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
  39MODULE_LICENSE("GPL");
  40MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
  41MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
  42MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
  43MODULE_ALIAS("ipt_hashlimit");
  44MODULE_ALIAS("ip6t_hashlimit");
  45
  46struct hashlimit_net {
  47	struct hlist_head	htables;
  48	struct proc_dir_entry	*ipt_hashlimit;
  49	struct proc_dir_entry	*ip6t_hashlimit;
  50};
  51
  52static unsigned int hashlimit_net_id;
  53static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
  54{
  55	return net_generic(net, hashlimit_net_id);
  56}
  57
  58/* need to declare this at the top */
  59static const struct file_operations dl_file_ops_v1;
  60static const struct file_operations dl_file_ops;
  61
  62/* hash table crap */
  63struct dsthash_dst {
  64	union {
  65		struct {
  66			__be32 src;
  67			__be32 dst;
  68		} ip;
  69#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
  70		struct {
  71			__be32 src[4];
  72			__be32 dst[4];
  73		} ip6;
  74#endif
  75	};
  76	__be16 src_port;
  77	__be16 dst_port;
  78};
  79
  80struct dsthash_ent {
  81	/* static / read-only parts in the beginning */
  82	struct hlist_node node;
  83	struct dsthash_dst dst;
  84
  85	/* modified structure members in the end */
  86	spinlock_t lock;
  87	unsigned long expires;		/* precalculated expiry time */
  88	struct {
  89		unsigned long prev;	/* last modification */
  90		u_int64_t credit;
  91		u_int64_t credit_cap, cost;
  92	} rateinfo;
  93	struct rcu_head rcu;
  94};
  95
  96struct xt_hashlimit_htable {
  97	struct hlist_node node;		/* global list of all htables */
  98	int use;
  99	u_int8_t family;
 100	bool rnd_initialized;
 101
 102	struct hashlimit_cfg2 cfg;	/* config */
 103
 104	/* used internally */
 105	spinlock_t lock;		/* lock for list_head */
 106	u_int32_t rnd;			/* random seed for hash */
 107	unsigned int count;		/* number entries in table */
 108	struct delayed_work gc_work;
 109
 110	/* seq_file stuff */
 111	struct proc_dir_entry *pde;
 112	const char *name;
 113	struct net *net;
 114
 115	struct hlist_head hash[0];	/* hashtable itself */
 116};
 117
 118static int
 119cfg_copy(struct hashlimit_cfg2 *to, void *from, int revision)
 120{
 121	if (revision == 1) {
 122		struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
 123
 124		to->mode = cfg->mode;
 125		to->avg = cfg->avg;
 126		to->burst = cfg->burst;
 127		to->size = cfg->size;
 128		to->max = cfg->max;
 129		to->gc_interval = cfg->gc_interval;
 130		to->expire = cfg->expire;
 131		to->srcmask = cfg->srcmask;
 132		to->dstmask = cfg->dstmask;
 133	} else if (revision == 2) {
 134		memcpy(to, from, sizeof(struct hashlimit_cfg2));
 135	} else {
 136		return -EINVAL;
 137	}
 138
 139	return 0;
 140}
 141
 142static DEFINE_MUTEX(hashlimit_mutex);	/* protects htables list */
 143static struct kmem_cache *hashlimit_cachep __read_mostly;
 144
 145static inline bool dst_cmp(const struct dsthash_ent *ent,
 146			   const struct dsthash_dst *b)
 147{
 148	return !memcmp(&ent->dst, b, sizeof(ent->dst));
 149}
 150
 151static u_int32_t
 152hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
 153{
 154	u_int32_t hash = jhash2((const u32 *)dst,
 155				sizeof(*dst)/sizeof(u32),
 156				ht->rnd);
 157	/*
 158	 * Instead of returning hash % ht->cfg.size (implying a divide)
 159	 * we return the high 32 bits of the (hash * ht->cfg.size) that will
 160	 * give results between [0 and cfg.size-1] and same hash distribution,
 161	 * but using a multiply, less expensive than a divide
 162	 */
 163	return reciprocal_scale(hash, ht->cfg.size);
 164}
 165
 166static struct dsthash_ent *
 167dsthash_find(const struct xt_hashlimit_htable *ht,
 168	     const struct dsthash_dst *dst)
 169{
 170	struct dsthash_ent *ent;
 
 171	u_int32_t hash = hash_dst(ht, dst);
 172
 173	if (!hlist_empty(&ht->hash[hash])) {
 174		hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
 175			if (dst_cmp(ent, dst)) {
 176				spin_lock(&ent->lock);
 177				return ent;
 178			}
 179	}
 180	return NULL;
 181}
 182
 183/* allocate dsthash_ent, initialize dst, put in htable and lock it */
 184static struct dsthash_ent *
 185dsthash_alloc_init(struct xt_hashlimit_htable *ht,
 186		   const struct dsthash_dst *dst, bool *race)
 187{
 188	struct dsthash_ent *ent;
 189
 190	spin_lock(&ht->lock);
 191
 192	/* Two or more packets may race to create the same entry in the
 193	 * hashtable, double check if this packet lost race.
 194	 */
 195	ent = dsthash_find(ht, dst);
 196	if (ent != NULL) {
 197		spin_unlock(&ht->lock);
 198		*race = true;
 199		return ent;
 200	}
 201
 202	/* initialize hash with random val at the time we allocate
 203	 * the first hashtable entry */
 204	if (unlikely(!ht->rnd_initialized)) {
 205		get_random_bytes(&ht->rnd, sizeof(ht->rnd));
 206		ht->rnd_initialized = true;
 207	}
 208
 209	if (ht->cfg.max && ht->count >= ht->cfg.max) {
 210		/* FIXME: do something. question is what.. */
 211		net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
 
 212		ent = NULL;
 213	} else
 214		ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
 215	if (ent) {
 
 
 
 216		memcpy(&ent->dst, dst, sizeof(ent->dst));
 217		spin_lock_init(&ent->lock);
 218
 219		spin_lock(&ent->lock);
 220		hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
 221		ht->count++;
 222	}
 223	spin_unlock(&ht->lock);
 224	return ent;
 225}
 226
 227static void dsthash_free_rcu(struct rcu_head *head)
 228{
 229	struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
 230
 231	kmem_cache_free(hashlimit_cachep, ent);
 232}
 233
 234static inline void
 235dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
 236{
 237	hlist_del_rcu(&ent->node);
 238	call_rcu_bh(&ent->rcu, dsthash_free_rcu);
 239	ht->count--;
 240}
 241static void htable_gc(struct work_struct *work);
 242
 243static int htable_create(struct net *net, struct hashlimit_cfg2 *cfg,
 244			 const char *name, u_int8_t family,
 245			 struct xt_hashlimit_htable **out_hinfo,
 246			 int revision)
 247{
 248	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 249	struct xt_hashlimit_htable *hinfo;
 250	unsigned int size, i;
 251	int ret;
 252
 253	if (cfg->size) {
 254		size = cfg->size;
 255	} else {
 256		size = (totalram_pages << PAGE_SHIFT) / 16384 /
 257		       sizeof(struct list_head);
 258		if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
 259			size = 8192;
 260		if (size < 16)
 261			size = 16;
 262	}
 263	/* FIXME: don't use vmalloc() here or anywhere else -HW */
 264	hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
 265	                sizeof(struct list_head) * size);
 266	if (hinfo == NULL)
 267		return -ENOMEM;
 268	*out_hinfo = hinfo;
 269
 270	/* copy match config into hashtable config */
 271	ret = cfg_copy(&hinfo->cfg, (void *)cfg, 2);
 272
 273	if (ret)
 274		return ret;
 275
 276	hinfo->cfg.size = size;
 277	if (hinfo->cfg.max == 0)
 278		hinfo->cfg.max = 8 * hinfo->cfg.size;
 279	else if (hinfo->cfg.max < hinfo->cfg.size)
 280		hinfo->cfg.max = hinfo->cfg.size;
 281
 282	for (i = 0; i < hinfo->cfg.size; i++)
 283		INIT_HLIST_HEAD(&hinfo->hash[i]);
 284
 285	hinfo->use = 1;
 286	hinfo->count = 0;
 287	hinfo->family = family;
 288	hinfo->rnd_initialized = false;
 289	hinfo->name = kstrdup(name, GFP_KERNEL);
 290	if (!hinfo->name) {
 291		vfree(hinfo);
 292		return -ENOMEM;
 293	}
 294	spin_lock_init(&hinfo->lock);
 295
 296	hinfo->pde = proc_create_data(name, 0,
 297		(family == NFPROTO_IPV4) ?
 298		hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
 299		(revision == 1) ? &dl_file_ops_v1 : &dl_file_ops,
 300		hinfo);
 301	if (hinfo->pde == NULL) {
 302		kfree(hinfo->name);
 303		vfree(hinfo);
 304		return -ENOMEM;
 305	}
 306	hinfo->net = net;
 307
 308	INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
 309	queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
 310			   msecs_to_jiffies(hinfo->cfg.gc_interval));
 311
 312	hlist_add_head(&hinfo->node, &hashlimit_net->htables);
 313
 314	return 0;
 315}
 316
 317static bool select_all(const struct xt_hashlimit_htable *ht,
 318		       const struct dsthash_ent *he)
 319{
 320	return 1;
 321}
 322
 323static bool select_gc(const struct xt_hashlimit_htable *ht,
 324		      const struct dsthash_ent *he)
 325{
 326	return time_after_eq(jiffies, he->expires);
 327}
 328
 329static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
 330			bool (*select)(const struct xt_hashlimit_htable *ht,
 331				      const struct dsthash_ent *he))
 332{
 333	unsigned int i;
 334
 
 
 335	for (i = 0; i < ht->cfg.size; i++) {
 336		struct dsthash_ent *dh;
 337		struct hlist_node *n;
 338
 339		spin_lock_bh(&ht->lock);
 340		hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
 341			if ((*select)(ht, dh))
 342				dsthash_free(ht, dh);
 343		}
 344		spin_unlock_bh(&ht->lock);
 345		cond_resched();
 346	}
 
 347}
 348
 349static void htable_gc(struct work_struct *work)
 
 350{
 351	struct xt_hashlimit_htable *ht;
 352
 353	ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
 354
 355	htable_selective_cleanup(ht, select_gc);
 356
 357	queue_delayed_work(system_power_efficient_wq,
 358			   &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
 
 359}
 360
 361static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
 362{
 363	struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
 364	struct proc_dir_entry *parent;
 365
 
 
 366	if (hinfo->family == NFPROTO_IPV4)
 367		parent = hashlimit_net->ipt_hashlimit;
 368	else
 369		parent = hashlimit_net->ip6t_hashlimit;
 370
 371	if (parent != NULL)
 372		remove_proc_entry(hinfo->name, parent);
 373}
 374
 375static void htable_destroy(struct xt_hashlimit_htable *hinfo)
 376{
 377	cancel_delayed_work_sync(&hinfo->gc_work);
 378	htable_remove_proc_entry(hinfo);
 379	htable_selective_cleanup(hinfo, select_all);
 380	kfree(hinfo->name);
 381	vfree(hinfo);
 382}
 383
 384static struct xt_hashlimit_htable *htable_find_get(struct net *net,
 385						   const char *name,
 386						   u_int8_t family)
 387{
 388	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
 389	struct xt_hashlimit_htable *hinfo;
 
 390
 391	hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
 392		if (!strcmp(name, hinfo->name) &&
 393		    hinfo->family == family) {
 394			hinfo->use++;
 395			return hinfo;
 396		}
 397	}
 398	return NULL;
 399}
 400
 401static void htable_put(struct xt_hashlimit_htable *hinfo)
 402{
 403	mutex_lock(&hashlimit_mutex);
 404	if (--hinfo->use == 0) {
 405		hlist_del(&hinfo->node);
 406		htable_destroy(hinfo);
 407	}
 408	mutex_unlock(&hashlimit_mutex);
 409}
 410
 411/* The algorithm used is the Simple Token Bucket Filter (TBF)
 412 * see net/sched/sch_tbf.c in the linux source tree
 413 */
 414
 415/* Rusty: This is my (non-mathematically-inclined) understanding of
 416   this algorithm.  The `average rate' in jiffies becomes your initial
 417   amount of credit `credit' and the most credit you can ever have
 418   `credit_cap'.  The `peak rate' becomes the cost of passing the
 419   test, `cost'.
 420
 421   `prev' tracks the last packet hit: you gain one credit per jiffy.
 422   If you get credit balance more than this, the extra credit is
 423   discarded.  Every time the match passes, you lose `cost' credits;
 424   if you don't have that many, the test fails.
 425
 426   See Alexey's formal explanation in net/sched/sch_tbf.c.
 427
 428   To get the maximum range, we multiply by this factor (ie. you get N
 429   credits per jiffy).  We want to allow a rate as low as 1 per day
 430   (slowest userspace tool allows), which means
 431   CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
 432*/
 433#define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
 434#define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
 435
 436/* Repeated shift and or gives us all 1s, final shift and add 1 gives
 437 * us the power of 2 below the theoretical max, so GCC simply does a
 438 * shift. */
 439#define _POW2_BELOW2(x) ((x)|((x)>>1))
 440#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
 441#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
 442#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
 443#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
 444#define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
 445#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
 446#define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
 447
 448#define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
 449#define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
 450
 451/* in byte mode, the lowest possible rate is one packet/second.
 452 * credit_cap is used as a counter that tells us how many times we can
 453 * refill the "credits available" counter when it becomes empty.
 454 */
 455#define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
 456#define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
 457
 458static u32 xt_hashlimit_len_to_chunks(u32 len)
 459{
 460	return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
 461}
 462
 463/* Precision saver. */
 464static u64 user2credits(u64 user, int revision)
 
 465{
 466	if (revision == 1) {
 467		/* If multiplying would overflow... */
 468		if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY_v1))
 469			/* Divide first. */
 470			return div64_u64(user, XT_HASHLIMIT_SCALE)
 471				* HZ * CREDITS_PER_JIFFY_v1;
 472
 473		return div64_u64(user * HZ * CREDITS_PER_JIFFY_v1,
 474				 XT_HASHLIMIT_SCALE);
 475	} else {
 476		if (user > 0xFFFFFFFFFFFFFFFFULL / (HZ*CREDITS_PER_JIFFY))
 477			return div64_u64(user, XT_HASHLIMIT_SCALE_v2)
 478				* HZ * CREDITS_PER_JIFFY;
 479
 480		return div64_u64(user * HZ * CREDITS_PER_JIFFY,
 481				 XT_HASHLIMIT_SCALE_v2);
 482	}
 483}
 484
 485static u32 user2credits_byte(u32 user)
 486{
 487	u64 us = user;
 488	us *= HZ * CREDITS_PER_JIFFY_BYTES;
 489	return (u32) (us >> 32);
 490}
 491
 492static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
 493			    u32 mode, int revision)
 494{
 495	unsigned long delta = now - dh->rateinfo.prev;
 496	u64 cap, cpj;
 497
 498	if (delta == 0)
 499		return;
 500
 501	dh->rateinfo.prev = now;
 502
 503	if (mode & XT_HASHLIMIT_BYTES) {
 504		u64 tmp = dh->rateinfo.credit;
 505		dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
 506		cap = CREDITS_PER_JIFFY_BYTES * HZ;
 507		if (tmp >= dh->rateinfo.credit) {/* overflow */
 508			dh->rateinfo.credit = cap;
 509			return;
 510		}
 511	} else {
 512		cpj = (revision == 1) ?
 513			CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
 514		dh->rateinfo.credit += delta * cpj;
 515		cap = dh->rateinfo.credit_cap;
 516	}
 517	if (dh->rateinfo.credit > cap)
 518		dh->rateinfo.credit = cap;
 519}
 520
 521static void rateinfo_init(struct dsthash_ent *dh,
 522			  struct xt_hashlimit_htable *hinfo, int revision)
 523{
 524	dh->rateinfo.prev = jiffies;
 525	if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
 526		dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
 527		dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
 528		dh->rateinfo.credit_cap = hinfo->cfg.burst;
 529	} else {
 530		dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
 531						   hinfo->cfg.burst, revision);
 532		dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
 533		dh->rateinfo.credit_cap = dh->rateinfo.credit;
 534	}
 535}
 536
 537static inline __be32 maskl(__be32 a, unsigned int l)
 538{
 539	return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
 540}
 541
 542#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 543static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
 544{
 545	switch (p) {
 546	case 0 ... 31:
 547		i[0] = maskl(i[0], p);
 548		i[1] = i[2] = i[3] = 0;
 549		break;
 550	case 32 ... 63:
 551		i[1] = maskl(i[1], p - 32);
 552		i[2] = i[3] = 0;
 553		break;
 554	case 64 ... 95:
 555		i[2] = maskl(i[2], p - 64);
 556		i[3] = 0;
 557		break;
 558	case 96 ... 127:
 559		i[3] = maskl(i[3], p - 96);
 560		break;
 561	case 128:
 562		break;
 563	}
 564}
 565#endif
 566
 567static int
 568hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
 569		   struct dsthash_dst *dst,
 570		   const struct sk_buff *skb, unsigned int protoff)
 571{
 572	__be16 _ports[2], *ports;
 573	u8 nexthdr;
 574	int poff;
 575
 576	memset(dst, 0, sizeof(*dst));
 577
 578	switch (hinfo->family) {
 579	case NFPROTO_IPV4:
 580		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
 581			dst->ip.dst = maskl(ip_hdr(skb)->daddr,
 582			              hinfo->cfg.dstmask);
 583		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
 584			dst->ip.src = maskl(ip_hdr(skb)->saddr,
 585			              hinfo->cfg.srcmask);
 586
 587		if (!(hinfo->cfg.mode &
 588		      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 589			return 0;
 590		nexthdr = ip_hdr(skb)->protocol;
 591		break;
 592#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 593	case NFPROTO_IPV6:
 594	{
 595		__be16 frag_off;
 596
 597		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
 598			memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
 599			       sizeof(dst->ip6.dst));
 600			hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
 601		}
 602		if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
 603			memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
 604			       sizeof(dst->ip6.src));
 605			hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
 606		}
 607
 608		if (!(hinfo->cfg.mode &
 609		      (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
 610			return 0;
 611		nexthdr = ipv6_hdr(skb)->nexthdr;
 612		protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
 613		if ((int)protoff < 0)
 614			return -1;
 615		break;
 616	}
 617#endif
 618	default:
 619		BUG();
 620		return 0;
 621	}
 622
 623	poff = proto_ports_offset(nexthdr);
 624	if (poff >= 0) {
 625		ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
 626					   &_ports);
 627	} else {
 628		_ports[0] = _ports[1] = 0;
 629		ports = _ports;
 630	}
 631	if (!ports)
 632		return -1;
 633	if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
 634		dst->src_port = ports[0];
 635	if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
 636		dst->dst_port = ports[1];
 637	return 0;
 638}
 639
 640static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
 641{
 642	u64 tmp = xt_hashlimit_len_to_chunks(len);
 643	tmp = tmp * dh->rateinfo.cost;
 644
 645	if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
 646		tmp = CREDITS_PER_JIFFY_BYTES * HZ;
 647
 648	if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
 649		dh->rateinfo.credit_cap--;
 650		dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
 651	}
 652	return (u32) tmp;
 653}
 654
 655static bool
 656hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
 657		    struct xt_hashlimit_htable *hinfo,
 658		    const struct hashlimit_cfg2 *cfg, int revision)
 659{
 
 
 660	unsigned long now = jiffies;
 661	struct dsthash_ent *dh;
 662	struct dsthash_dst dst;
 663	bool race = false;
 664	u64 cost;
 665
 666	if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
 667		goto hotdrop;
 668
 669	rcu_read_lock_bh();
 670	dh = dsthash_find(hinfo, &dst);
 671	if (dh == NULL) {
 672		dh = dsthash_alloc_init(hinfo, &dst, &race);
 673		if (dh == NULL) {
 674			rcu_read_unlock_bh();
 675			goto hotdrop;
 676		} else if (race) {
 677			/* Already got an entry, update expiration timeout */
 678			dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 679			rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
 680		} else {
 681			dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
 682			rateinfo_init(dh, hinfo, revision);
 683		}
 
 
 
 
 
 
 
 684	} else {
 685		/* update expiration timeout */
 686		dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
 687		rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
 688	}
 689
 690	if (cfg->mode & XT_HASHLIMIT_BYTES)
 691		cost = hashlimit_byte_cost(skb->len, dh);
 692	else
 693		cost = dh->rateinfo.cost;
 694
 695	if (dh->rateinfo.credit >= cost) {
 696		/* below the limit */
 697		dh->rateinfo.credit -= cost;
 698		spin_unlock(&dh->lock);
 699		rcu_read_unlock_bh();
 700		return !(cfg->mode & XT_HASHLIMIT_INVERT);
 701	}
 702
 703	spin_unlock(&dh->lock);
 704	rcu_read_unlock_bh();
 705	/* default match is underlimit - so over the limit, we need to invert */
 706	return cfg->mode & XT_HASHLIMIT_INVERT;
 707
 708 hotdrop:
 709	par->hotdrop = true;
 710	return false;
 711}
 712
 713static bool
 714hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
 715{
 716	const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 717	struct xt_hashlimit_htable *hinfo = info->hinfo;
 718	struct hashlimit_cfg2 cfg = {};
 719	int ret;
 720
 721	ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
 722
 723	if (ret)
 724		return ret;
 725
 726	return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
 727}
 728
 729static bool
 730hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
 731{
 732	const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
 733	struct xt_hashlimit_htable *hinfo = info->hinfo;
 734
 735	return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 2);
 736}
 737
 738static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
 739				     struct xt_hashlimit_htable **hinfo,
 740				     struct hashlimit_cfg2 *cfg,
 741				     const char *name, int revision)
 742{
 743	struct net *net = par->net;
 
 744	int ret;
 745
 746	if (cfg->gc_interval == 0 || cfg->expire == 0)
 
 
 
 
 
 
 
 
 
 
 747		return -EINVAL;
 748	if (par->family == NFPROTO_IPV4) {
 749		if (cfg->srcmask > 32 || cfg->dstmask > 32)
 750			return -EINVAL;
 751	} else {
 752		if (cfg->srcmask > 128 || cfg->dstmask > 128)
 753			return -EINVAL;
 754	}
 755
 756	if (cfg->mode & ~XT_HASHLIMIT_ALL) {
 757		pr_info("Unknown mode mask %X, kernel too old?\n",
 758						cfg->mode);
 759		return -EINVAL;
 760	}
 761
 762	/* Check for overflow. */
 763	if (cfg->mode & XT_HASHLIMIT_BYTES) {
 764		if (user2credits_byte(cfg->avg) == 0) {
 765			pr_info("overflow, rate too high: %llu\n", cfg->avg);
 766			return -EINVAL;
 767		}
 768	} else if (cfg->burst == 0 ||
 769		    user2credits(cfg->avg * cfg->burst, revision) <
 770		    user2credits(cfg->avg, revision)) {
 771			pr_info("overflow, try lower: %llu/%llu\n",
 772				cfg->avg, cfg->burst);
 773			return -ERANGE;
 774	}
 775
 776	mutex_lock(&hashlimit_mutex);
 777	*hinfo = htable_find_get(net, name, par->family);
 778	if (*hinfo == NULL) {
 779		ret = htable_create(net, cfg, name, par->family,
 780				    hinfo, revision);
 781		if (ret < 0) {
 782			mutex_unlock(&hashlimit_mutex);
 783			return ret;
 784		}
 785	}
 786	mutex_unlock(&hashlimit_mutex);
 787
 788	return 0;
 789}
 790
 791static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
 792{
 793	struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 794	struct hashlimit_cfg2 cfg = {};
 795	int ret;
 796
 797	if (info->name[sizeof(info->name) - 1] != '\0')
 798		return -EINVAL;
 799
 800	ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
 801
 802	if (ret)
 803		return ret;
 804
 805	return hashlimit_mt_check_common(par, &info->hinfo,
 806					 &cfg, info->name, 1);
 807}
 808
 809static int hashlimit_mt_check(const struct xt_mtchk_param *par)
 810{
 811	struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
 812
 813	if (info->name[sizeof(info->name) - 1] != '\0')
 814		return -EINVAL;
 815
 816	return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
 817					 info->name, 2);
 818}
 819
 820static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
 821{
 822	const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
 823
 824	htable_put(info->hinfo);
 825}
 826
 827static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
 828{
 829	const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
 830
 831	htable_put(info->hinfo);
 832}
 833
 834static struct xt_match hashlimit_mt_reg[] __read_mostly = {
 835	{
 836		.name           = "hashlimit",
 837		.revision       = 1,
 838		.family         = NFPROTO_IPV4,
 839		.match          = hashlimit_mt_v1,
 840		.matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 841		.checkentry     = hashlimit_mt_check_v1,
 842		.destroy        = hashlimit_mt_destroy_v1,
 843		.me             = THIS_MODULE,
 844	},
 845	{
 846		.name           = "hashlimit",
 847		.revision       = 2,
 848		.family         = NFPROTO_IPV4,
 849		.match          = hashlimit_mt,
 850		.matchsize      = sizeof(struct xt_hashlimit_mtinfo2),
 851		.checkentry     = hashlimit_mt_check,
 852		.destroy        = hashlimit_mt_destroy,
 853		.me             = THIS_MODULE,
 854	},
 855#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 856	{
 857		.name           = "hashlimit",
 858		.revision       = 1,
 859		.family         = NFPROTO_IPV6,
 860		.match          = hashlimit_mt_v1,
 861		.matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
 862		.checkentry     = hashlimit_mt_check_v1,
 863		.destroy        = hashlimit_mt_destroy_v1,
 864		.me             = THIS_MODULE,
 865	},
 866	{
 867		.name           = "hashlimit",
 868		.revision       = 2,
 869		.family         = NFPROTO_IPV6,
 870		.match          = hashlimit_mt,
 871		.matchsize      = sizeof(struct xt_hashlimit_mtinfo2),
 872		.checkentry     = hashlimit_mt_check,
 873		.destroy        = hashlimit_mt_destroy,
 874		.me             = THIS_MODULE,
 875	},
 876#endif
 877};
 878
 879/* PROC stuff */
 880static void *dl_seq_start(struct seq_file *s, loff_t *pos)
 881	__acquires(htable->lock)
 882{
 883	struct xt_hashlimit_htable *htable = s->private;
 884	unsigned int *bucket;
 885
 886	spin_lock_bh(&htable->lock);
 887	if (*pos >= htable->cfg.size)
 888		return NULL;
 889
 890	bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
 891	if (!bucket)
 892		return ERR_PTR(-ENOMEM);
 893
 894	*bucket = *pos;
 895	return bucket;
 896}
 897
 898static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
 899{
 900	struct xt_hashlimit_htable *htable = s->private;
 901	unsigned int *bucket = (unsigned int *)v;
 902
 903	*pos = ++(*bucket);
 904	if (*pos >= htable->cfg.size) {
 905		kfree(v);
 906		return NULL;
 907	}
 908	return bucket;
 909}
 910
 911static void dl_seq_stop(struct seq_file *s, void *v)
 912	__releases(htable->lock)
 913{
 914	struct xt_hashlimit_htable *htable = s->private;
 915	unsigned int *bucket = (unsigned int *)v;
 916
 917	if (!IS_ERR(bucket))
 918		kfree(bucket);
 919	spin_unlock_bh(&htable->lock);
 920}
 921
 922static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
 923			 struct seq_file *s)
 924{
 
 
 
 
 
 
 925	switch (family) {
 926	case NFPROTO_IPV4:
 927		seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
 928			   (long)(ent->expires - jiffies)/HZ,
 929			   &ent->dst.ip.src,
 930			   ntohs(ent->dst.src_port),
 931			   &ent->dst.ip.dst,
 932			   ntohs(ent->dst.dst_port),
 933			   ent->rateinfo.credit, ent->rateinfo.credit_cap,
 934			   ent->rateinfo.cost);
 935		break;
 936#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 937	case NFPROTO_IPV6:
 938		seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
 939			   (long)(ent->expires - jiffies)/HZ,
 940			   &ent->dst.ip6.src,
 941			   ntohs(ent->dst.src_port),
 942			   &ent->dst.ip6.dst,
 943			   ntohs(ent->dst.dst_port),
 944			   ent->rateinfo.credit, ent->rateinfo.credit_cap,
 945			   ent->rateinfo.cost);
 946		break;
 947#endif
 948	default:
 949		BUG();
 
 950	}
 951}
 952
 953static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
 954			       struct seq_file *s)
 955{
 956	const struct xt_hashlimit_htable *ht = s->private;
 957
 958	spin_lock(&ent->lock);
 959	/* recalculate to show accurate numbers */
 960	rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
 961
 962	dl_seq_print(ent, family, s);
 963
 964	spin_unlock(&ent->lock);
 965	return seq_has_overflowed(s);
 966}
 967
 968static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
 969			    struct seq_file *s)
 970{
 971	const struct xt_hashlimit_htable *ht = s->private;
 972
 973	spin_lock(&ent->lock);
 974	/* recalculate to show accurate numbers */
 975	rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
 976
 977	dl_seq_print(ent, family, s);
 978
 979	spin_unlock(&ent->lock);
 980	return seq_has_overflowed(s);
 981}
 982
 983static int dl_seq_show_v1(struct seq_file *s, void *v)
 984{
 985	struct xt_hashlimit_htable *htable = s->private;
 986	unsigned int *bucket = (unsigned int *)v;
 987	struct dsthash_ent *ent;
 988
 989	if (!hlist_empty(&htable->hash[*bucket])) {
 990		hlist_for_each_entry(ent, &htable->hash[*bucket], node)
 991			if (dl_seq_real_show_v1(ent, htable->family, s))
 992				return -1;
 993	}
 994	return 0;
 995}
 996
 997static int dl_seq_show(struct seq_file *s, void *v)
 998{
 999	struct xt_hashlimit_htable *htable = s->private;
1000	unsigned int *bucket = (unsigned int *)v;
1001	struct dsthash_ent *ent;
 
1002
1003	if (!hlist_empty(&htable->hash[*bucket])) {
1004		hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1005			if (dl_seq_real_show(ent, htable->family, s))
1006				return -1;
1007	}
1008	return 0;
1009}
1010
1011static const struct seq_operations dl_seq_ops_v1 = {
1012	.start = dl_seq_start,
1013	.next  = dl_seq_next,
1014	.stop  = dl_seq_stop,
1015	.show  = dl_seq_show_v1
1016};
1017
1018static const struct seq_operations dl_seq_ops = {
1019	.start = dl_seq_start,
1020	.next  = dl_seq_next,
1021	.stop  = dl_seq_stop,
1022	.show  = dl_seq_show
1023};
1024
1025static int dl_proc_open_v1(struct inode *inode, struct file *file)
1026{
1027	int ret = seq_open(file, &dl_seq_ops_v1);
1028
1029	if (!ret) {
1030		struct seq_file *sf = file->private_data;
1031		sf->private = PDE_DATA(inode);
1032	}
1033	return ret;
1034}
1035
1036static int dl_proc_open(struct inode *inode, struct file *file)
1037{
1038	int ret = seq_open(file, &dl_seq_ops);
1039
1040	if (!ret) {
1041		struct seq_file *sf = file->private_data;
1042
1043		sf->private = PDE_DATA(inode);
1044	}
1045	return ret;
1046}
1047
1048static const struct file_operations dl_file_ops_v1 = {
1049	.owner   = THIS_MODULE,
1050	.open    = dl_proc_open_v1,
1051	.read    = seq_read,
1052	.llseek  = seq_lseek,
1053	.release = seq_release
1054};
1055
1056static const struct file_operations dl_file_ops = {
1057	.owner   = THIS_MODULE,
1058	.open    = dl_proc_open,
1059	.read    = seq_read,
1060	.llseek  = seq_lseek,
1061	.release = seq_release
1062};
1063
1064static int __net_init hashlimit_proc_net_init(struct net *net)
1065{
1066	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1067
1068	hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1069	if (!hashlimit_net->ipt_hashlimit)
1070		return -ENOMEM;
1071#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1072	hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1073	if (!hashlimit_net->ip6t_hashlimit) {
1074		remove_proc_entry("ipt_hashlimit", net->proc_net);
1075		return -ENOMEM;
1076	}
1077#endif
1078	return 0;
1079}
1080
1081static void __net_exit hashlimit_proc_net_exit(struct net *net)
1082{
1083	struct xt_hashlimit_htable *hinfo;
1084	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1085
1086	/* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1087	 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1088	 * entries is empty before trying to remove it.
1089	 */
1090	mutex_lock(&hashlimit_mutex);
1091	hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1092		htable_remove_proc_entry(hinfo);
1093	hashlimit_net->ipt_hashlimit = NULL;
1094	hashlimit_net->ip6t_hashlimit = NULL;
1095	mutex_unlock(&hashlimit_mutex);
1096
1097	remove_proc_entry("ipt_hashlimit", net->proc_net);
1098#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1099	remove_proc_entry("ip6t_hashlimit", net->proc_net);
1100#endif
1101}
1102
1103static int __net_init hashlimit_net_init(struct net *net)
1104{
1105	struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1106
1107	INIT_HLIST_HEAD(&hashlimit_net->htables);
1108	return hashlimit_proc_net_init(net);
1109}
1110
1111static void __net_exit hashlimit_net_exit(struct net *net)
1112{
 
 
 
1113	hashlimit_proc_net_exit(net);
1114}
1115
1116static struct pernet_operations hashlimit_net_ops = {
1117	.init	= hashlimit_net_init,
1118	.exit	= hashlimit_net_exit,
1119	.id	= &hashlimit_net_id,
1120	.size	= sizeof(struct hashlimit_net),
1121};
1122
1123static int __init hashlimit_mt_init(void)
1124{
1125	int err;
1126
1127	err = register_pernet_subsys(&hashlimit_net_ops);
1128	if (err < 0)
1129		return err;
1130	err = xt_register_matches(hashlimit_mt_reg,
1131	      ARRAY_SIZE(hashlimit_mt_reg));
1132	if (err < 0)
1133		goto err1;
1134
1135	err = -ENOMEM;
1136	hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1137					    sizeof(struct dsthash_ent), 0, 0,
1138					    NULL);
1139	if (!hashlimit_cachep) {
1140		pr_warn("unable to create slab cache\n");
1141		goto err2;
1142	}
1143	return 0;
1144
1145err2:
1146	xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1147err1:
1148	unregister_pernet_subsys(&hashlimit_net_ops);
1149	return err;
1150
1151}
1152
1153static void __exit hashlimit_mt_exit(void)
1154{
1155	xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1156	unregister_pernet_subsys(&hashlimit_net_ops);
1157
1158	rcu_barrier_bh();
1159	kmem_cache_destroy(hashlimit_cachep);
1160}
1161
1162module_init(hashlimit_mt_init);
1163module_exit(hashlimit_mt_exit);