xref: /openbmc/linux/net/netfilter/xt_hashlimit.c (revision 31af04cd)
1 /*
2  *	xt_hashlimit - Netfilter module to limit the number of packets per time
3  *	separately for each hashbucket (sourceip/sourceport/dstip/dstport)
4  *
5  *	(C) 2003-2004 by Harald Welte <laforge@netfilter.org>
6  *	(C) 2006-2012 Patrick McHardy <kaber@trash.net>
7  *	Copyright © CC Computer Consultants GmbH, 2007 - 2008
8  *
9  * Development of this code was funded by Astaro AG, http://www.astaro.com/
10  */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/module.h>
13 #include <linux/spinlock.h>
14 #include <linux/random.h>
15 #include <linux/jhash.h>
16 #include <linux/slab.h>
17 #include <linux/vmalloc.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/list.h>
21 #include <linux/skbuff.h>
22 #include <linux/mm.h>
23 #include <linux/in.h>
24 #include <linux/ip.h>
25 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
26 #include <linux/ipv6.h>
27 #include <net/ipv6.h>
28 #endif
29 
30 #include <net/net_namespace.h>
31 #include <net/netns/generic.h>
32 
33 #include <linux/netfilter/x_tables.h>
34 #include <linux/netfilter_ipv4/ip_tables.h>
35 #include <linux/netfilter_ipv6/ip6_tables.h>
36 #include <linux/netfilter/xt_hashlimit.h>
37 #include <linux/mutex.h>
38 #include <linux/kernel.h>
39 
40 MODULE_LICENSE("GPL");
41 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
42 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
43 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
44 MODULE_ALIAS("ipt_hashlimit");
45 MODULE_ALIAS("ip6t_hashlimit");
46 
47 struct hashlimit_net {
48 	struct hlist_head	htables;
49 	struct proc_dir_entry	*ipt_hashlimit;
50 	struct proc_dir_entry	*ip6t_hashlimit;
51 };
52 
53 static unsigned int hashlimit_net_id;
54 static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
55 {
56 	return net_generic(net, hashlimit_net_id);
57 }
58 
59 /* need to declare this at the top */
60 static const struct seq_operations dl_seq_ops_v2;
61 static const struct seq_operations dl_seq_ops_v1;
62 static const struct seq_operations dl_seq_ops;
63 
64 /* hash table crap */
65 struct dsthash_dst {
66 	union {
67 		struct {
68 			__be32 src;
69 			__be32 dst;
70 		} ip;
71 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
72 		struct {
73 			__be32 src[4];
74 			__be32 dst[4];
75 		} ip6;
76 #endif
77 	};
78 	__be16 src_port;
79 	__be16 dst_port;
80 };
81 
82 struct dsthash_ent {
83 	/* static / read-only parts in the beginning */
84 	struct hlist_node node;
85 	struct dsthash_dst dst;
86 
87 	/* modified structure members in the end */
88 	spinlock_t lock;
89 	unsigned long expires;		/* precalculated expiry time */
90 	struct {
91 		unsigned long prev;	/* last modification */
92 		union {
93 			struct {
94 				u_int64_t credit;
95 				u_int64_t credit_cap;
96 				u_int64_t cost;
97 			};
98 			struct {
99 				u_int32_t interval, prev_window;
100 				u_int64_t current_rate;
101 				u_int64_t rate;
102 				int64_t burst;
103 			};
104 		};
105 	} rateinfo;
106 	struct rcu_head rcu;
107 };
108 
109 struct xt_hashlimit_htable {
110 	struct hlist_node node;		/* global list of all htables */
111 	int use;
112 	u_int8_t family;
113 	bool rnd_initialized;
114 
115 	struct hashlimit_cfg3 cfg;	/* config */
116 
117 	/* used internally */
118 	spinlock_t lock;		/* lock for list_head */
119 	u_int32_t rnd;			/* random seed for hash */
120 	unsigned int count;		/* number entries in table */
121 	struct delayed_work gc_work;
122 
123 	/* seq_file stuff */
124 	struct proc_dir_entry *pde;
125 	const char *name;
126 	struct net *net;
127 
128 	struct hlist_head hash[0];	/* hashtable itself */
129 };
130 
131 static int
132 cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
133 {
134 	if (revision == 1) {
135 		struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
136 
137 		to->mode = cfg->mode;
138 		to->avg = cfg->avg;
139 		to->burst = cfg->burst;
140 		to->size = cfg->size;
141 		to->max = cfg->max;
142 		to->gc_interval = cfg->gc_interval;
143 		to->expire = cfg->expire;
144 		to->srcmask = cfg->srcmask;
145 		to->dstmask = cfg->dstmask;
146 	} else if (revision == 2) {
147 		struct hashlimit_cfg2 *cfg = (struct hashlimit_cfg2 *)from;
148 
149 		to->mode = cfg->mode;
150 		to->avg = cfg->avg;
151 		to->burst = cfg->burst;
152 		to->size = cfg->size;
153 		to->max = cfg->max;
154 		to->gc_interval = cfg->gc_interval;
155 		to->expire = cfg->expire;
156 		to->srcmask = cfg->srcmask;
157 		to->dstmask = cfg->dstmask;
158 	} else if (revision == 3) {
159 		memcpy(to, from, sizeof(struct hashlimit_cfg3));
160 	} else {
161 		return -EINVAL;
162 	}
163 
164 	return 0;
165 }
166 
167 static DEFINE_MUTEX(hashlimit_mutex);	/* protects htables list */
168 static struct kmem_cache *hashlimit_cachep __read_mostly;
169 
170 static inline bool dst_cmp(const struct dsthash_ent *ent,
171 			   const struct dsthash_dst *b)
172 {
173 	return !memcmp(&ent->dst, b, sizeof(ent->dst));
174 }
175 
176 static u_int32_t
177 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
178 {
179 	u_int32_t hash = jhash2((const u32 *)dst,
180 				sizeof(*dst)/sizeof(u32),
181 				ht->rnd);
182 	/*
183 	 * Instead of returning hash % ht->cfg.size (implying a divide)
184 	 * we return the high 32 bits of the (hash * ht->cfg.size) that will
185 	 * give results between [0 and cfg.size-1] and same hash distribution,
186 	 * but using a multiply, less expensive than a divide
187 	 */
188 	return reciprocal_scale(hash, ht->cfg.size);
189 }
190 
191 static struct dsthash_ent *
192 dsthash_find(const struct xt_hashlimit_htable *ht,
193 	     const struct dsthash_dst *dst)
194 {
195 	struct dsthash_ent *ent;
196 	u_int32_t hash = hash_dst(ht, dst);
197 
198 	if (!hlist_empty(&ht->hash[hash])) {
199 		hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
200 			if (dst_cmp(ent, dst)) {
201 				spin_lock(&ent->lock);
202 				return ent;
203 			}
204 	}
205 	return NULL;
206 }
207 
208 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
209 static struct dsthash_ent *
210 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
211 		   const struct dsthash_dst *dst, bool *race)
212 {
213 	struct dsthash_ent *ent;
214 
215 	spin_lock(&ht->lock);
216 
217 	/* Two or more packets may race to create the same entry in the
218 	 * hashtable, double check if this packet lost race.
219 	 */
220 	ent = dsthash_find(ht, dst);
221 	if (ent != NULL) {
222 		spin_unlock(&ht->lock);
223 		*race = true;
224 		return ent;
225 	}
226 
227 	/* initialize hash with random val at the time we allocate
228 	 * the first hashtable entry */
229 	if (unlikely(!ht->rnd_initialized)) {
230 		get_random_bytes(&ht->rnd, sizeof(ht->rnd));
231 		ht->rnd_initialized = true;
232 	}
233 
234 	if (ht->cfg.max && ht->count >= ht->cfg.max) {
235 		/* FIXME: do something. question is what.. */
236 		net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
237 		ent = NULL;
238 	} else
239 		ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
240 	if (ent) {
241 		memcpy(&ent->dst, dst, sizeof(ent->dst));
242 		spin_lock_init(&ent->lock);
243 
244 		spin_lock(&ent->lock);
245 		hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
246 		ht->count++;
247 	}
248 	spin_unlock(&ht->lock);
249 	return ent;
250 }
251 
252 static void dsthash_free_rcu(struct rcu_head *head)
253 {
254 	struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
255 
256 	kmem_cache_free(hashlimit_cachep, ent);
257 }
258 
259 static inline void
260 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
261 {
262 	hlist_del_rcu(&ent->node);
263 	call_rcu(&ent->rcu, dsthash_free_rcu);
264 	ht->count--;
265 }
266 static void htable_gc(struct work_struct *work);
267 
268 static int htable_create(struct net *net, struct hashlimit_cfg3 *cfg,
269 			 const char *name, u_int8_t family,
270 			 struct xt_hashlimit_htable **out_hinfo,
271 			 int revision)
272 {
273 	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
274 	struct xt_hashlimit_htable *hinfo;
275 	const struct seq_operations *ops;
276 	unsigned int size, i;
277 	unsigned long nr_pages = totalram_pages();
278 	int ret;
279 
280 	if (cfg->size) {
281 		size = cfg->size;
282 	} else {
283 		size = (nr_pages << PAGE_SHIFT) / 16384 /
284 		       sizeof(struct hlist_head);
285 		if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
286 			size = 8192;
287 		if (size < 16)
288 			size = 16;
289 	}
290 	/* FIXME: don't use vmalloc() here or anywhere else -HW */
291 	hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
292 	                sizeof(struct hlist_head) * size);
293 	if (hinfo == NULL)
294 		return -ENOMEM;
295 	*out_hinfo = hinfo;
296 
297 	/* copy match config into hashtable config */
298 	ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
299 	if (ret) {
300 		vfree(hinfo);
301 		return ret;
302 	}
303 
304 	hinfo->cfg.size = size;
305 	if (hinfo->cfg.max == 0)
306 		hinfo->cfg.max = 8 * hinfo->cfg.size;
307 	else if (hinfo->cfg.max < hinfo->cfg.size)
308 		hinfo->cfg.max = hinfo->cfg.size;
309 
310 	for (i = 0; i < hinfo->cfg.size; i++)
311 		INIT_HLIST_HEAD(&hinfo->hash[i]);
312 
313 	hinfo->use = 1;
314 	hinfo->count = 0;
315 	hinfo->family = family;
316 	hinfo->rnd_initialized = false;
317 	hinfo->name = kstrdup(name, GFP_KERNEL);
318 	if (!hinfo->name) {
319 		vfree(hinfo);
320 		return -ENOMEM;
321 	}
322 	spin_lock_init(&hinfo->lock);
323 
324 	switch (revision) {
325 	case 1:
326 		ops = &dl_seq_ops_v1;
327 		break;
328 	case 2:
329 		ops = &dl_seq_ops_v2;
330 		break;
331 	default:
332 		ops = &dl_seq_ops;
333 	}
334 
335 	hinfo->pde = proc_create_seq_data(name, 0,
336 		(family == NFPROTO_IPV4) ?
337 		hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
338 		ops, hinfo);
339 	if (hinfo->pde == NULL) {
340 		kfree(hinfo->name);
341 		vfree(hinfo);
342 		return -ENOMEM;
343 	}
344 	hinfo->net = net;
345 
346 	INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
347 	queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
348 			   msecs_to_jiffies(hinfo->cfg.gc_interval));
349 
350 	hlist_add_head(&hinfo->node, &hashlimit_net->htables);
351 
352 	return 0;
353 }
354 
355 static bool select_all(const struct xt_hashlimit_htable *ht,
356 		       const struct dsthash_ent *he)
357 {
358 	return true;
359 }
360 
361 static bool select_gc(const struct xt_hashlimit_htable *ht,
362 		      const struct dsthash_ent *he)
363 {
364 	return time_after_eq(jiffies, he->expires);
365 }
366 
367 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
368 			bool (*select)(const struct xt_hashlimit_htable *ht,
369 				      const struct dsthash_ent *he))
370 {
371 	unsigned int i;
372 
373 	for (i = 0; i < ht->cfg.size; i++) {
374 		struct dsthash_ent *dh;
375 		struct hlist_node *n;
376 
377 		spin_lock_bh(&ht->lock);
378 		hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
379 			if ((*select)(ht, dh))
380 				dsthash_free(ht, dh);
381 		}
382 		spin_unlock_bh(&ht->lock);
383 		cond_resched();
384 	}
385 }
386 
387 static void htable_gc(struct work_struct *work)
388 {
389 	struct xt_hashlimit_htable *ht;
390 
391 	ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
392 
393 	htable_selective_cleanup(ht, select_gc);
394 
395 	queue_delayed_work(system_power_efficient_wq,
396 			   &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
397 }
398 
399 static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
400 {
401 	struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
402 	struct proc_dir_entry *parent;
403 
404 	if (hinfo->family == NFPROTO_IPV4)
405 		parent = hashlimit_net->ipt_hashlimit;
406 	else
407 		parent = hashlimit_net->ip6t_hashlimit;
408 
409 	if (parent != NULL)
410 		remove_proc_entry(hinfo->name, parent);
411 }
412 
413 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
414 {
415 	cancel_delayed_work_sync(&hinfo->gc_work);
416 	htable_remove_proc_entry(hinfo);
417 	htable_selective_cleanup(hinfo, select_all);
418 	kfree(hinfo->name);
419 	vfree(hinfo);
420 }
421 
422 static struct xt_hashlimit_htable *htable_find_get(struct net *net,
423 						   const char *name,
424 						   u_int8_t family)
425 {
426 	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
427 	struct xt_hashlimit_htable *hinfo;
428 
429 	hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
430 		if (!strcmp(name, hinfo->name) &&
431 		    hinfo->family == family) {
432 			hinfo->use++;
433 			return hinfo;
434 		}
435 	}
436 	return NULL;
437 }
438 
439 static void htable_put(struct xt_hashlimit_htable *hinfo)
440 {
441 	mutex_lock(&hashlimit_mutex);
442 	if (--hinfo->use == 0) {
443 		hlist_del(&hinfo->node);
444 		htable_destroy(hinfo);
445 	}
446 	mutex_unlock(&hashlimit_mutex);
447 }
448 
449 /* The algorithm used is the Simple Token Bucket Filter (TBF)
450  * see net/sched/sch_tbf.c in the linux source tree
451  */
452 
453 /* Rusty: This is my (non-mathematically-inclined) understanding of
454    this algorithm.  The `average rate' in jiffies becomes your initial
455    amount of credit `credit' and the most credit you can ever have
456    `credit_cap'.  The `peak rate' becomes the cost of passing the
457    test, `cost'.
458 
459    `prev' tracks the last packet hit: you gain one credit per jiffy.
460    If you get credit balance more than this, the extra credit is
461    discarded.  Every time the match passes, you lose `cost' credits;
462    if you don't have that many, the test fails.
463 
464    See Alexey's formal explanation in net/sched/sch_tbf.c.
465 
466    To get the maximum range, we multiply by this factor (ie. you get N
467    credits per jiffy).  We want to allow a rate as low as 1 per day
468    (slowest userspace tool allows), which means
469    CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
470 */
471 #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
472 #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
473 
474 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
475  * us the power of 2 below the theoretical max, so GCC simply does a
476  * shift. */
477 #define _POW2_BELOW2(x) ((x)|((x)>>1))
478 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
479 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
480 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
481 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
482 #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
483 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
484 #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
485 
486 #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
487 #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
488 
489 /* in byte mode, the lowest possible rate is one packet/second.
490  * credit_cap is used as a counter that tells us how many times we can
491  * refill the "credits available" counter when it becomes empty.
492  */
493 #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
494 #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
495 
496 static u32 xt_hashlimit_len_to_chunks(u32 len)
497 {
498 	return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
499 }
500 
501 /* Precision saver. */
502 static u64 user2credits(u64 user, int revision)
503 {
504 	u64 scale = (revision == 1) ?
505 		XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
506 	u64 cpj = (revision == 1) ?
507 		CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
508 
509 	/* Avoid overflow: divide the constant operands first */
510 	if (scale >= HZ * cpj)
511 		return div64_u64(user, div64_u64(scale, HZ * cpj));
512 
513 	return user * div64_u64(HZ * cpj, scale);
514 }
515 
516 static u32 user2credits_byte(u32 user)
517 {
518 	u64 us = user;
519 	us *= HZ * CREDITS_PER_JIFFY_BYTES;
520 	return (u32) (us >> 32);
521 }
522 
523 static u64 user2rate(u64 user)
524 {
525 	if (user != 0) {
526 		return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
527 	} else {
528 		pr_info_ratelimited("invalid rate from userspace: %llu\n",
529 				    user);
530 		return 0;
531 	}
532 }
533 
534 static u64 user2rate_bytes(u32 user)
535 {
536 	u64 r;
537 
538 	r = user ? U32_MAX / user : U32_MAX;
539 	return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
540 }
541 
542 static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
543 			    u32 mode, int revision)
544 {
545 	unsigned long delta = now - dh->rateinfo.prev;
546 	u64 cap, cpj;
547 
548 	if (delta == 0)
549 		return;
550 
551 	if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
552 		u64 interval = dh->rateinfo.interval * HZ;
553 
554 		if (delta < interval)
555 			return;
556 
557 		dh->rateinfo.prev = now;
558 		dh->rateinfo.prev_window =
559 			((dh->rateinfo.current_rate * interval) >
560 			 (delta * dh->rateinfo.rate));
561 		dh->rateinfo.current_rate = 0;
562 
563 		return;
564 	}
565 
566 	dh->rateinfo.prev = now;
567 
568 	if (mode & XT_HASHLIMIT_BYTES) {
569 		u64 tmp = dh->rateinfo.credit;
570 		dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
571 		cap = CREDITS_PER_JIFFY_BYTES * HZ;
572 		if (tmp >= dh->rateinfo.credit) {/* overflow */
573 			dh->rateinfo.credit = cap;
574 			return;
575 		}
576 	} else {
577 		cpj = (revision == 1) ?
578 			CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
579 		dh->rateinfo.credit += delta * cpj;
580 		cap = dh->rateinfo.credit_cap;
581 	}
582 	if (dh->rateinfo.credit > cap)
583 		dh->rateinfo.credit = cap;
584 }
585 
586 static void rateinfo_init(struct dsthash_ent *dh,
587 			  struct xt_hashlimit_htable *hinfo, int revision)
588 {
589 	dh->rateinfo.prev = jiffies;
590 	if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
591 		dh->rateinfo.prev_window = 0;
592 		dh->rateinfo.current_rate = 0;
593 		if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
594 			dh->rateinfo.rate =
595 				user2rate_bytes((u32)hinfo->cfg.avg);
596 			if (hinfo->cfg.burst)
597 				dh->rateinfo.burst =
598 					hinfo->cfg.burst * dh->rateinfo.rate;
599 			else
600 				dh->rateinfo.burst = dh->rateinfo.rate;
601 		} else {
602 			dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
603 			dh->rateinfo.burst =
604 				hinfo->cfg.burst + dh->rateinfo.rate;
605 		}
606 		dh->rateinfo.interval = hinfo->cfg.interval;
607 	} else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
608 		dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
609 		dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
610 		dh->rateinfo.credit_cap = hinfo->cfg.burst;
611 	} else {
612 		dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
613 						   hinfo->cfg.burst, revision);
614 		dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
615 		dh->rateinfo.credit_cap = dh->rateinfo.credit;
616 	}
617 }
618 
619 static inline __be32 maskl(__be32 a, unsigned int l)
620 {
621 	return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
622 }
623 
624 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
625 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
626 {
627 	switch (p) {
628 	case 0 ... 31:
629 		i[0] = maskl(i[0], p);
630 		i[1] = i[2] = i[3] = 0;
631 		break;
632 	case 32 ... 63:
633 		i[1] = maskl(i[1], p - 32);
634 		i[2] = i[3] = 0;
635 		break;
636 	case 64 ... 95:
637 		i[2] = maskl(i[2], p - 64);
638 		i[3] = 0;
639 		break;
640 	case 96 ... 127:
641 		i[3] = maskl(i[3], p - 96);
642 		break;
643 	case 128:
644 		break;
645 	}
646 }
647 #endif
648 
649 static int
650 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
651 		   struct dsthash_dst *dst,
652 		   const struct sk_buff *skb, unsigned int protoff)
653 {
654 	__be16 _ports[2], *ports;
655 	u8 nexthdr;
656 	int poff;
657 
658 	memset(dst, 0, sizeof(*dst));
659 
660 	switch (hinfo->family) {
661 	case NFPROTO_IPV4:
662 		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
663 			dst->ip.dst = maskl(ip_hdr(skb)->daddr,
664 			              hinfo->cfg.dstmask);
665 		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
666 			dst->ip.src = maskl(ip_hdr(skb)->saddr,
667 			              hinfo->cfg.srcmask);
668 
669 		if (!(hinfo->cfg.mode &
670 		      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
671 			return 0;
672 		nexthdr = ip_hdr(skb)->protocol;
673 		break;
674 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
675 	case NFPROTO_IPV6:
676 	{
677 		__be16 frag_off;
678 
679 		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
680 			memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
681 			       sizeof(dst->ip6.dst));
682 			hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
683 		}
684 		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
685 			memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
686 			       sizeof(dst->ip6.src));
687 			hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
688 		}
689 
690 		if (!(hinfo->cfg.mode &
691 		      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
692 			return 0;
693 		nexthdr = ipv6_hdr(skb)->nexthdr;
694 		protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
695 		if ((int)protoff < 0)
696 			return -1;
697 		break;
698 	}
699 #endif
700 	default:
701 		BUG();
702 		return 0;
703 	}
704 
705 	poff = proto_ports_offset(nexthdr);
706 	if (poff >= 0) {
707 		ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
708 					   &_ports);
709 	} else {
710 		_ports[0] = _ports[1] = 0;
711 		ports = _ports;
712 	}
713 	if (!ports)
714 		return -1;
715 	if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
716 		dst->src_port = ports[0];
717 	if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
718 		dst->dst_port = ports[1];
719 	return 0;
720 }
721 
722 static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
723 {
724 	u64 tmp = xt_hashlimit_len_to_chunks(len);
725 	tmp = tmp * dh->rateinfo.cost;
726 
727 	if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
728 		tmp = CREDITS_PER_JIFFY_BYTES * HZ;
729 
730 	if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
731 		dh->rateinfo.credit_cap--;
732 		dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
733 	}
734 	return (u32) tmp;
735 }
736 
737 static bool
738 hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
739 		    struct xt_hashlimit_htable *hinfo,
740 		    const struct hashlimit_cfg3 *cfg, int revision)
741 {
742 	unsigned long now = jiffies;
743 	struct dsthash_ent *dh;
744 	struct dsthash_dst dst;
745 	bool race = false;
746 	u64 cost;
747 
748 	if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
749 		goto hotdrop;
750 
751 	local_bh_disable();
752 	dh = dsthash_find(hinfo, &dst);
753 	if (dh == NULL) {
754 		dh = dsthash_alloc_init(hinfo, &dst, &race);
755 		if (dh == NULL) {
756 			local_bh_enable();
757 			goto hotdrop;
758 		} else if (race) {
759 			/* Already got an entry, update expiration timeout */
760 			dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
761 			rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
762 		} else {
763 			dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
764 			rateinfo_init(dh, hinfo, revision);
765 		}
766 	} else {
767 		/* update expiration timeout */
768 		dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
769 		rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
770 	}
771 
772 	if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
773 		cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
774 		dh->rateinfo.current_rate += cost;
775 
776 		if (!dh->rateinfo.prev_window &&
777 		    (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
778 			spin_unlock(&dh->lock);
779 			local_bh_enable();
780 			return !(cfg->mode & XT_HASHLIMIT_INVERT);
781 		} else {
782 			goto overlimit;
783 		}
784 	}
785 
786 	if (cfg->mode & XT_HASHLIMIT_BYTES)
787 		cost = hashlimit_byte_cost(skb->len, dh);
788 	else
789 		cost = dh->rateinfo.cost;
790 
791 	if (dh->rateinfo.credit >= cost) {
792 		/* below the limit */
793 		dh->rateinfo.credit -= cost;
794 		spin_unlock(&dh->lock);
795 		local_bh_enable();
796 		return !(cfg->mode & XT_HASHLIMIT_INVERT);
797 	}
798 
799 overlimit:
800 	spin_unlock(&dh->lock);
801 	local_bh_enable();
802 	/* default match is underlimit - so over the limit, we need to invert */
803 	return cfg->mode & XT_HASHLIMIT_INVERT;
804 
805  hotdrop:
806 	par->hotdrop = true;
807 	return false;
808 }
809 
810 static bool
811 hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
812 {
813 	const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
814 	struct xt_hashlimit_htable *hinfo = info->hinfo;
815 	struct hashlimit_cfg3 cfg = {};
816 	int ret;
817 
818 	ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
819 	if (ret)
820 		return ret;
821 
822 	return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
823 }
824 
825 static bool
826 hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
827 {
828 	const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
829 	struct xt_hashlimit_htable *hinfo = info->hinfo;
830 	struct hashlimit_cfg3 cfg = {};
831 	int ret;
832 
833 	ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
834 	if (ret)
835 		return ret;
836 
837 	return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
838 }
839 
840 static bool
841 hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
842 {
843 	const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
844 	struct xt_hashlimit_htable *hinfo = info->hinfo;
845 
846 	return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
847 }
848 
849 static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
850 				     struct xt_hashlimit_htable **hinfo,
851 				     struct hashlimit_cfg3 *cfg,
852 				     const char *name, int revision)
853 {
854 	struct net *net = par->net;
855 	int ret;
856 
857 	if (cfg->gc_interval == 0 || cfg->expire == 0)
858 		return -EINVAL;
859 	if (par->family == NFPROTO_IPV4) {
860 		if (cfg->srcmask > 32 || cfg->dstmask > 32)
861 			return -EINVAL;
862 	} else {
863 		if (cfg->srcmask > 128 || cfg->dstmask > 128)
864 			return -EINVAL;
865 	}
866 
867 	if (cfg->mode & ~XT_HASHLIMIT_ALL) {
868 		pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
869 				    cfg->mode);
870 		return -EINVAL;
871 	}
872 
873 	/* Check for overflow. */
874 	if (revision >= 3 && cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
875 		if (cfg->avg == 0 || cfg->avg > U32_MAX) {
876 			pr_info_ratelimited("invalid rate\n");
877 			return -ERANGE;
878 		}
879 
880 		if (cfg->interval == 0) {
881 			pr_info_ratelimited("invalid interval\n");
882 			return -EINVAL;
883 		}
884 	} else if (cfg->mode & XT_HASHLIMIT_BYTES) {
885 		if (user2credits_byte(cfg->avg) == 0) {
886 			pr_info_ratelimited("overflow, rate too high: %llu\n",
887 					    cfg->avg);
888 			return -EINVAL;
889 		}
890 	} else if (cfg->burst == 0 ||
891 		   user2credits(cfg->avg * cfg->burst, revision) <
892 		   user2credits(cfg->avg, revision)) {
893 		pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
894 				    cfg->avg, cfg->burst);
895 		return -ERANGE;
896 	}
897 
898 	mutex_lock(&hashlimit_mutex);
899 	*hinfo = htable_find_get(net, name, par->family);
900 	if (*hinfo == NULL) {
901 		ret = htable_create(net, cfg, name, par->family,
902 				    hinfo, revision);
903 		if (ret < 0) {
904 			mutex_unlock(&hashlimit_mutex);
905 			return ret;
906 		}
907 	}
908 	mutex_unlock(&hashlimit_mutex);
909 
910 	return 0;
911 }
912 
913 static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
914 {
915 	struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
916 	struct hashlimit_cfg3 cfg = {};
917 	int ret;
918 
919 	ret = xt_check_proc_name(info->name, sizeof(info->name));
920 	if (ret)
921 		return ret;
922 
923 	ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
924 	if (ret)
925 		return ret;
926 
927 	return hashlimit_mt_check_common(par, &info->hinfo,
928 					 &cfg, info->name, 1);
929 }
930 
931 static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
932 {
933 	struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
934 	struct hashlimit_cfg3 cfg = {};
935 	int ret;
936 
937 	ret = xt_check_proc_name(info->name, sizeof(info->name));
938 	if (ret)
939 		return ret;
940 
941 	ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
942 	if (ret)
943 		return ret;
944 
945 	return hashlimit_mt_check_common(par, &info->hinfo,
946 					 &cfg, info->name, 2);
947 }
948 
949 static int hashlimit_mt_check(const struct xt_mtchk_param *par)
950 {
951 	struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
952 	int ret;
953 
954 	ret = xt_check_proc_name(info->name, sizeof(info->name));
955 	if (ret)
956 		return ret;
957 
958 	return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
959 					 info->name, 3);
960 }
961 
962 static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
963 {
964 	const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
965 
966 	htable_put(info->hinfo);
967 }
968 
969 static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
970 {
971 	const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
972 
973 	htable_put(info->hinfo);
974 }
975 
976 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
977 {
978 	const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
979 
980 	htable_put(info->hinfo);
981 }
982 
983 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
984 	{
985 		.name           = "hashlimit",
986 		.revision       = 1,
987 		.family         = NFPROTO_IPV4,
988 		.match          = hashlimit_mt_v1,
989 		.matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
990 		.usersize	= offsetof(struct xt_hashlimit_mtinfo1, hinfo),
991 		.checkentry     = hashlimit_mt_check_v1,
992 		.destroy        = hashlimit_mt_destroy_v1,
993 		.me             = THIS_MODULE,
994 	},
995 	{
996 		.name           = "hashlimit",
997 		.revision       = 2,
998 		.family         = NFPROTO_IPV4,
999 		.match          = hashlimit_mt_v2,
1000 		.matchsize      = sizeof(struct xt_hashlimit_mtinfo2),
1001 		.usersize	= offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1002 		.checkentry     = hashlimit_mt_check_v2,
1003 		.destroy        = hashlimit_mt_destroy_v2,
1004 		.me             = THIS_MODULE,
1005 	},
1006 	{
1007 		.name           = "hashlimit",
1008 		.revision       = 3,
1009 		.family         = NFPROTO_IPV4,
1010 		.match          = hashlimit_mt,
1011 		.matchsize      = sizeof(struct xt_hashlimit_mtinfo3),
1012 		.usersize	= offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1013 		.checkentry     = hashlimit_mt_check,
1014 		.destroy        = hashlimit_mt_destroy,
1015 		.me             = THIS_MODULE,
1016 	},
1017 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1018 	{
1019 		.name           = "hashlimit",
1020 		.revision       = 1,
1021 		.family         = NFPROTO_IPV6,
1022 		.match          = hashlimit_mt_v1,
1023 		.matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
1024 		.usersize	= offsetof(struct xt_hashlimit_mtinfo1, hinfo),
1025 		.checkentry     = hashlimit_mt_check_v1,
1026 		.destroy        = hashlimit_mt_destroy_v1,
1027 		.me             = THIS_MODULE,
1028 	},
1029 	{
1030 		.name           = "hashlimit",
1031 		.revision       = 2,
1032 		.family         = NFPROTO_IPV6,
1033 		.match          = hashlimit_mt_v2,
1034 		.matchsize      = sizeof(struct xt_hashlimit_mtinfo2),
1035 		.usersize	= offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1036 		.checkentry     = hashlimit_mt_check_v2,
1037 		.destroy        = hashlimit_mt_destroy_v2,
1038 		.me             = THIS_MODULE,
1039 	},
1040 	{
1041 		.name           = "hashlimit",
1042 		.revision       = 3,
1043 		.family         = NFPROTO_IPV6,
1044 		.match          = hashlimit_mt,
1045 		.matchsize      = sizeof(struct xt_hashlimit_mtinfo3),
1046 		.usersize	= offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1047 		.checkentry     = hashlimit_mt_check,
1048 		.destroy        = hashlimit_mt_destroy,
1049 		.me             = THIS_MODULE,
1050 	},
1051 #endif
1052 };
1053 
1054 /* PROC stuff */
1055 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1056 	__acquires(htable->lock)
1057 {
1058 	struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1059 	unsigned int *bucket;
1060 
1061 	spin_lock_bh(&htable->lock);
1062 	if (*pos >= htable->cfg.size)
1063 		return NULL;
1064 
1065 	bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1066 	if (!bucket)
1067 		return ERR_PTR(-ENOMEM);
1068 
1069 	*bucket = *pos;
1070 	return bucket;
1071 }
1072 
1073 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1074 {
1075 	struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1076 	unsigned int *bucket = v;
1077 
1078 	*pos = ++(*bucket);
1079 	if (*pos >= htable->cfg.size) {
1080 		kfree(v);
1081 		return NULL;
1082 	}
1083 	return bucket;
1084 }
1085 
1086 static void dl_seq_stop(struct seq_file *s, void *v)
1087 	__releases(htable->lock)
1088 {
1089 	struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1090 	unsigned int *bucket = v;
1091 
1092 	if (!IS_ERR(bucket))
1093 		kfree(bucket);
1094 	spin_unlock_bh(&htable->lock);
1095 }
1096 
1097 static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1098 			 struct seq_file *s)
1099 {
1100 	switch (family) {
1101 	case NFPROTO_IPV4:
1102 		seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1103 			   (long)(ent->expires - jiffies)/HZ,
1104 			   &ent->dst.ip.src,
1105 			   ntohs(ent->dst.src_port),
1106 			   &ent->dst.ip.dst,
1107 			   ntohs(ent->dst.dst_port),
1108 			   ent->rateinfo.credit, ent->rateinfo.credit_cap,
1109 			   ent->rateinfo.cost);
1110 		break;
1111 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1112 	case NFPROTO_IPV6:
1113 		seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1114 			   (long)(ent->expires - jiffies)/HZ,
1115 			   &ent->dst.ip6.src,
1116 			   ntohs(ent->dst.src_port),
1117 			   &ent->dst.ip6.dst,
1118 			   ntohs(ent->dst.dst_port),
1119 			   ent->rateinfo.credit, ent->rateinfo.credit_cap,
1120 			   ent->rateinfo.cost);
1121 		break;
1122 #endif
1123 	default:
1124 		BUG();
1125 	}
1126 }
1127 
1128 static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1129 			       struct seq_file *s)
1130 {
1131 	struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1132 
1133 	spin_lock(&ent->lock);
1134 	/* recalculate to show accurate numbers */
1135 	rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1136 
1137 	dl_seq_print(ent, family, s);
1138 
1139 	spin_unlock(&ent->lock);
1140 	return seq_has_overflowed(s);
1141 }
1142 
1143 static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1144 			       struct seq_file *s)
1145 {
1146 	struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1147 
1148 	spin_lock(&ent->lock);
1149 	/* recalculate to show accurate numbers */
1150 	rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1151 
1152 	dl_seq_print(ent, family, s);
1153 
1154 	spin_unlock(&ent->lock);
1155 	return seq_has_overflowed(s);
1156 }
1157 
1158 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1159 			    struct seq_file *s)
1160 {
1161 	struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1162 
1163 	spin_lock(&ent->lock);
1164 	/* recalculate to show accurate numbers */
1165 	rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1166 
1167 	dl_seq_print(ent, family, s);
1168 
1169 	spin_unlock(&ent->lock);
1170 	return seq_has_overflowed(s);
1171 }
1172 
1173 static int dl_seq_show_v2(struct seq_file *s, void *v)
1174 {
1175 	struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1176 	unsigned int *bucket = (unsigned int *)v;
1177 	struct dsthash_ent *ent;
1178 
1179 	if (!hlist_empty(&htable->hash[*bucket])) {
1180 		hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1181 			if (dl_seq_real_show_v2(ent, htable->family, s))
1182 				return -1;
1183 	}
1184 	return 0;
1185 }
1186 
1187 static int dl_seq_show_v1(struct seq_file *s, void *v)
1188 {
1189 	struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1190 	unsigned int *bucket = v;
1191 	struct dsthash_ent *ent;
1192 
1193 	if (!hlist_empty(&htable->hash[*bucket])) {
1194 		hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1195 			if (dl_seq_real_show_v1(ent, htable->family, s))
1196 				return -1;
1197 	}
1198 	return 0;
1199 }
1200 
1201 static int dl_seq_show(struct seq_file *s, void *v)
1202 {
1203 	struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1204 	unsigned int *bucket = v;
1205 	struct dsthash_ent *ent;
1206 
1207 	if (!hlist_empty(&htable->hash[*bucket])) {
1208 		hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1209 			if (dl_seq_real_show(ent, htable->family, s))
1210 				return -1;
1211 	}
1212 	return 0;
1213 }
1214 
1215 static const struct seq_operations dl_seq_ops_v1 = {
1216 	.start = dl_seq_start,
1217 	.next  = dl_seq_next,
1218 	.stop  = dl_seq_stop,
1219 	.show  = dl_seq_show_v1
1220 };
1221 
1222 static const struct seq_operations dl_seq_ops_v2 = {
1223 	.start = dl_seq_start,
1224 	.next  = dl_seq_next,
1225 	.stop  = dl_seq_stop,
1226 	.show  = dl_seq_show_v2
1227 };
1228 
1229 static const struct seq_operations dl_seq_ops = {
1230 	.start = dl_seq_start,
1231 	.next  = dl_seq_next,
1232 	.stop  = dl_seq_stop,
1233 	.show  = dl_seq_show
1234 };
1235 
1236 static int __net_init hashlimit_proc_net_init(struct net *net)
1237 {
1238 	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1239 
1240 	hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1241 	if (!hashlimit_net->ipt_hashlimit)
1242 		return -ENOMEM;
1243 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1244 	hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1245 	if (!hashlimit_net->ip6t_hashlimit) {
1246 		remove_proc_entry("ipt_hashlimit", net->proc_net);
1247 		return -ENOMEM;
1248 	}
1249 #endif
1250 	return 0;
1251 }
1252 
1253 static void __net_exit hashlimit_proc_net_exit(struct net *net)
1254 {
1255 	struct xt_hashlimit_htable *hinfo;
1256 	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1257 
1258 	/* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1259 	 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1260 	 * entries is empty before trying to remove it.
1261 	 */
1262 	mutex_lock(&hashlimit_mutex);
1263 	hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1264 		htable_remove_proc_entry(hinfo);
1265 	hashlimit_net->ipt_hashlimit = NULL;
1266 	hashlimit_net->ip6t_hashlimit = NULL;
1267 	mutex_unlock(&hashlimit_mutex);
1268 
1269 	remove_proc_entry("ipt_hashlimit", net->proc_net);
1270 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1271 	remove_proc_entry("ip6t_hashlimit", net->proc_net);
1272 #endif
1273 }
1274 
1275 static int __net_init hashlimit_net_init(struct net *net)
1276 {
1277 	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1278 
1279 	INIT_HLIST_HEAD(&hashlimit_net->htables);
1280 	return hashlimit_proc_net_init(net);
1281 }
1282 
1283 static void __net_exit hashlimit_net_exit(struct net *net)
1284 {
1285 	hashlimit_proc_net_exit(net);
1286 }
1287 
1288 static struct pernet_operations hashlimit_net_ops = {
1289 	.init	= hashlimit_net_init,
1290 	.exit	= hashlimit_net_exit,
1291 	.id	= &hashlimit_net_id,
1292 	.size	= sizeof(struct hashlimit_net),
1293 };
1294 
1295 static int __init hashlimit_mt_init(void)
1296 {
1297 	int err;
1298 
1299 	err = register_pernet_subsys(&hashlimit_net_ops);
1300 	if (err < 0)
1301 		return err;
1302 	err = xt_register_matches(hashlimit_mt_reg,
1303 	      ARRAY_SIZE(hashlimit_mt_reg));
1304 	if (err < 0)
1305 		goto err1;
1306 
1307 	err = -ENOMEM;
1308 	hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1309 					    sizeof(struct dsthash_ent), 0, 0,
1310 					    NULL);
1311 	if (!hashlimit_cachep) {
1312 		pr_warn("unable to create slab cache\n");
1313 		goto err2;
1314 	}
1315 	return 0;
1316 
1317 err2:
1318 	xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1319 err1:
1320 	unregister_pernet_subsys(&hashlimit_net_ops);
1321 	return err;
1322 
1323 }
1324 
1325 static void __exit hashlimit_mt_exit(void)
1326 {
1327 	xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1328 	unregister_pernet_subsys(&hashlimit_net_ops);
1329 
1330 	rcu_barrier();
1331 	kmem_cache_destroy(hashlimit_cachep);
1332 }
1333 
1334 module_init(hashlimit_mt_init);
1335 module_exit(hashlimit_mt_exit);
1336