Linux Audio

Check our new training course

Loading...
v5.14.15
   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);
v4.17
 
   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
  40MODULE_LICENSE("GPL");
  41MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
  42MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
  43MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
  44MODULE_ALIAS("ipt_hashlimit");
  45MODULE_ALIAS("ip6t_hashlimit");
  46
  47struct hashlimit_net {
  48	struct hlist_head	htables;
  49	struct proc_dir_entry	*ipt_hashlimit;
  50	struct proc_dir_entry	*ip6t_hashlimit;
  51};
  52
  53static unsigned int hashlimit_net_id;
  54static 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 */
  60static const struct file_operations dl_file_ops_v2;
  61static const struct file_operations dl_file_ops_v1;
  62static const struct file_operations dl_file_ops;
  63
  64/* hash table crap */
  65struct 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
  82struct 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
 109struct 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
 131static int
 132cfg_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
 167static DEFINE_MUTEX(hashlimit_mutex);	/* protects htables list */
 168static struct kmem_cache *hashlimit_cachep __read_mostly;
 169
 170static 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
 176static u_int32_t
 177hash_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
 191static struct dsthash_ent *
 192dsthash_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 */
 209static struct dsthash_ent *
 210dsthash_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
 252static 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
 259static inline void
 260dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
 261{
 262	hlist_del_rcu(&ent->node);
 263	call_rcu_bh(&ent->rcu, dsthash_free_rcu);
 264	ht->count--;
 265}
 266static void htable_gc(struct work_struct *work);
 267
 268static 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 file_operations *fops;
 276	unsigned int size, i;
 
 277	int ret;
 278
 279	if (cfg->size) {
 280		size = cfg->size;
 281	} else {
 282		size = (totalram_pages << PAGE_SHIFT) / 16384 /
 283		       sizeof(struct hlist_head);
 284		if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
 285			size = 8192;
 286		if (size < 16)
 287			size = 16;
 288	}
 289	/* FIXME: don't use vmalloc() here or anywhere else -HW */
 290	hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
 291	                sizeof(struct hlist_head) * size);
 292	if (hinfo == NULL)
 293		return -ENOMEM;
 294	*out_hinfo = hinfo;
 295
 296	/* copy match config into hashtable config */
 297	ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
 298
 299	if (ret)
 300		return ret;
 
 301
 302	hinfo->cfg.size = size;
 303	if (hinfo->cfg.max == 0)
 304		hinfo->cfg.max = 8 * hinfo->cfg.size;
 305	else if (hinfo->cfg.max < hinfo->cfg.size)
 306		hinfo->cfg.max = hinfo->cfg.size;
 307
 308	for (i = 0; i < hinfo->cfg.size; i++)
 309		INIT_HLIST_HEAD(&hinfo->hash[i]);
 310
 311	hinfo->use = 1;
 312	hinfo->count = 0;
 313	hinfo->family = family;
 314	hinfo->rnd_initialized = false;
 315	hinfo->name = kstrdup(name, GFP_KERNEL);
 316	if (!hinfo->name) {
 317		vfree(hinfo);
 318		return -ENOMEM;
 319	}
 320	spin_lock_init(&hinfo->lock);
 321
 322	switch (revision) {
 323	case 1:
 324		fops = &dl_file_ops_v1;
 325		break;
 326	case 2:
 327		fops = &dl_file_ops_v2;
 328		break;
 329	default:
 330		fops = &dl_file_ops;
 331	}
 332
 333	hinfo->pde = proc_create_data(name, 0,
 334		(family == NFPROTO_IPV4) ?
 335		hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
 336		fops, hinfo);
 337	if (hinfo->pde == NULL) {
 338		kfree(hinfo->name);
 339		vfree(hinfo);
 340		return -ENOMEM;
 341	}
 342	hinfo->net = net;
 343
 344	INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
 345	queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
 346			   msecs_to_jiffies(hinfo->cfg.gc_interval));
 347
 348	hlist_add_head(&hinfo->node, &hashlimit_net->htables);
 349
 350	return 0;
 351}
 352
 353static bool select_all(const struct xt_hashlimit_htable *ht,
 354		       const struct dsthash_ent *he)
 355{
 356	return true;
 357}
 358
 359static bool select_gc(const struct xt_hashlimit_htable *ht,
 360		      const struct dsthash_ent *he)
 361{
 362	return time_after_eq(jiffies, he->expires);
 363}
 364
 365static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
 366			bool (*select)(const struct xt_hashlimit_htable *ht,
 367				      const struct dsthash_ent *he))
 368{
 369	unsigned int i;
 370
 371	for (i = 0; i < ht->cfg.size; i++) {
 372		struct dsthash_ent *dh;
 373		struct hlist_node *n;
 374
 375		spin_lock_bh(&ht->lock);
 376		hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
 377			if ((*select)(ht, dh))
 378				dsthash_free(ht, dh);
 379		}
 380		spin_unlock_bh(&ht->lock);
 381		cond_resched();
 382	}
 383}
 384
 385static void htable_gc(struct work_struct *work)
 386{
 387	struct xt_hashlimit_htable *ht;
 388
 389	ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
 390
 391	htable_selective_cleanup(ht, select_gc);
 392
 393	queue_delayed_work(system_power_efficient_wq,
 394			   &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
 395}
 396
 397static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
 398{
 399	struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
 400	struct proc_dir_entry *parent;
 401
 402	if (hinfo->family == NFPROTO_IPV4)
 403		parent = hashlimit_net->ipt_hashlimit;
 404	else
 405		parent = hashlimit_net->ip6t_hashlimit;
 406
 407	if (parent != NULL)
 408		remove_proc_entry(hinfo->name, parent);
 409}
 410
 411static void htable_destroy(struct xt_hashlimit_htable *hinfo)
 412{
 413	cancel_delayed_work_sync(&hinfo->gc_work);
 414	htable_remove_proc_entry(hinfo);
 415	htable_selective_cleanup(hinfo, select_all);
 416	kfree(hinfo->name);
 417	vfree(hinfo);
 418}
 419
 420static struct xt_hashlimit_htable *htable_find_get(struct net *net,
 421						   const char *name,
 422						   u_int8_t family)
 423{
 424	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 425	struct xt_hashlimit_htable *hinfo;
 426
 427	hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
 428		if (!strcmp(name, hinfo->name) &&
 429		    hinfo->family == family) {
 430			hinfo->use++;
 431			return hinfo;
 432		}
 433	}
 434	return NULL;
 435}
 436
 437static void htable_put(struct xt_hashlimit_htable *hinfo)
 438{
 439	mutex_lock(&hashlimit_mutex);
 440	if (--hinfo->use == 0) {
 441		hlist_del(&hinfo->node);
 442		htable_destroy(hinfo);
 
 
 
 
 
 
 443	}
 444	mutex_unlock(&hashlimit_mutex);
 445}
 446
 447/* The algorithm used is the Simple Token Bucket Filter (TBF)
 448 * see net/sched/sch_tbf.c in the linux source tree
 449 */
 450
 451/* Rusty: This is my (non-mathematically-inclined) understanding of
 452   this algorithm.  The `average rate' in jiffies becomes your initial
 453   amount of credit `credit' and the most credit you can ever have
 454   `credit_cap'.  The `peak rate' becomes the cost of passing the
 455   test, `cost'.
 456
 457   `prev' tracks the last packet hit: you gain one credit per jiffy.
 458   If you get credit balance more than this, the extra credit is
 459   discarded.  Every time the match passes, you lose `cost' credits;
 460   if you don't have that many, the test fails.
 461
 462   See Alexey's formal explanation in net/sched/sch_tbf.c.
 463
 464   To get the maximum range, we multiply by this factor (ie. you get N
 465   credits per jiffy).  We want to allow a rate as low as 1 per day
 466   (slowest userspace tool allows), which means
 467   CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
 468*/
 469#define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
 470#define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
 471
 472/* Repeated shift and or gives us all 1s, final shift and add 1 gives
 473 * us the power of 2 below the theoretical max, so GCC simply does a
 474 * shift. */
 475#define _POW2_BELOW2(x) ((x)|((x)>>1))
 476#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
 477#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
 478#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
 479#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
 480#define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
 481#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
 482#define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
 483
 484#define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
 485#define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
 486
 487/* in byte mode, the lowest possible rate is one packet/second.
 488 * credit_cap is used as a counter that tells us how many times we can
 489 * refill the "credits available" counter when it becomes empty.
 490 */
 491#define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
 492#define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
 493
 494static u32 xt_hashlimit_len_to_chunks(u32 len)
 495{
 496	return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
 497}
 498
 499/* Precision saver. */
 500static u64 user2credits(u64 user, int revision)
 501{
 502	u64 scale = (revision == 1) ?
 503		XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
 504	u64 cpj = (revision == 1) ?
 505		CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
 506
 507	/* Avoid overflow: divide the constant operands first */
 508	if (scale >= HZ * cpj)
 509		return div64_u64(user, div64_u64(scale, HZ * cpj));
 510
 511	return user * div64_u64(HZ * cpj, scale);
 512}
 513
 514static u32 user2credits_byte(u32 user)
 515{
 516	u64 us = user;
 517	us *= HZ * CREDITS_PER_JIFFY_BYTES;
 518	return (u32) (us >> 32);
 519}
 520
 521static u64 user2rate(u64 user)
 522{
 523	if (user != 0) {
 524		return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
 525	} else {
 526		pr_info_ratelimited("invalid rate from userspace: %llu\n",
 527				    user);
 528		return 0;
 529	}
 530}
 531
 532static u64 user2rate_bytes(u32 user)
 533{
 534	u64 r;
 535
 536	r = user ? U32_MAX / user : U32_MAX;
 537	return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
 538}
 539
 540static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
 541			    u32 mode, int revision)
 542{
 543	unsigned long delta = now - dh->rateinfo.prev;
 544	u64 cap, cpj;
 545
 546	if (delta == 0)
 547		return;
 548
 549	if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
 550		u64 interval = dh->rateinfo.interval * HZ;
 551
 552		if (delta < interval)
 553			return;
 554
 555		dh->rateinfo.prev = now;
 556		dh->rateinfo.prev_window =
 557			((dh->rateinfo.current_rate * interval) >
 558			 (delta * dh->rateinfo.rate));
 559		dh->rateinfo.current_rate = 0;
 560
 561		return;
 562	}
 563
 564	dh->rateinfo.prev = now;
 565
 566	if (mode & XT_HASHLIMIT_BYTES) {
 567		u64 tmp = dh->rateinfo.credit;
 568		dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
 569		cap = CREDITS_PER_JIFFY_BYTES * HZ;
 570		if (tmp >= dh->rateinfo.credit) {/* overflow */
 571			dh->rateinfo.credit = cap;
 572			return;
 573		}
 574	} else {
 575		cpj = (revision == 1) ?
 576			CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
 577		dh->rateinfo.credit += delta * cpj;
 578		cap = dh->rateinfo.credit_cap;
 579	}
 580	if (dh->rateinfo.credit > cap)
 581		dh->rateinfo.credit = cap;
 582}
 583
 584static void rateinfo_init(struct dsthash_ent *dh,
 585			  struct xt_hashlimit_htable *hinfo, int revision)
 586{
 587	dh->rateinfo.prev = jiffies;
 588	if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
 589		dh->rateinfo.prev_window = 0;
 590		dh->rateinfo.current_rate = 0;
 591		if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
 592			dh->rateinfo.rate =
 593				user2rate_bytes((u32)hinfo->cfg.avg);
 594			if (hinfo->cfg.burst)
 595				dh->rateinfo.burst =
 596					hinfo->cfg.burst * dh->rateinfo.rate;
 597			else
 598				dh->rateinfo.burst = dh->rateinfo.rate;
 599		} else {
 600			dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
 601			dh->rateinfo.burst =
 602				hinfo->cfg.burst + dh->rateinfo.rate;
 603		}
 604		dh->rateinfo.interval = hinfo->cfg.interval;
 605	} else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
 606		dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
 607		dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
 608		dh->rateinfo.credit_cap = hinfo->cfg.burst;
 609	} else {
 610		dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
 611						   hinfo->cfg.burst, revision);
 612		dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
 613		dh->rateinfo.credit_cap = dh->rateinfo.credit;
 614	}
 615}
 616
 617static inline __be32 maskl(__be32 a, unsigned int l)
 618{
 619	return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
 620}
 621
 622#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 623static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
 624{
 625	switch (p) {
 626	case 0 ... 31:
 627		i[0] = maskl(i[0], p);
 628		i[1] = i[2] = i[3] = 0;
 629		break;
 630	case 32 ... 63:
 631		i[1] = maskl(i[1], p - 32);
 632		i[2] = i[3] = 0;
 633		break;
 634	case 64 ... 95:
 635		i[2] = maskl(i[2], p - 64);
 636		i[3] = 0;
 637		break;
 638	case 96 ... 127:
 639		i[3] = maskl(i[3], p - 96);
 640		break;
 641	case 128:
 642		break;
 643	}
 644}
 645#endif
 646
 647static int
 648hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
 649		   struct dsthash_dst *dst,
 650		   const struct sk_buff *skb, unsigned int protoff)
 651{
 652	__be16 _ports[2], *ports;
 653	u8 nexthdr;
 654	int poff;
 655
 656	memset(dst, 0, sizeof(*dst));
 657
 658	switch (hinfo->family) {
 659	case NFPROTO_IPV4:
 660		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
 661			dst->ip.dst = maskl(ip_hdr(skb)->daddr,
 662			              hinfo->cfg.dstmask);
 663		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
 664			dst->ip.src = maskl(ip_hdr(skb)->saddr,
 665			              hinfo->cfg.srcmask);
 666
 667		if (!(hinfo->cfg.mode &
 668		      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 669			return 0;
 670		nexthdr = ip_hdr(skb)->protocol;
 671		break;
 672#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 673	case NFPROTO_IPV6:
 674	{
 675		__be16 frag_off;
 676
 677		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
 678			memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
 679			       sizeof(dst->ip6.dst));
 680			hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
 681		}
 682		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
 683			memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
 684			       sizeof(dst->ip6.src));
 685			hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
 686		}
 687
 688		if (!(hinfo->cfg.mode &
 689		      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 690			return 0;
 691		nexthdr = ipv6_hdr(skb)->nexthdr;
 692		protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
 693		if ((int)protoff < 0)
 694			return -1;
 695		break;
 696	}
 697#endif
 698	default:
 699		BUG();
 700		return 0;
 701	}
 702
 703	poff = proto_ports_offset(nexthdr);
 704	if (poff >= 0) {
 705		ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
 706					   &_ports);
 707	} else {
 708		_ports[0] = _ports[1] = 0;
 709		ports = _ports;
 710	}
 711	if (!ports)
 712		return -1;
 713	if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
 714		dst->src_port = ports[0];
 715	if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
 716		dst->dst_port = ports[1];
 717	return 0;
 718}
 719
 720static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
 721{
 722	u64 tmp = xt_hashlimit_len_to_chunks(len);
 723	tmp = tmp * dh->rateinfo.cost;
 724
 725	if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
 726		tmp = CREDITS_PER_JIFFY_BYTES * HZ;
 727
 728	if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
 729		dh->rateinfo.credit_cap--;
 730		dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
 731	}
 732	return (u32) tmp;
 733}
 734
 735static bool
 736hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
 737		    struct xt_hashlimit_htable *hinfo,
 738		    const struct hashlimit_cfg3 *cfg, int revision)
 739{
 740	unsigned long now = jiffies;
 741	struct dsthash_ent *dh;
 742	struct dsthash_dst dst;
 743	bool race = false;
 744	u64 cost;
 745
 746	if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
 747		goto hotdrop;
 748
 749	local_bh_disable();
 750	dh = dsthash_find(hinfo, &dst);
 751	if (dh == NULL) {
 752		dh = dsthash_alloc_init(hinfo, &dst, &race);
 753		if (dh == NULL) {
 754			local_bh_enable();
 755			goto hotdrop;
 756		} else if (race) {
 757			/* Already got an entry, update expiration timeout */
 758			dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 759			rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
 760		} else {
 761			dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
 762			rateinfo_init(dh, hinfo, revision);
 763		}
 764	} else {
 765		/* update expiration timeout */
 766		dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 767		rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
 768	}
 769
 770	if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
 771		cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
 772		dh->rateinfo.current_rate += cost;
 773
 774		if (!dh->rateinfo.prev_window &&
 775		    (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
 776			spin_unlock(&dh->lock);
 777			local_bh_enable();
 778			return !(cfg->mode & XT_HASHLIMIT_INVERT);
 779		} else {
 780			goto overlimit;
 781		}
 782	}
 783
 784	if (cfg->mode & XT_HASHLIMIT_BYTES)
 785		cost = hashlimit_byte_cost(skb->len, dh);
 786	else
 787		cost = dh->rateinfo.cost;
 788
 789	if (dh->rateinfo.credit >= cost) {
 790		/* below the limit */
 791		dh->rateinfo.credit -= cost;
 792		spin_unlock(&dh->lock);
 793		local_bh_enable();
 794		return !(cfg->mode & XT_HASHLIMIT_INVERT);
 795	}
 796
 797overlimit:
 798	spin_unlock(&dh->lock);
 799	local_bh_enable();
 800	/* default match is underlimit - so over the limit, we need to invert */
 801	return cfg->mode & XT_HASHLIMIT_INVERT;
 802
 803 hotdrop:
 804	par->hotdrop = true;
 805	return false;
 806}
 807
 808static bool
 809hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
 810{
 811	const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 812	struct xt_hashlimit_htable *hinfo = info->hinfo;
 813	struct hashlimit_cfg3 cfg = {};
 814	int ret;
 815
 816	ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
 817
 818	if (ret)
 819		return ret;
 820
 821	return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
 822}
 823
 824static bool
 825hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
 826{
 827	const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
 828	struct xt_hashlimit_htable *hinfo = info->hinfo;
 829	struct hashlimit_cfg3 cfg = {};
 830	int ret;
 831
 832	ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
 833
 834	if (ret)
 835		return ret;
 836
 837	return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
 838}
 839
 840static bool
 841hashlimit_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
 
 
 849static 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
 913static 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
 925	if (ret)
 926		return ret;
 927
 928	return hashlimit_mt_check_common(par, &info->hinfo,
 929					 &cfg, info->name, 1);
 930}
 931
 932static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
 933{
 934	struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
 935	struct hashlimit_cfg3 cfg = {};
 936	int ret;
 937
 938	ret = xt_check_proc_name(info->name, sizeof(info->name));
 939	if (ret)
 940		return ret;
 941
 942	ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
 943
 944	if (ret)
 945		return ret;
 946
 947	return hashlimit_mt_check_common(par, &info->hinfo,
 948					 &cfg, info->name, 2);
 949}
 950
 951static int hashlimit_mt_check(const struct xt_mtchk_param *par)
 952{
 953	struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
 954	int ret;
 955
 956	ret = xt_check_proc_name(info->name, sizeof(info->name));
 957	if (ret)
 958		return ret;
 959
 960	return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
 961					 info->name, 3);
 962}
 963
 964static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
 965{
 966	const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
 967
 968	htable_put(info->hinfo);
 969}
 970
 971static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
 972{
 973	const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 974
 975	htable_put(info->hinfo);
 976}
 977
 978static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
 979{
 980	const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
 981
 982	htable_put(info->hinfo);
 983}
 984
 985static struct xt_match hashlimit_mt_reg[] __read_mostly = {
 986	{
 987		.name           = "hashlimit",
 988		.revision       = 1,
 989		.family         = NFPROTO_IPV4,
 990		.match          = hashlimit_mt_v1,
 991		.matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 992		.usersize	= offsetof(struct xt_hashlimit_mtinfo1, hinfo),
 993		.checkentry     = hashlimit_mt_check_v1,
 994		.destroy        = hashlimit_mt_destroy_v1,
 995		.me             = THIS_MODULE,
 996	},
 997	{
 998		.name           = "hashlimit",
 999		.revision       = 2,
1000		.family         = NFPROTO_IPV4,
1001		.match          = hashlimit_mt_v2,
1002		.matchsize      = sizeof(struct xt_hashlimit_mtinfo2),
1003		.usersize	= offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1004		.checkentry     = hashlimit_mt_check_v2,
1005		.destroy        = hashlimit_mt_destroy_v2,
1006		.me             = THIS_MODULE,
1007	},
1008	{
1009		.name           = "hashlimit",
1010		.revision       = 3,
1011		.family         = NFPROTO_IPV4,
1012		.match          = hashlimit_mt,
1013		.matchsize      = sizeof(struct xt_hashlimit_mtinfo3),
1014		.usersize	= offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1015		.checkentry     = hashlimit_mt_check,
1016		.destroy        = hashlimit_mt_destroy,
1017		.me             = THIS_MODULE,
1018	},
1019#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1020	{
1021		.name           = "hashlimit",
1022		.revision       = 1,
1023		.family         = NFPROTO_IPV6,
1024		.match          = hashlimit_mt_v1,
1025		.matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
1026		.usersize	= offsetof(struct xt_hashlimit_mtinfo1, hinfo),
1027		.checkentry     = hashlimit_mt_check_v1,
1028		.destroy        = hashlimit_mt_destroy_v1,
1029		.me             = THIS_MODULE,
1030	},
1031	{
1032		.name           = "hashlimit",
1033		.revision       = 2,
1034		.family         = NFPROTO_IPV6,
1035		.match          = hashlimit_mt_v2,
1036		.matchsize      = sizeof(struct xt_hashlimit_mtinfo2),
1037		.usersize	= offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1038		.checkentry     = hashlimit_mt_check_v2,
1039		.destroy        = hashlimit_mt_destroy_v2,
1040		.me             = THIS_MODULE,
1041	},
1042	{
1043		.name           = "hashlimit",
1044		.revision       = 3,
1045		.family         = NFPROTO_IPV6,
1046		.match          = hashlimit_mt,
1047		.matchsize      = sizeof(struct xt_hashlimit_mtinfo3),
1048		.usersize	= offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1049		.checkentry     = hashlimit_mt_check,
1050		.destroy        = hashlimit_mt_destroy,
1051		.me             = THIS_MODULE,
1052	},
1053#endif
1054};
1055
1056/* PROC stuff */
1057static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1058	__acquires(htable->lock)
1059{
1060	struct xt_hashlimit_htable *htable = s->private;
1061	unsigned int *bucket;
1062
1063	spin_lock_bh(&htable->lock);
1064	if (*pos >= htable->cfg.size)
1065		return NULL;
1066
1067	bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1068	if (!bucket)
1069		return ERR_PTR(-ENOMEM);
1070
1071	*bucket = *pos;
1072	return bucket;
1073}
1074
1075static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1076{
1077	struct xt_hashlimit_htable *htable = s->private;
1078	unsigned int *bucket = v;
1079
1080	*pos = ++(*bucket);
1081	if (*pos >= htable->cfg.size) {
1082		kfree(v);
1083		return NULL;
1084	}
1085	return bucket;
1086}
1087
1088static void dl_seq_stop(struct seq_file *s, void *v)
1089	__releases(htable->lock)
1090{
1091	struct xt_hashlimit_htable *htable = s->private;
1092	unsigned int *bucket = v;
1093
1094	if (!IS_ERR(bucket))
1095		kfree(bucket);
1096	spin_unlock_bh(&htable->lock);
1097}
1098
1099static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1100			 struct seq_file *s)
1101{
1102	switch (family) {
1103	case NFPROTO_IPV4:
1104		seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1105			   (long)(ent->expires - jiffies)/HZ,
1106			   &ent->dst.ip.src,
1107			   ntohs(ent->dst.src_port),
1108			   &ent->dst.ip.dst,
1109			   ntohs(ent->dst.dst_port),
1110			   ent->rateinfo.credit, ent->rateinfo.credit_cap,
1111			   ent->rateinfo.cost);
1112		break;
1113#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1114	case NFPROTO_IPV6:
1115		seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1116			   (long)(ent->expires - jiffies)/HZ,
1117			   &ent->dst.ip6.src,
1118			   ntohs(ent->dst.src_port),
1119			   &ent->dst.ip6.dst,
1120			   ntohs(ent->dst.dst_port),
1121			   ent->rateinfo.credit, ent->rateinfo.credit_cap,
1122			   ent->rateinfo.cost);
1123		break;
1124#endif
1125	default:
1126		BUG();
1127	}
1128}
1129
1130static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1131			       struct seq_file *s)
1132{
1133	const struct xt_hashlimit_htable *ht = s->private;
1134
1135	spin_lock(&ent->lock);
1136	/* recalculate to show accurate numbers */
1137	rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1138
1139	dl_seq_print(ent, family, s);
1140
1141	spin_unlock(&ent->lock);
1142	return seq_has_overflowed(s);
1143}
1144
1145static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1146			       struct seq_file *s)
1147{
1148	const struct xt_hashlimit_htable *ht = s->private;
1149
1150	spin_lock(&ent->lock);
1151	/* recalculate to show accurate numbers */
1152	rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1153
1154	dl_seq_print(ent, family, s);
1155
1156	spin_unlock(&ent->lock);
1157	return seq_has_overflowed(s);
1158}
1159
1160static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1161			    struct seq_file *s)
1162{
1163	const struct xt_hashlimit_htable *ht = s->private;
1164
1165	spin_lock(&ent->lock);
1166	/* recalculate to show accurate numbers */
1167	rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1168
1169	dl_seq_print(ent, family, s);
1170
1171	spin_unlock(&ent->lock);
1172	return seq_has_overflowed(s);
1173}
1174
1175static int dl_seq_show_v2(struct seq_file *s, void *v)
1176{
1177	struct xt_hashlimit_htable *htable = s->private;
1178	unsigned int *bucket = (unsigned int *)v;
1179	struct dsthash_ent *ent;
1180
1181	if (!hlist_empty(&htable->hash[*bucket])) {
1182		hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1183			if (dl_seq_real_show_v2(ent, htable->family, s))
1184				return -1;
1185	}
1186	return 0;
1187}
1188
1189static int dl_seq_show_v1(struct seq_file *s, void *v)
1190{
1191	struct xt_hashlimit_htable *htable = s->private;
1192	unsigned int *bucket = v;
1193	struct dsthash_ent *ent;
1194
1195	if (!hlist_empty(&htable->hash[*bucket])) {
1196		hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1197			if (dl_seq_real_show_v1(ent, htable->family, s))
1198				return -1;
1199	}
1200	return 0;
1201}
1202
1203static int dl_seq_show(struct seq_file *s, void *v)
1204{
1205	struct xt_hashlimit_htable *htable = s->private;
1206	unsigned int *bucket = v;
1207	struct dsthash_ent *ent;
1208
1209	if (!hlist_empty(&htable->hash[*bucket])) {
1210		hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1211			if (dl_seq_real_show(ent, htable->family, s))
1212				return -1;
1213	}
1214	return 0;
1215}
1216
1217static const struct seq_operations dl_seq_ops_v1 = {
1218	.start = dl_seq_start,
1219	.next  = dl_seq_next,
1220	.stop  = dl_seq_stop,
1221	.show  = dl_seq_show_v1
1222};
1223
1224static const struct seq_operations dl_seq_ops_v2 = {
1225	.start = dl_seq_start,
1226	.next  = dl_seq_next,
1227	.stop  = dl_seq_stop,
1228	.show  = dl_seq_show_v2
1229};
1230
1231static const struct seq_operations dl_seq_ops = {
1232	.start = dl_seq_start,
1233	.next  = dl_seq_next,
1234	.stop  = dl_seq_stop,
1235	.show  = dl_seq_show
1236};
1237
1238static int dl_proc_open_v2(struct inode *inode, struct file *file)
1239{
1240	int ret = seq_open(file, &dl_seq_ops_v2);
1241
1242	if (!ret) {
1243		struct seq_file *sf = file->private_data;
1244
1245		sf->private = PDE_DATA(inode);
1246	}
1247	return ret;
1248}
1249
1250static int dl_proc_open_v1(struct inode *inode, struct file *file)
1251{
1252	int ret = seq_open(file, &dl_seq_ops_v1);
1253
1254	if (!ret) {
1255		struct seq_file *sf = file->private_data;
1256		sf->private = PDE_DATA(inode);
1257	}
1258	return ret;
1259}
1260
1261static int dl_proc_open(struct inode *inode, struct file *file)
1262{
1263	int ret = seq_open(file, &dl_seq_ops);
1264
1265	if (!ret) {
1266		struct seq_file *sf = file->private_data;
1267
1268		sf->private = PDE_DATA(inode);
1269	}
1270	return ret;
1271}
1272
1273static const struct file_operations dl_file_ops_v2 = {
1274	.open    = dl_proc_open_v2,
1275	.read    = seq_read,
1276	.llseek  = seq_lseek,
1277	.release = seq_release
1278};
1279
1280static const struct file_operations dl_file_ops_v1 = {
1281	.open    = dl_proc_open_v1,
1282	.read    = seq_read,
1283	.llseek  = seq_lseek,
1284	.release = seq_release
1285};
1286
1287static const struct file_operations dl_file_ops = {
1288	.open    = dl_proc_open,
1289	.read    = seq_read,
1290	.llseek  = seq_lseek,
1291	.release = seq_release
1292};
1293
1294static int __net_init hashlimit_proc_net_init(struct net *net)
1295{
1296	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1297
1298	hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1299	if (!hashlimit_net->ipt_hashlimit)
1300		return -ENOMEM;
1301#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1302	hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1303	if (!hashlimit_net->ip6t_hashlimit) {
1304		remove_proc_entry("ipt_hashlimit", net->proc_net);
1305		return -ENOMEM;
1306	}
1307#endif
1308	return 0;
1309}
1310
1311static void __net_exit hashlimit_proc_net_exit(struct net *net)
1312{
1313	struct xt_hashlimit_htable *hinfo;
1314	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1315
1316	/* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1317	 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1318	 * entries is empty before trying to remove it.
1319	 */
1320	mutex_lock(&hashlimit_mutex);
1321	hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1322		htable_remove_proc_entry(hinfo);
1323	hashlimit_net->ipt_hashlimit = NULL;
1324	hashlimit_net->ip6t_hashlimit = NULL;
1325	mutex_unlock(&hashlimit_mutex);
1326
1327	remove_proc_entry("ipt_hashlimit", net->proc_net);
1328#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1329	remove_proc_entry("ip6t_hashlimit", net->proc_net);
1330#endif
1331}
1332
1333static int __net_init hashlimit_net_init(struct net *net)
1334{
1335	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1336
1337	INIT_HLIST_HEAD(&hashlimit_net->htables);
1338	return hashlimit_proc_net_init(net);
1339}
1340
1341static void __net_exit hashlimit_net_exit(struct net *net)
1342{
1343	hashlimit_proc_net_exit(net);
1344}
1345
1346static struct pernet_operations hashlimit_net_ops = {
1347	.init	= hashlimit_net_init,
1348	.exit	= hashlimit_net_exit,
1349	.id	= &hashlimit_net_id,
1350	.size	= sizeof(struct hashlimit_net),
1351};
1352
1353static int __init hashlimit_mt_init(void)
1354{
1355	int err;
1356
1357	err = register_pernet_subsys(&hashlimit_net_ops);
1358	if (err < 0)
1359		return err;
1360	err = xt_register_matches(hashlimit_mt_reg,
1361	      ARRAY_SIZE(hashlimit_mt_reg));
1362	if (err < 0)
1363		goto err1;
1364
1365	err = -ENOMEM;
1366	hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1367					    sizeof(struct dsthash_ent), 0, 0,
1368					    NULL);
1369	if (!hashlimit_cachep) {
1370		pr_warn("unable to create slab cache\n");
1371		goto err2;
1372	}
1373	return 0;
1374
1375err2:
1376	xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1377err1:
1378	unregister_pernet_subsys(&hashlimit_net_ops);
1379	return err;
1380
1381}
1382
1383static void __exit hashlimit_mt_exit(void)
1384{
1385	xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1386	unregister_pernet_subsys(&hashlimit_net_ops);
1387
1388	rcu_barrier_bh();
1389	kmem_cache_destroy(hashlimit_cachep);
1390}
1391
1392module_init(hashlimit_mt_init);
1393module_exit(hashlimit_mt_exit);