double check if this packet lost race.
225 */
226 ent = dsthash_find(ht, dst);
227 if (ent != NULL) {
228 spin_unlock(&ht->lock);
229 *race = true;
230 return ent;
231 }
232
233 /* initialize hash with random val at the time we allocate
234 * the first hashtable entry */
235 if (unlikely(!ht->rnd_initialized)) {
236 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
237 ht->rnd_initialized = true;
238 }
239
240 if (ht->cfg.max && ht->count >= ht->cfg.max) {
241 /* FIXME: do something. question is what.. */
242 net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
243 ent = NULL;
244 } else
245 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
246 if (ent) {
247 memcpy(&ent->dst, dst, sizeof(ent->dst));
248 spin_lock_init(&ent->lock);
249
250 spin_lock(&ent->lock);
251 hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
252 ht->count++;
253 }
254 spin_unlock(&ht->lock);
255 return ent;
256}
257
258static void dsthash_free_rcu(struct rcu_head *head)
259{
260 struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
261
262 kmem_cache_free(hashlimit_cachep, ent);
263}
264
265static inline void
266dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
267{
268 hlist_del_rcu(&ent->node);
269 call_rcu(&ent->rcu, dsthash_free_rcu);
270 ht->count--;
271}
272static void htable_gc(struct work_struct *work);
273
274static int htable_create(struct net *net, struct hashlimit_cfg3 *cfg,
275 const char *name, u_int8_t family,
276 struct xt_hashlimit_htable **out_hinfo,
277 int revision)
278{
279 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
280 struct xt_hashlimit_htable *hinfo;
281 const struct seq_operations *ops;
282 unsigned int size, i;
283 unsigned long nr_pages = totalram_pages();
284 int ret;
285
286 if (cfg->size) {
287 size = cfg->size;
288 } else {
289 size = (nr_pages << PAGE_SHIFT) / 16384 /
290 sizeof(struct hlist_head);
291 if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
292 size = 8192;
293 if (size < 16)
294 size = 16;
295 }
296 /* FIXME: don't use vmalloc() here or anywhere else -HW */
297 hinfo = vmalloc(struct_size(hinfo, hash, size));
298 if (hinfo == NULL)
299 return -ENOMEM;
300 *out_hinfo = hinfo;
301
302 /* copy match config into hashtable config */
303 ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
304 if (ret) {
305 vfree(hinfo);
306 return ret;
307 }
308
309 hinfo->cfg.size = size;
310 if (hinfo->cfg.max == 0)
311 hinfo->cfg.max = 8 * hinfo->cfg.size;
312 else if (hinfo->cfg.max < hinfo->cfg.size)
313 hinfo->cfg.max = hinfo->cfg.size;
314
315 for (i = 0; i < hinfo->cfg.size; i++)
316 INIT_HLIST_HEAD(&hinfo->hash[i]);
317
318 hinfo->use = 1;
319 hinfo->count = 0;
320 hinfo->family = family;
321 hinfo->rnd_initialized = false;
322 hinfo->name = kstrdup(name, GFP_KERNEL);
323 if (!hinfo->name) {
324 vfree(hinfo);
325 return -ENOMEM;
326 }
327 spin_lock_init(&hinfo->lock);
328
329 switch (revision) {
330 case 1:
331 ops = &dl_seq_ops_v1;
332 break;
333 case 2:
334 ops = &dl_seq_ops_v2;
335 break;
336 default:
337 ops = &dl_seq_ops;
338 }
339
340 hinfo->pde = proc_create_seq_data(name, 0,
341 (family == NFPROTO_IPV4) ?
342 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
343 ops, hinfo);
344 if (hinfo->pde == NULL) {
345 kfree(hinfo->name);
346 vfree(hinfo);
347 return -ENOMEM;
348 }
349 hinfo->net = net;
350
351 INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
352 queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
353 msecs_to_jiffies(hinfo->cfg.gc_interval));
354
355 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
356
357 return 0;
358}
359
360static bool select_all(const struct xt_hashlimit_htable *ht,
361 const struct dsthash_ent *he)
362{
363 return true;
364}
365
366static bool select_gc(const struct xt_hashlimit_htable *ht,
367 const struct dsthash_ent *he)
368{
369 return time_after_eq(jiffies, he->expires);
370}
371
372static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
373 bool (*select)(const struct xt_hashlimit_htable *ht,
374 const struct dsthash_ent *he))
375{
376 unsigned int i;
377
378 for (i = 0; i < ht->cfg.size; i++) {
379 struct dsthash_ent *dh;
380 struct hlist_node *n;
381
382 spin_lock_bh(&ht->lock);
383 hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
384 if ((*select)(ht, dh))
385 dsthash_free(ht, dh);
386 }
387 spin_unlock_bh(&ht->lock);
388 cond_resched();
389 }
390}
391
392static void htable_gc(struct work_struct *work)
393{
394 struct xt_hashlimit_htable *ht;
395
396 ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
397
398 htable_selective_cleanup(ht, select_gc);
399
400 queue_delayed_work(system_power_efficient_wq,
401 &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
402}
403
404static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
405{
406 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
407 struct proc_dir_entry *parent;
408
409 if (hinfo->family == NFPROTO_IPV4)
410 parent = hashlimit_net->ipt_hashlimit;
411 else
412 parent = hashlimit_net->ip6t_hashlimit;
413
414 if (parent != NULL)
415 remove_proc_entry(hinfo->name, parent);
416}
417
418static void htable_destroy(struct xt_hashlimit_htable *hinfo)
419{
420 cancel_delayed_work_sync(&hinfo->gc_work);
421 htable_remove_proc_entry(hinfo);
422 htable_selective_cleanup(hinfo, select_all);
423 kfree(hinfo->name);
424 vfree(hinfo);
425}
426
427static struct xt_hashlimit_htable *htable_find_get(struct net *net,
428 const char *name,
429 u_int8_t family)
430{
431 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
432 struct xt_hashlimit_htable *hinfo;
433
434 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
435 if (!strcmp(name, hinfo->name) &&
436 hinfo->family == family) {
437 hinfo->use++;
438 return hinfo;
439 }
440 }
441 return NULL;
442}
443
444static void htable_put(struct xt_hashlimit_htable *hinfo)
445{
446 mutex_lock(&hashlimit_mutex);
447 if (--hinfo->use == 0) {
448 hlist_del(&hinfo->node);
449 htable_destroy(hinfo);
450 }
451 mutex_unlock(&hashlimit_mutex);
452}
453
454/* The algorithm used is the Simple Token Bucket Filter (TBF)
455 * see net/sched/sch_tbf.c in the linux source tree
456 */
457
458/* Rusty: This is my (non-mathematically-inclined) understanding of
459 this algorithm. The `average rate' in jiffies becomes your initial
460 amount of credit `credit' and the most credit you can ever have
461 `credit_cap'. The `peak rate' becomes the cost of passing the
462 test, `cost'.
463
464 `prev' tracks the last packet hit: you gain one credit per jiffy.
465 If you get credit balance more than this, the extra credit is
466 discarded. Every time the match passes, you lose `cost' credits;
467 if you don't have that many, the test fails.
468
469 See Alexey's formal explanation in net/sched/sch_tbf.c.
470
471 To get the maximum range, we multiply by this factor (ie. you get N
472 credits per jiffy). We want to allow a rate as low as 1 per day
473 (slowest userspace tool allows), which means
474 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
475*/
476#define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
477#define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
478
479/* Repeated shift and or gives us all 1s, final shift and add 1 gives
480 * us the power of 2 below the theoretical max, so GCC simply does a
481 * shift. */
482#define _POW2_BELOW2(x) ((x)|((x)>>1))
483#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
484#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
485#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
486#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
487#define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
488#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
489#define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
490
491#define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
492#define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
493
494/* in byte mode, the lowest possible rate is one packet/second.
495 * credit_cap is used as a counter that tells us how many times we can
496 * refill the "credits available" counter when it becomes empty.
497 */
498#define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
499#define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
500
501static u32 xt_hashlimit_len_to_chunks(u32 len)
502{
503 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
504}
505
506/* Precision saver. */
507static u64 user2credits(u64 user, int revision)
508{
509 u64 scale = (revision == 1) ?
510 XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
511 u64 cpj = (revision == 1) ?
512 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
513
514 /* Avoid overflow: divide the constant operands first */
515 if (scale >= HZ * cpj)
516 return div64_u64(user, div64_u64(scale, HZ * cpj));
517
518 return user * div64_u64(HZ * cpj, scale);
519}
520
521static u32 user2credits_byte(u32 user)
522{
523 u64 us = user;
524 us *= HZ * CREDITS_PER_JIFFY_BYTES;
525 return (u32) (us >> 32);
526}
527
528static u64 user2rate(u64 user)
529{
530 if (user != 0) {
531 return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
532 } else {
533 pr_info_ratelimited("invalid rate from userspace: %llu\n",
534 user);
535 return 0;
536 }
537}
538
539static u64 user2rate_bytes(u32 user)
540{
541 u64 r;
542
543 r = user ? U32_MAX / user : U32_MAX;
544 return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
545}
546
547static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
548 u32 mode, int revision)
549{
550 unsigned long delta = now - dh->rateinfo.prev;
551 u64 cap, cpj;
552
553 if (delta == 0)
554 return;
555
556 if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
557 u64 interval = dh->rateinfo.interval * HZ;
558
559 if (delta < interval)
560 return;
561
562 dh->rateinfo.prev = now;
563 dh->rateinfo.prev_window =
564 ((dh->rateinfo.current_rate * interval) >
565 (delta * dh->rateinfo.rate));
566 dh->rateinfo.current_rate = 0;
567
568 return;
569 }
570
571 dh->rateinfo.prev = now;
572
573 if (mode & XT_HASHLIMIT_BYTES) {
574 u64 tmp = dh->rateinfo.credit;
575 dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
576 cap = CREDITS_PER_JIFFY_BYTES * HZ;
577 if (tmp >= dh->rateinfo.credit) {/* overflow */
578 dh->rateinfo.credit = cap;
579 return;
580 }
581 } else {
582 cpj = (revision == 1) ?
583 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
584 dh->rateinfo.credit += delta * cpj;
585 cap = dh->rateinfo.credit_cap;
586 }
587 if (dh->rateinfo.credit > cap)
588 dh->rateinfo.credit = cap;
589}
590
591static void rateinfo_init(struct dsthash_ent *dh,
592 struct xt_hashlimit_htable *hinfo, int revision)
593{
594 dh->rateinfo.prev = jiffies;
595 if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
596 dh->rateinfo.prev_window = 0;
597 dh->rateinfo.current_rate = 0;
598 if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
599 dh->rateinfo.rate =
600 user2rate_bytes((u32)hinfo->cfg.avg);
601 if (hinfo->cfg.burst)
602 dh->rateinfo.burst =
603 hinfo->cfg.burst * dh->rateinfo.rate;
604 else
605 dh->rateinfo.burst = dh->rateinfo.rate;
606 } else {
607 dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
608 dh->rateinfo.burst =
609 hinfo->cfg.burst + dh->rateinfo.rate;
610 }
611 dh->rateinfo.interval = hinfo->cfg.interval;
612 } else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
613 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
614 dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
615 dh->rateinfo.credit_cap = hinfo->cfg.burst;
616 } else {
617 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
618 hinfo->cfg.burst, revision);
619 dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
620 dh->rateinfo.credit_cap = dh->rateinfo.credit;
621 }
622}
623
624static inline __be32 maskl(__be32 a, unsigned int l)
625{
626 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
627}
628
629#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
630static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
631{
632 switch (p) {
633 case 0 ... 31:
634 i[0] = maskl(i[0], p);
635 i[1] = i[2] = i[3] = 0;
636 break;
637 case 32 ... 63:
638 i[1] = maskl(i[1], p - 32);
639 i[2] = i[3] = 0;
640 break;
641 case 64 ... 95:
642 i[2] = maskl(i[2], p - 64);
643 i[3] = 0;
644 break;
645 case 96 ... 127:
646 i[3] = maskl(i[3], p - 96);
647 break;
648 case 128:
649 break;
650 }
651}
652#endif
653
654static int
655hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
656 struct dsthash_dst *dst,
657 const struct sk_buff *skb, unsigned int protoff)
658{
659 __be16 _ports[2], *ports;
660 u8 nexthdr;
661 int poff;
662
663 memset(dst, 0, sizeof(*dst));
664
665 switch (hinfo->family) {
666 case NFPROTO_IPV4:
667 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
668 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
669 hinfo->cfg.dstmask);
670 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
671 dst->ip.src = maskl(ip_hdr(skb)->saddr,
672 hinfo->cfg.srcmask);
673
674 if (!(hinfo->cfg.mode &
675 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
676 return 0;
677 nexthdr = ip_hdr(skb)->protocol;
678 break;
679#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
680 case NFPROTO_IPV6:
681 {
682 __be16 frag_off;
683
684 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
685 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
686 sizeof(dst->ip6.dst));
687 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
688 }
689 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
690 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
691 sizeof(dst->ip6.src));
692 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
693 }
694
695 if (!(hinfo->cfg.mode &
696 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
697 return 0;
698 nexthdr = ipv6_hdr(skb)->nexthdr;
699 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
700 if ((int)protoff < 0)
701 return -1;
702 break;
703 }
704#endif
705 default:
706 BUG();
707 return 0;
708 }
709
710 poff = proto_ports_offset(nexthdr);
711 if (poff >= 0) {
712 ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
713 &_ports);
714 } else {
715 _ports[0] = _ports[1] = 0;
716 ports = _ports;
717 }
718 if (!ports)
719 return -1;
720 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
721 dst->src_port = ports[0];
722 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
723 dst->dst_port = ports[1];
724 return 0;
725}
726
727static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
728{
729 u64 tmp = xt_hashlimit_len_to_chunks(len);
730 tmp = tmp * dh->rateinfo.cost;
731
732 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
733 tmp = CREDITS_PER_JIFFY_BYTES * HZ;
734
735 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
736 dh->rateinfo.credit_cap--;
737 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
738 }
739 return (u32) tmp;
740}
741
742static bool
743hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
744 struct xt_hashlimit_htable *hinfo,
745 const struct hashlimit_cfg3 *cfg, int revision)
746{
747 unsigned long now = jiffies;
748 struct dsthash_ent *dh;
749 struct dsthash_dst dst;
750 bool race = false;
751 u64 cost;
752
753 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
754 goto hotdrop;
755
756 local_bh_disable();
757 dh = dsthash_find(hinfo, &dst);
758 if (dh == NULL) {
759 dh = dsthash_alloc_init(hinfo, &dst, &race);
760 if (dh == NULL) {
761 local_bh_enable();
762 goto hotdrop;
763 } else if (race) {
764 /* Already got an entry, update expiration timeout */
765 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
766 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
767 } else {
768 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
769 rateinfo_init(dh, hinfo, revision);
770 }
771 } else {
772 /* update expiration timeout */
773 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
774 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
775 }
776
777 if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
778 cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
779 dh->rateinfo.current_rate += cost;
780
781 if (!dh->rateinfo.prev_window &&
782 (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
783 spin_unlock(&dh->lock);
784 local_bh_enable();
785 return !(cfg->mode & XT_HASHLIMIT_INVERT);
786 } else {
787 goto overlimit;
788 }
789 }
790
791 if (cfg->mode & XT_HASHLIMIT_BYTES)
792 cost = hashlimit_byte_cost(skb->len, dh);
793 else
794 cost = dh->rateinfo.cost;
795
796 if (dh->rateinfo.credit >= cost) {
797 /* below the limit */
798 dh->rateinfo.credit -= cost;
799 spin_unlock(&dh->lock);
800 local_bh_enable();
801 return !(cfg->mode & XT_HASHLIMIT_INVERT);
802 }
803
804overlimit:
805 spin_unlock(&dh->lock);
806 local_bh_enable();
807 /* default match is underlimit - so over the limit, we need to invert */
808 return cfg->mode & XT_HASHLIMIT_INVERT;
809
810 hotdrop:
811 par->hotdrop = true;
812 return false;
813}
814
815static bool
816hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
817{
818 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
819 struct xt_hashlimit_htable *hinfo = info->hinfo;
820 struct hashlimit_cfg3 cfg = {};
821 int ret;
822
823 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
824 if (ret)
825 return ret;
826
827 return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
828}
829
830static bool
831hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
832{
833 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
834 struct xt_hashlimit_htable *hinfo = info->hinfo;
835 struct hashlimit_cfg3 cfg = {};
836 int ret;
837
838 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
839 if (ret)
840 return ret;
841
842 return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
843}
844
845static bool
846hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
847{
848 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
849 struct xt_hashlimit_htable *hinfo = info->hinfo;
850
851 return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
852}
853
854static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
855 struct xt_hashlimit_htable **hinfo,
856 struct hashlimit_cfg3 *cfg,
857 const char *name, int revision)
858{
859 struct net *net = par->net;
860 int ret;
861
862 if (cfg->gc_interval == 0 || cfg->expire == 0)
863 return -EINVAL;
864 if (par->family == NFPROTO_IPV4) {
865 if (cfg->srcmask > 32 || cfg->dstmask > 32)
866 return -EINVAL;
867 } else {
868 if (cfg->srcmask > 128 || cfg->dstmask > 128)
869 return -EINVAL;
870 }
871
872 if (cfg->mode & ~XT_HASHLIMIT_ALL) {
873 pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
874 cfg->mode);
875 return -EINVAL;
876 }
877
878 /* Check for overflow. */
879 if (revision >= 3 && cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
880 if (cfg->avg == 0 || cfg->avg > U32_MAX) {
881 pr_info_ratelimited("invalid rate\n");
882 return -ERANGE;
883 }
884
885 if (cfg->interval == 0) {
886 pr_info_ratelimited("invalid interval\n");
887 return -EINVAL;
888 }
889 } else if (cfg->mode & XT_HASHLIMIT_BYTES) {
890 if (user2credits_byte(cfg->avg) == 0) {
891 pr_info_ratelimited("overflow, rate too high: %llu\n",
892 cfg->avg);
893 return -EINVAL;
894 }
895 } else if (cfg->burst == 0 ||
896 user2credits(cfg->avg * cfg->burst, revision) <
897 user2credits(cfg->avg, revision)) {
898 pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
899 cfg->avg, cfg->burst);
900 return -ERANGE;
901 }
902
903 mutex_lock(&hashlimit_mutex);
904 *hinfo = htable_find_get(net, name, par->family);
905 if (*hinfo == NULL) {
906 ret = htable_create(net, cfg, name, par->family,
907 hinfo, revision);
908 if (ret < 0) {
909 mutex_unlock(&hashlimit_mutex);
910 return ret;
911 }
912 }
913 mutex_unlock(&hashlimit_mutex);
914
915 return 0;
916}
917
918static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
919{
920 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
921 struct hashlimit_cfg3 cfg = {};
922 int ret;
923
924 ret = xt_check_proc_name(info->name, sizeof(info->name));
925 if (ret)
926 return ret;
927
928 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
929 if (ret)
930 return ret;
931
932 return hashlimit_mt_check_common(par, &info->hinfo,
933 &cfg, info->name, 1);
934}
935
936static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
937{
938 struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
939 struct hashlimit_cfg3 cfg = {};
940 int ret;
941
942 ret = xt_check_proc_name(info->name, sizeof(info->name));
943 if (ret)
944 return ret;
945
946 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
947 if (ret)
948 return ret;
949
950 return hashlimit_mt_check_common(par, &info->hinfo,
951 &cfg, info->name, 2);
952}
953
954static int hashlimit_mt_check(const struct xt_mtchk_param *par)
955{
956 struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
957 int ret;
958
959 ret = xt_check_proc_name(info->name, sizeof(info->name));
960 if (ret)
961 return ret;
962
963 return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
964 info->name, 3);
965}
966
967static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
968{
969 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
970
971 htable_put(info->hinfo);
972}
973
974static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
975{
976 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
977
978 htable_put(info->hinfo);
979}
980
981static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
982{
983 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
984
985 htable_put(info->hinfo);
986}
987
988static struct xt_match hashlimit_mt_reg[] __read_mostly = {
989 {
990 .name = "hashlimit",
991 .revision = 1,
992 .family = NFPROTO_IPV4,
993 .match = hashlimit_mt_v1,
994 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
995 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
996 .checkentry = hashlimit_mt_check_v1,
997 .destroy = hashlimit_mt_destroy_v1,
998 .me = THIS_MODULE,
999 },
1000 {
1001 .name = "hashlimit",
1002 .revision = 2,
1003 .family = NFPROTO_IPV4,
1004 .match = hashlimit_mt_v2,
1005 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1006 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1007 .checkentry = hashlimit_mt_check_v2,
1008 .destroy = hashlimit_mt_destroy_v2,
1009 .me = THIS_MODULE,
1010 },
1011 {
1012 .name = "hashlimit",
1013 .revision = 3,
1014 .family = NFPROTO_IPV4,
1015 .match = hashlimit_mt,
1016 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1017 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1018 .checkentry = hashlimit_mt_check,
1019 .destroy = hashlimit_mt_destroy,
1020 .me = THIS_MODULE,
1021 },
1022#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1023 {
1024 .name = "hashlimit",
1025 .revision = 1,
1026 .family = NFPROTO_IPV6,
1027 .match = hashlimit_mt_v1,
1028 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
1029 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
1030 .checkentry = hashlimit_mt_check_v1,
1031 .destroy = hashlimit_mt_destroy_v1,
1032 .me = THIS_MODULE,
1033 },
1034 {
1035 .name = "hashlimit",
1036 .revision = 2,
1037 .family = NFPROTO_IPV6,
1038 .match = hashlimit_mt_v2,
1039 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1040 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1041 .checkentry = hashlimit_mt_check_v2,
1042 .destroy = hashlimit_mt_destroy_v2,
1043 .me = THIS_MODULE,
1044 },
1045 {
1046 .name = "hashlimit",
1047 .revision = 3,
1048 .family = NFPROTO_IPV6,
1049 .match = hashlimit_mt,
1050 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1051 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1052 .checkentry = hashlimit_mt_check,
1053 .destroy = hashlimit_mt_destroy,
1054 .me = THIS_MODULE,
1055 },
1056#endif
1057};
1058
1059/* PROC stuff */
1060static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1061 __acquires(htable->lock)
1062{
1063 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1064 unsigned int *bucket;
1065
1066 spin_lock_bh(&htable->lock);
1067 if (*pos >= htable->cfg.size)
1068 return NULL;
1069
1070 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1071 if (!bucket)
1072 return ERR_PTR(-ENOMEM);
1073
1074 *bucket = *pos;
1075 return bucket;
1076}
1077
1078static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1079{
1080 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1081 unsigned int *bucket = v;
1082
1083 *pos = ++(*bucket);
1084 if (*pos >= htable->cfg.size) {
1085 kfree(v);
1086 return NULL;
1087 }
1088 return bucket;
1089}
1090
1091static void dl_seq_stop(struct seq_file *s, void *v)
1092 __releases(htable->lock)
1093{
1094 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1095 unsigned int *bucket = v;
1096
1097 if (!IS_ERR(bucket))
1098 kfree(bucket);
1099 spin_unlock_bh(&htable->lock);
1100}
1101
1102static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1103 struct seq_file *s)
1104{
1105 switch (family) {
1106 case NFPROTO_IPV4:
1107 seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1108 (long)(ent->expires - jiffies)/HZ,
1109 &ent->dst.ip.src,
1110 ntohs(ent->dst.src_port),
1111 &ent->dst.ip.dst,
1112 ntohs(ent->dst.dst_port),
1113 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1114 ent->rateinfo.cost);
1115 break;
1116#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1117 case NFPROTO_IPV6:
1118 seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1119 (long)(ent->expires - jiffies)/HZ,
1120 &ent->dst.ip6.src,
1121 ntohs(ent->dst.src_port),
1122 &ent->dst.ip6.dst,
1123 ntohs(ent->dst.dst_port),
1124 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1125 ent->rateinfo.cost);
1126 break;
1127#endif
1128 default:
1129 BUG();
1130 }
1131}
1132
1133static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1134 struct seq_file *s)
1135{
1136 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1137
1138 spin_lock(&ent->lock);
1139 /* recalculate to show accurate numbers */
1140 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1141
1142 dl_seq_print(ent, family, s);
1143
1144 spin_unlock(&ent->lock);
1145 return seq_has_overflowed(s);
1146}
1147
1148static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1149 struct seq_file *s)
1150{
1151 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1152
1153 spin_lock(&ent->lock);
1154 /* recalculate to show accurate numbers */
1155 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1156
1157 dl_seq_print(ent, family, s);
1158
1159 spin_unlock(&ent->lock);
1160 return seq_has_overflowed(s);
1161}
1162
1163static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1164 struct seq_file *s)
1165{
1166 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1167
1168 spin_lock(&ent->lock);
1169 /* recalculate to show accurate numbers */
1170 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1171
1172 dl_seq_print(ent, family, s);
1173
1174 spin_unlock(&ent->lock);
1175 return seq_has_overflowed(s);
1176}
1177
1178static int dl_seq_show_v2(struct seq_file *s, void *v)
1179{
1180 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1181 unsigned int *bucket = (unsigned int *)v;
1182 struct dsthash_ent *ent;
1183
1184 if (!hlist_empty(&htable->hash[*bucket])) {
1185 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1186 if (dl_seq_real_show_v2(ent, htable->family, s))
1187 return -1;
1188 }
1189 return 0;
1190}
1191
1192static int dl_seq_show_v1(struct seq_file *s, void *v)
1193{
1194 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1195 unsigned int *bucket = v;
1196 struct dsthash_ent *ent;
1197
1198 if (!hlist_empty(&htable->hash[*bucket])) {
1199 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1200 if (dl_seq_real_show_v1(ent, htable->family, s))
1201 return -1;
1202 }
1203 return 0;
1204}
1205
1206static int dl_seq_show(struct seq_file *s, void *v)
1207{
1208 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1209 unsigned int *bucket = v;
1210 struct dsthash_ent *ent;
1211
1212 if (!hlist_empty(&htable->hash[*bucket])) {
1213 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1214 if (dl_seq_real_show(ent, htable->family, s))
1215 return -1;
1216 }
1217 return 0;
1218}
1219
1220static const struct seq_operations dl_seq_ops_v1 = {
1221 .start = dl_seq_start,
1222 .next = dl_seq_next,
1223 .stop = dl_seq_stop,
1224 .show = dl_seq_show_v1
1225};
1226
1227static const struct seq_operations dl_seq_ops_v2 = {
1228 .start = dl_seq_start,
1229 .next = dl_seq_next,
1230 .stop = dl_seq_stop,
1231 .show = dl_seq_show_v2
1232};
1233
1234static const struct seq_operations dl_seq_ops = {
1235 .start = dl_seq_start,
1236 .next = dl_seq_next,
1237 .stop = dl_seq_stop,
1238 .show = dl_seq_show
1239};
1240
1241static int __net_init hashlimit_proc_net_init(struct net *net)
1242{
1243 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1244
1245 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1246 if (!hashlimit_net->ipt_hashlimit)
1247 return -ENOMEM;
1248#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1249 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1250 if (!hashlimit_net->ip6t_hashlimit) {
1251 remove_proc_entry("ipt_hashlimit", net->proc_net);
1252 return -ENOMEM;
1253 }
1254#endif
1255 return 0;
1256}
1257
1258static void __net_exit hashlimit_proc_net_exit(struct net *net)
1259{
1260 struct xt_hashlimit_htable *hinfo;
1261 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1262
1263 /* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1264 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1265 * entries is empty before trying to remove it.
1266 */
1267 mutex_lock(&hashlimit_mutex);
1268 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1269 htable_remove_proc_entry(hinfo);
1270 hashlimit_net->ipt_hashlimit = NULL;
1271 hashlimit_net->ip6t_hashlimit = NULL;
1272 mutex_unlock(&hashlimit_mutex);
1273
1274 remove_proc_entry("ipt_hashlimit", net->proc_net);
1275#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1276 remove_proc_entry("ip6t_hashlimit", net->proc_net);
1277#endif
1278}
1279
1280static int __net_init hashlimit_net_init(struct net *net)
1281{
1282 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1283
1284 INIT_HLIST_HEAD(&hashlimit_net->htables);
1285 return hashlimit_proc_net_init(net);
1286}
1287
1288static void __net_exit hashlimit_net_exit(struct net *net)
1289{
1290 hashlimit_proc_net_exit(net);
1291}
1292
1293static struct pernet_operations hashlimit_net_ops = {
1294 .init = hashlimit_net_init,
1295 .exit = hashlimit_net_exit,
1296 .id = &hashlimit_net_id,
1297 .size = sizeof(struct hashlimit_net),
1298};
1299
1300static int __init hashlimit_mt_init(void)
1301{
1302 int err;
1303
1304 err = register_pernet_subsys(&hashlimit_net_ops);
1305 if (err < 0)
1306 return err;
1307 err = xt_register_matches(hashlimit_mt_reg,
1308 ARRAY_SIZE(hashlimit_mt_reg));
1309 if (err < 0)
1310 goto err1;
1311
1312 err = -ENOMEM;
1313 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1314 sizeof(struct dsthash_ent), 0, 0,
1315 NULL);
1316 if (!hashlimit_cachep) {
1317 pr_warn("unable to create slab cache\n");
1318 goto err2;
1319 }
1320 return 0;
1321
1322err2:
1323 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1324err1:
1325 unregister_pernet_subsys(&hashlimit_net_ops);
1326 return err;
1327
1328}
1329
1330static void __exit hashlimit_mt_exit(void)
1331{
1332 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1333 unregister_pernet_subsys(&hashlimit_net_ops);
1334
1335 rcu_barrier();
1336 kmem_cache_destroy(hashlimit_cachep);
1337}
1338
1339module_init(hashlimit_mt_init);
1340module_exit(hashlimit_mt_exit);