ARM: dma-api: fix max_pfn off-by-one error in __dma_supported()
[linux/fpc-iii.git] / net / netfilter / xt_hashlimit.c
blobbccd47cd7190810af06dea672842be7f58f72a4e
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * xt_hashlimit - Netfilter module to limit the number of packets per time
4 * separately for each hashbucket (sourceip/sourceport/dstip/dstport)
6 * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
7 * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
8 * Copyright © CC Computer Consultants GmbH, 2007 - 2008
10 * Development of this code was funded by Astaro AG, http://www.astaro.com/
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/module.h>
14 #include <linux/spinlock.h>
15 #include <linux/random.h>
16 #include <linux/jhash.h>
17 #include <linux/slab.h>
18 #include <linux/vmalloc.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/list.h>
22 #include <linux/skbuff.h>
23 #include <linux/mm.h>
24 #include <linux/in.h>
25 #include <linux/ip.h>
26 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
27 #include <linux/ipv6.h>
28 #include <net/ipv6.h>
29 #endif
31 #include <net/net_namespace.h>
32 #include <net/netns/generic.h>
34 #include <linux/netfilter/x_tables.h>
35 #include <linux/netfilter_ipv4/ip_tables.h>
36 #include <linux/netfilter_ipv6/ip6_tables.h>
37 #include <linux/mutex.h>
38 #include <linux/kernel.h>
39 #include <uapi/linux/netfilter/xt_hashlimit.h>
41 #define XT_HASHLIMIT_ALL (XT_HASHLIMIT_HASH_DIP | XT_HASHLIMIT_HASH_DPT | \
42 XT_HASHLIMIT_HASH_SIP | XT_HASHLIMIT_HASH_SPT | \
43 XT_HASHLIMIT_INVERT | XT_HASHLIMIT_BYTES |\
44 XT_HASHLIMIT_RATE_MATCH)
46 MODULE_LICENSE("GPL");
47 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
48 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
49 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
50 MODULE_ALIAS("ipt_hashlimit");
51 MODULE_ALIAS("ip6t_hashlimit");
53 struct hashlimit_net {
54 struct hlist_head htables;
55 struct proc_dir_entry *ipt_hashlimit;
56 struct proc_dir_entry *ip6t_hashlimit;
59 static unsigned int hashlimit_net_id;
60 static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
62 return net_generic(net, hashlimit_net_id);
65 /* need to declare this at the top */
66 static const struct seq_operations dl_seq_ops_v2;
67 static const struct seq_operations dl_seq_ops_v1;
68 static const struct seq_operations dl_seq_ops;
70 /* hash table crap */
71 struct dsthash_dst {
72 union {
73 struct {
74 __be32 src;
75 __be32 dst;
76 } ip;
77 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
78 struct {
79 __be32 src[4];
80 __be32 dst[4];
81 } ip6;
82 #endif
84 __be16 src_port;
85 __be16 dst_port;
88 struct dsthash_ent {
89 /* static / read-only parts in the beginning */
90 struct hlist_node node;
91 struct dsthash_dst dst;
93 /* modified structure members in the end */
94 spinlock_t lock;
95 unsigned long expires; /* precalculated expiry time */
96 struct {
97 unsigned long prev; /* last modification */
98 union {
99 struct {
100 u_int64_t credit;
101 u_int64_t credit_cap;
102 u_int64_t cost;
104 struct {
105 u_int32_t interval, prev_window;
106 u_int64_t current_rate;
107 u_int64_t rate;
108 int64_t burst;
111 } rateinfo;
112 struct rcu_head rcu;
115 struct xt_hashlimit_htable {
116 struct hlist_node node; /* global list of all htables */
117 int use;
118 u_int8_t family;
119 bool rnd_initialized;
121 struct hashlimit_cfg3 cfg; /* config */
123 /* used internally */
124 spinlock_t lock; /* lock for list_head */
125 u_int32_t rnd; /* random seed for hash */
126 unsigned int count; /* number entries in table */
127 struct delayed_work gc_work;
129 /* seq_file stuff */
130 struct proc_dir_entry *pde;
131 const char *name;
132 struct net *net;
134 struct hlist_head hash[0]; /* hashtable itself */
137 static int
138 cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
140 if (revision == 1) {
141 struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
143 to->mode = cfg->mode;
144 to->avg = cfg->avg;
145 to->burst = cfg->burst;
146 to->size = cfg->size;
147 to->max = cfg->max;
148 to->gc_interval = cfg->gc_interval;
149 to->expire = cfg->expire;
150 to->srcmask = cfg->srcmask;
151 to->dstmask = cfg->dstmask;
152 } else if (revision == 2) {
153 struct hashlimit_cfg2 *cfg = (struct hashlimit_cfg2 *)from;
155 to->mode = cfg->mode;
156 to->avg = cfg->avg;
157 to->burst = cfg->burst;
158 to->size = cfg->size;
159 to->max = cfg->max;
160 to->gc_interval = cfg->gc_interval;
161 to->expire = cfg->expire;
162 to->srcmask = cfg->srcmask;
163 to->dstmask = cfg->dstmask;
164 } else if (revision == 3) {
165 memcpy(to, from, sizeof(struct hashlimit_cfg3));
166 } else {
167 return -EINVAL;
170 return 0;
173 static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */
174 static struct kmem_cache *hashlimit_cachep __read_mostly;
176 static inline bool dst_cmp(const struct dsthash_ent *ent,
177 const struct dsthash_dst *b)
179 return !memcmp(&ent->dst, b, sizeof(ent->dst));
182 static u_int32_t
183 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
185 u_int32_t hash = jhash2((const u32 *)dst,
186 sizeof(*dst)/sizeof(u32),
187 ht->rnd);
189 * Instead of returning hash % ht->cfg.size (implying a divide)
190 * we return the high 32 bits of the (hash * ht->cfg.size) that will
191 * give results between [0 and cfg.size-1] and same hash distribution,
192 * but using a multiply, less expensive than a divide
194 return reciprocal_scale(hash, ht->cfg.size);
197 static struct dsthash_ent *
198 dsthash_find(const struct xt_hashlimit_htable *ht,
199 const struct dsthash_dst *dst)
201 struct dsthash_ent *ent;
202 u_int32_t hash = hash_dst(ht, dst);
204 if (!hlist_empty(&ht->hash[hash])) {
205 hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
206 if (dst_cmp(ent, dst)) {
207 spin_lock(&ent->lock);
208 return ent;
211 return NULL;
214 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
215 static struct dsthash_ent *
216 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
217 const struct dsthash_dst *dst, bool *race)
219 struct dsthash_ent *ent;
221 spin_lock(&ht->lock);
223 /* Two or more packets may race to create the same entry in the
224 * hashtable, double check if this packet lost race.
226 ent = dsthash_find(ht, dst);
227 if (ent != NULL) {
228 spin_unlock(&ht->lock);
229 *race = true;
230 return ent;
233 /* initialize hash with random val at the time we allocate
234 * the first hashtable entry */
235 if (unlikely(!ht->rnd_initialized)) {
236 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
237 ht->rnd_initialized = true;
240 if (ht->cfg.max && ht->count >= ht->cfg.max) {
241 /* FIXME: do something. question is what.. */
242 net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
243 ent = NULL;
244 } else
245 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
246 if (ent) {
247 memcpy(&ent->dst, dst, sizeof(ent->dst));
248 spin_lock_init(&ent->lock);
250 spin_lock(&ent->lock);
251 hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
252 ht->count++;
254 spin_unlock(&ht->lock);
255 return ent;
258 static void dsthash_free_rcu(struct rcu_head *head)
260 struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
262 kmem_cache_free(hashlimit_cachep, ent);
265 static inline void
266 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
268 hlist_del_rcu(&ent->node);
269 call_rcu(&ent->rcu, dsthash_free_rcu);
270 ht->count--;
272 static void htable_gc(struct work_struct *work);
274 static int htable_create(struct net *net, struct hashlimit_cfg3 *cfg,
275 const char *name, u_int8_t family,
276 struct xt_hashlimit_htable **out_hinfo,
277 int revision)
279 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
280 struct xt_hashlimit_htable *hinfo;
281 const struct seq_operations *ops;
282 unsigned int size, i;
283 unsigned long nr_pages = totalram_pages();
284 int ret;
286 if (cfg->size) {
287 size = cfg->size;
288 } else {
289 size = (nr_pages << PAGE_SHIFT) / 16384 /
290 sizeof(struct hlist_head);
291 if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
292 size = 8192;
293 if (size < 16)
294 size = 16;
296 /* FIXME: don't use vmalloc() here or anywhere else -HW */
297 hinfo = vmalloc(struct_size(hinfo, hash, size));
298 if (hinfo == NULL)
299 return -ENOMEM;
300 *out_hinfo = hinfo;
302 /* copy match config into hashtable config */
303 ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
304 if (ret) {
305 vfree(hinfo);
306 return ret;
309 hinfo->cfg.size = size;
310 if (hinfo->cfg.max == 0)
311 hinfo->cfg.max = 8 * hinfo->cfg.size;
312 else if (hinfo->cfg.max < hinfo->cfg.size)
313 hinfo->cfg.max = hinfo->cfg.size;
315 for (i = 0; i < hinfo->cfg.size; i++)
316 INIT_HLIST_HEAD(&hinfo->hash[i]);
318 hinfo->use = 1;
319 hinfo->count = 0;
320 hinfo->family = family;
321 hinfo->rnd_initialized = false;
322 hinfo->name = kstrdup(name, GFP_KERNEL);
323 if (!hinfo->name) {
324 vfree(hinfo);
325 return -ENOMEM;
327 spin_lock_init(&hinfo->lock);
329 switch (revision) {
330 case 1:
331 ops = &dl_seq_ops_v1;
332 break;
333 case 2:
334 ops = &dl_seq_ops_v2;
335 break;
336 default:
337 ops = &dl_seq_ops;
340 hinfo->pde = proc_create_seq_data(name, 0,
341 (family == NFPROTO_IPV4) ?
342 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
343 ops, hinfo);
344 if (hinfo->pde == NULL) {
345 kfree(hinfo->name);
346 vfree(hinfo);
347 return -ENOMEM;
349 hinfo->net = net;
351 INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
352 queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
353 msecs_to_jiffies(hinfo->cfg.gc_interval));
355 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
357 return 0;
360 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, bool select_all)
362 unsigned int i;
364 for (i = 0; i < ht->cfg.size; i++) {
365 struct dsthash_ent *dh;
366 struct hlist_node *n;
368 spin_lock_bh(&ht->lock);
369 hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
370 if (time_after_eq(jiffies, dh->expires) || select_all)
371 dsthash_free(ht, dh);
373 spin_unlock_bh(&ht->lock);
374 cond_resched();
378 static void htable_gc(struct work_struct *work)
380 struct xt_hashlimit_htable *ht;
382 ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
384 htable_selective_cleanup(ht, false);
386 queue_delayed_work(system_power_efficient_wq,
387 &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
390 static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
392 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
393 struct proc_dir_entry *parent;
395 if (hinfo->family == NFPROTO_IPV4)
396 parent = hashlimit_net->ipt_hashlimit;
397 else
398 parent = hashlimit_net->ip6t_hashlimit;
400 if (parent != NULL)
401 remove_proc_entry(hinfo->name, parent);
404 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
406 cancel_delayed_work_sync(&hinfo->gc_work);
407 htable_remove_proc_entry(hinfo);
408 htable_selective_cleanup(hinfo, true);
409 kfree(hinfo->name);
410 vfree(hinfo);
413 static struct xt_hashlimit_htable *htable_find_get(struct net *net,
414 const char *name,
415 u_int8_t family)
417 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
418 struct xt_hashlimit_htable *hinfo;
420 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
421 if (!strcmp(name, hinfo->name) &&
422 hinfo->family == family) {
423 hinfo->use++;
424 return hinfo;
427 return NULL;
430 static void htable_put(struct xt_hashlimit_htable *hinfo)
432 mutex_lock(&hashlimit_mutex);
433 if (--hinfo->use == 0) {
434 hlist_del(&hinfo->node);
435 htable_destroy(hinfo);
437 mutex_unlock(&hashlimit_mutex);
440 /* The algorithm used is the Simple Token Bucket Filter (TBF)
441 * see net/sched/sch_tbf.c in the linux source tree
444 /* Rusty: This is my (non-mathematically-inclined) understanding of
445 this algorithm. The `average rate' in jiffies becomes your initial
446 amount of credit `credit' and the most credit you can ever have
447 `credit_cap'. The `peak rate' becomes the cost of passing the
448 test, `cost'.
450 `prev' tracks the last packet hit: you gain one credit per jiffy.
451 If you get credit balance more than this, the extra credit is
452 discarded. Every time the match passes, you lose `cost' credits;
453 if you don't have that many, the test fails.
455 See Alexey's formal explanation in net/sched/sch_tbf.c.
457 To get the maximum range, we multiply by this factor (ie. you get N
458 credits per jiffy). We want to allow a rate as low as 1 per day
459 (slowest userspace tool allows), which means
460 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
462 #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
463 #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
465 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
466 * us the power of 2 below the theoretical max, so GCC simply does a
467 * shift. */
468 #define _POW2_BELOW2(x) ((x)|((x)>>1))
469 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
470 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
471 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
472 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
473 #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
474 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
475 #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
477 #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
478 #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
480 /* in byte mode, the lowest possible rate is one packet/second.
481 * credit_cap is used as a counter that tells us how many times we can
482 * refill the "credits available" counter when it becomes empty.
484 #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
485 #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
487 static u32 xt_hashlimit_len_to_chunks(u32 len)
489 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
492 /* Precision saver. */
493 static u64 user2credits(u64 user, int revision)
495 u64 scale = (revision == 1) ?
496 XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
497 u64 cpj = (revision == 1) ?
498 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
500 /* Avoid overflow: divide the constant operands first */
501 if (scale >= HZ * cpj)
502 return div64_u64(user, div64_u64(scale, HZ * cpj));
504 return user * div64_u64(HZ * cpj, scale);
507 static u32 user2credits_byte(u32 user)
509 u64 us = user;
510 us *= HZ * CREDITS_PER_JIFFY_BYTES;
511 return (u32) (us >> 32);
514 static u64 user2rate(u64 user)
516 if (user != 0) {
517 return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
518 } else {
519 pr_info_ratelimited("invalid rate from userspace: %llu\n",
520 user);
521 return 0;
525 static u64 user2rate_bytes(u32 user)
527 u64 r;
529 r = user ? U32_MAX / user : U32_MAX;
530 return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
533 static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
534 u32 mode, int revision)
536 unsigned long delta = now - dh->rateinfo.prev;
537 u64 cap, cpj;
539 if (delta == 0)
540 return;
542 if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
543 u64 interval = dh->rateinfo.interval * HZ;
545 if (delta < interval)
546 return;
548 dh->rateinfo.prev = now;
549 dh->rateinfo.prev_window =
550 ((dh->rateinfo.current_rate * interval) >
551 (delta * dh->rateinfo.rate));
552 dh->rateinfo.current_rate = 0;
554 return;
557 dh->rateinfo.prev = now;
559 if (mode & XT_HASHLIMIT_BYTES) {
560 u64 tmp = dh->rateinfo.credit;
561 dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
562 cap = CREDITS_PER_JIFFY_BYTES * HZ;
563 if (tmp >= dh->rateinfo.credit) {/* overflow */
564 dh->rateinfo.credit = cap;
565 return;
567 } else {
568 cpj = (revision == 1) ?
569 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
570 dh->rateinfo.credit += delta * cpj;
571 cap = dh->rateinfo.credit_cap;
573 if (dh->rateinfo.credit > cap)
574 dh->rateinfo.credit = cap;
577 static void rateinfo_init(struct dsthash_ent *dh,
578 struct xt_hashlimit_htable *hinfo, int revision)
580 dh->rateinfo.prev = jiffies;
581 if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
582 dh->rateinfo.prev_window = 0;
583 dh->rateinfo.current_rate = 0;
584 if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
585 dh->rateinfo.rate =
586 user2rate_bytes((u32)hinfo->cfg.avg);
587 if (hinfo->cfg.burst)
588 dh->rateinfo.burst =
589 hinfo->cfg.burst * dh->rateinfo.rate;
590 else
591 dh->rateinfo.burst = dh->rateinfo.rate;
592 } else {
593 dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
594 dh->rateinfo.burst =
595 hinfo->cfg.burst + dh->rateinfo.rate;
597 dh->rateinfo.interval = hinfo->cfg.interval;
598 } else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
599 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
600 dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
601 dh->rateinfo.credit_cap = hinfo->cfg.burst;
602 } else {
603 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
604 hinfo->cfg.burst, revision);
605 dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
606 dh->rateinfo.credit_cap = dh->rateinfo.credit;
610 static inline __be32 maskl(__be32 a, unsigned int l)
612 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
615 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
616 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
618 switch (p) {
619 case 0 ... 31:
620 i[0] = maskl(i[0], p);
621 i[1] = i[2] = i[3] = 0;
622 break;
623 case 32 ... 63:
624 i[1] = maskl(i[1], p - 32);
625 i[2] = i[3] = 0;
626 break;
627 case 64 ... 95:
628 i[2] = maskl(i[2], p - 64);
629 i[3] = 0;
630 break;
631 case 96 ... 127:
632 i[3] = maskl(i[3], p - 96);
633 break;
634 case 128:
635 break;
638 #endif
640 static int
641 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
642 struct dsthash_dst *dst,
643 const struct sk_buff *skb, unsigned int protoff)
645 __be16 _ports[2], *ports;
646 u8 nexthdr;
647 int poff;
649 memset(dst, 0, sizeof(*dst));
651 switch (hinfo->family) {
652 case NFPROTO_IPV4:
653 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
654 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
655 hinfo->cfg.dstmask);
656 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
657 dst->ip.src = maskl(ip_hdr(skb)->saddr,
658 hinfo->cfg.srcmask);
660 if (!(hinfo->cfg.mode &
661 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
662 return 0;
663 nexthdr = ip_hdr(skb)->protocol;
664 break;
665 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
666 case NFPROTO_IPV6:
668 __be16 frag_off;
670 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
671 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
672 sizeof(dst->ip6.dst));
673 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
675 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
676 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
677 sizeof(dst->ip6.src));
678 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
681 if (!(hinfo->cfg.mode &
682 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
683 return 0;
684 nexthdr = ipv6_hdr(skb)->nexthdr;
685 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
686 if ((int)protoff < 0)
687 return -1;
688 break;
690 #endif
691 default:
692 BUG();
693 return 0;
696 poff = proto_ports_offset(nexthdr);
697 if (poff >= 0) {
698 ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
699 &_ports);
700 } else {
701 _ports[0] = _ports[1] = 0;
702 ports = _ports;
704 if (!ports)
705 return -1;
706 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
707 dst->src_port = ports[0];
708 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
709 dst->dst_port = ports[1];
710 return 0;
713 static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
715 u64 tmp = xt_hashlimit_len_to_chunks(len);
716 tmp = tmp * dh->rateinfo.cost;
718 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
719 tmp = CREDITS_PER_JIFFY_BYTES * HZ;
721 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
722 dh->rateinfo.credit_cap--;
723 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
725 return (u32) tmp;
728 static bool
729 hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
730 struct xt_hashlimit_htable *hinfo,
731 const struct hashlimit_cfg3 *cfg, int revision)
733 unsigned long now = jiffies;
734 struct dsthash_ent *dh;
735 struct dsthash_dst dst;
736 bool race = false;
737 u64 cost;
739 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
740 goto hotdrop;
742 local_bh_disable();
743 dh = dsthash_find(hinfo, &dst);
744 if (dh == NULL) {
745 dh = dsthash_alloc_init(hinfo, &dst, &race);
746 if (dh == NULL) {
747 local_bh_enable();
748 goto hotdrop;
749 } else if (race) {
750 /* Already got an entry, update expiration timeout */
751 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
752 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
753 } else {
754 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
755 rateinfo_init(dh, hinfo, revision);
757 } else {
758 /* update expiration timeout */
759 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
760 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
763 if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
764 cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
765 dh->rateinfo.current_rate += cost;
767 if (!dh->rateinfo.prev_window &&
768 (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
769 spin_unlock(&dh->lock);
770 local_bh_enable();
771 return !(cfg->mode & XT_HASHLIMIT_INVERT);
772 } else {
773 goto overlimit;
777 if (cfg->mode & XT_HASHLIMIT_BYTES)
778 cost = hashlimit_byte_cost(skb->len, dh);
779 else
780 cost = dh->rateinfo.cost;
782 if (dh->rateinfo.credit >= cost) {
783 /* below the limit */
784 dh->rateinfo.credit -= cost;
785 spin_unlock(&dh->lock);
786 local_bh_enable();
787 return !(cfg->mode & XT_HASHLIMIT_INVERT);
790 overlimit:
791 spin_unlock(&dh->lock);
792 local_bh_enable();
793 /* default match is underlimit - so over the limit, we need to invert */
794 return cfg->mode & XT_HASHLIMIT_INVERT;
796 hotdrop:
797 par->hotdrop = true;
798 return false;
801 static bool
802 hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
804 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
805 struct xt_hashlimit_htable *hinfo = info->hinfo;
806 struct hashlimit_cfg3 cfg = {};
807 int ret;
809 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
810 if (ret)
811 return ret;
813 return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
816 static bool
817 hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
819 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
820 struct xt_hashlimit_htable *hinfo = info->hinfo;
821 struct hashlimit_cfg3 cfg = {};
822 int ret;
824 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
825 if (ret)
826 return ret;
828 return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
831 static bool
832 hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
834 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
835 struct xt_hashlimit_htable *hinfo = info->hinfo;
837 return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
840 static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
841 struct xt_hashlimit_htable **hinfo,
842 struct hashlimit_cfg3 *cfg,
843 const char *name, int revision)
845 struct net *net = par->net;
846 int ret;
848 if (cfg->gc_interval == 0 || cfg->expire == 0)
849 return -EINVAL;
850 if (par->family == NFPROTO_IPV4) {
851 if (cfg->srcmask > 32 || cfg->dstmask > 32)
852 return -EINVAL;
853 } else {
854 if (cfg->srcmask > 128 || cfg->dstmask > 128)
855 return -EINVAL;
858 if (cfg->mode & ~XT_HASHLIMIT_ALL) {
859 pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
860 cfg->mode);
861 return -EINVAL;
864 /* Check for overflow. */
865 if (revision >= 3 && cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
866 if (cfg->avg == 0 || cfg->avg > U32_MAX) {
867 pr_info_ratelimited("invalid rate\n");
868 return -ERANGE;
871 if (cfg->interval == 0) {
872 pr_info_ratelimited("invalid interval\n");
873 return -EINVAL;
875 } else if (cfg->mode & XT_HASHLIMIT_BYTES) {
876 if (user2credits_byte(cfg->avg) == 0) {
877 pr_info_ratelimited("overflow, rate too high: %llu\n",
878 cfg->avg);
879 return -EINVAL;
881 } else if (cfg->burst == 0 ||
882 user2credits(cfg->avg * cfg->burst, revision) <
883 user2credits(cfg->avg, revision)) {
884 pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
885 cfg->avg, cfg->burst);
886 return -ERANGE;
889 mutex_lock(&hashlimit_mutex);
890 *hinfo = htable_find_get(net, name, par->family);
891 if (*hinfo == NULL) {
892 ret = htable_create(net, cfg, name, par->family,
893 hinfo, revision);
894 if (ret < 0) {
895 mutex_unlock(&hashlimit_mutex);
896 return ret;
899 mutex_unlock(&hashlimit_mutex);
901 return 0;
904 static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
906 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
907 struct hashlimit_cfg3 cfg = {};
908 int ret;
910 ret = xt_check_proc_name(info->name, sizeof(info->name));
911 if (ret)
912 return ret;
914 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
915 if (ret)
916 return ret;
918 return hashlimit_mt_check_common(par, &info->hinfo,
919 &cfg, info->name, 1);
922 static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
924 struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
925 struct hashlimit_cfg3 cfg = {};
926 int ret;
928 ret = xt_check_proc_name(info->name, sizeof(info->name));
929 if (ret)
930 return ret;
932 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
933 if (ret)
934 return ret;
936 return hashlimit_mt_check_common(par, &info->hinfo,
937 &cfg, info->name, 2);
940 static int hashlimit_mt_check(const struct xt_mtchk_param *par)
942 struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
943 int ret;
945 ret = xt_check_proc_name(info->name, sizeof(info->name));
946 if (ret)
947 return ret;
949 return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
950 info->name, 3);
953 static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
955 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
957 htable_put(info->hinfo);
960 static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
962 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
964 htable_put(info->hinfo);
967 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
969 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
971 htable_put(info->hinfo);
974 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
976 .name = "hashlimit",
977 .revision = 1,
978 .family = NFPROTO_IPV4,
979 .match = hashlimit_mt_v1,
980 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
981 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
982 .checkentry = hashlimit_mt_check_v1,
983 .destroy = hashlimit_mt_destroy_v1,
984 .me = THIS_MODULE,
987 .name = "hashlimit",
988 .revision = 2,
989 .family = NFPROTO_IPV4,
990 .match = hashlimit_mt_v2,
991 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
992 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
993 .checkentry = hashlimit_mt_check_v2,
994 .destroy = hashlimit_mt_destroy_v2,
995 .me = THIS_MODULE,
998 .name = "hashlimit",
999 .revision = 3,
1000 .family = NFPROTO_IPV4,
1001 .match = hashlimit_mt,
1002 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1003 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1004 .checkentry = hashlimit_mt_check,
1005 .destroy = hashlimit_mt_destroy,
1006 .me = THIS_MODULE,
1008 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1010 .name = "hashlimit",
1011 .revision = 1,
1012 .family = NFPROTO_IPV6,
1013 .match = hashlimit_mt_v1,
1014 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
1015 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
1016 .checkentry = hashlimit_mt_check_v1,
1017 .destroy = hashlimit_mt_destroy_v1,
1018 .me = THIS_MODULE,
1021 .name = "hashlimit",
1022 .revision = 2,
1023 .family = NFPROTO_IPV6,
1024 .match = hashlimit_mt_v2,
1025 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1026 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1027 .checkentry = hashlimit_mt_check_v2,
1028 .destroy = hashlimit_mt_destroy_v2,
1029 .me = THIS_MODULE,
1032 .name = "hashlimit",
1033 .revision = 3,
1034 .family = NFPROTO_IPV6,
1035 .match = hashlimit_mt,
1036 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1037 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1038 .checkentry = hashlimit_mt_check,
1039 .destroy = hashlimit_mt_destroy,
1040 .me = THIS_MODULE,
1042 #endif
1045 /* PROC stuff */
1046 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1047 __acquires(htable->lock)
1049 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1050 unsigned int *bucket;
1052 spin_lock_bh(&htable->lock);
1053 if (*pos >= htable->cfg.size)
1054 return NULL;
1056 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1057 if (!bucket)
1058 return ERR_PTR(-ENOMEM);
1060 *bucket = *pos;
1061 return bucket;
1064 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1066 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1067 unsigned int *bucket = v;
1069 *pos = ++(*bucket);
1070 if (*pos >= htable->cfg.size) {
1071 kfree(v);
1072 return NULL;
1074 return bucket;
1077 static void dl_seq_stop(struct seq_file *s, void *v)
1078 __releases(htable->lock)
1080 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1081 unsigned int *bucket = v;
1083 if (!IS_ERR(bucket))
1084 kfree(bucket);
1085 spin_unlock_bh(&htable->lock);
1088 static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1089 struct seq_file *s)
1091 switch (family) {
1092 case NFPROTO_IPV4:
1093 seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1094 (long)(ent->expires - jiffies)/HZ,
1095 &ent->dst.ip.src,
1096 ntohs(ent->dst.src_port),
1097 &ent->dst.ip.dst,
1098 ntohs(ent->dst.dst_port),
1099 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1100 ent->rateinfo.cost);
1101 break;
1102 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1103 case NFPROTO_IPV6:
1104 seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1105 (long)(ent->expires - jiffies)/HZ,
1106 &ent->dst.ip6.src,
1107 ntohs(ent->dst.src_port),
1108 &ent->dst.ip6.dst,
1109 ntohs(ent->dst.dst_port),
1110 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1111 ent->rateinfo.cost);
1112 break;
1113 #endif
1114 default:
1115 BUG();
1119 static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1120 struct seq_file *s)
1122 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1124 spin_lock(&ent->lock);
1125 /* recalculate to show accurate numbers */
1126 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1128 dl_seq_print(ent, family, s);
1130 spin_unlock(&ent->lock);
1131 return seq_has_overflowed(s);
1134 static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1135 struct seq_file *s)
1137 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1139 spin_lock(&ent->lock);
1140 /* recalculate to show accurate numbers */
1141 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1143 dl_seq_print(ent, family, s);
1145 spin_unlock(&ent->lock);
1146 return seq_has_overflowed(s);
1149 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1150 struct seq_file *s)
1152 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1154 spin_lock(&ent->lock);
1155 /* recalculate to show accurate numbers */
1156 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1158 dl_seq_print(ent, family, s);
1160 spin_unlock(&ent->lock);
1161 return seq_has_overflowed(s);
1164 static int dl_seq_show_v2(struct seq_file *s, void *v)
1166 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1167 unsigned int *bucket = (unsigned int *)v;
1168 struct dsthash_ent *ent;
1170 if (!hlist_empty(&htable->hash[*bucket])) {
1171 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1172 if (dl_seq_real_show_v2(ent, htable->family, s))
1173 return -1;
1175 return 0;
1178 static int dl_seq_show_v1(struct seq_file *s, void *v)
1180 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1181 unsigned int *bucket = v;
1182 struct dsthash_ent *ent;
1184 if (!hlist_empty(&htable->hash[*bucket])) {
1185 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1186 if (dl_seq_real_show_v1(ent, htable->family, s))
1187 return -1;
1189 return 0;
1192 static int dl_seq_show(struct seq_file *s, void *v)
1194 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1195 unsigned int *bucket = v;
1196 struct dsthash_ent *ent;
1198 if (!hlist_empty(&htable->hash[*bucket])) {
1199 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1200 if (dl_seq_real_show(ent, htable->family, s))
1201 return -1;
1203 return 0;
1206 static const struct seq_operations dl_seq_ops_v1 = {
1207 .start = dl_seq_start,
1208 .next = dl_seq_next,
1209 .stop = dl_seq_stop,
1210 .show = dl_seq_show_v1
1213 static const struct seq_operations dl_seq_ops_v2 = {
1214 .start = dl_seq_start,
1215 .next = dl_seq_next,
1216 .stop = dl_seq_stop,
1217 .show = dl_seq_show_v2
1220 static const struct seq_operations dl_seq_ops = {
1221 .start = dl_seq_start,
1222 .next = dl_seq_next,
1223 .stop = dl_seq_stop,
1224 .show = dl_seq_show
1227 static int __net_init hashlimit_proc_net_init(struct net *net)
1229 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1231 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1232 if (!hashlimit_net->ipt_hashlimit)
1233 return -ENOMEM;
1234 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1235 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1236 if (!hashlimit_net->ip6t_hashlimit) {
1237 remove_proc_entry("ipt_hashlimit", net->proc_net);
1238 return -ENOMEM;
1240 #endif
1241 return 0;
1244 static void __net_exit hashlimit_proc_net_exit(struct net *net)
1246 struct xt_hashlimit_htable *hinfo;
1247 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1249 /* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1250 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1251 * entries is empty before trying to remove it.
1253 mutex_lock(&hashlimit_mutex);
1254 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1255 htable_remove_proc_entry(hinfo);
1256 hashlimit_net->ipt_hashlimit = NULL;
1257 hashlimit_net->ip6t_hashlimit = NULL;
1258 mutex_unlock(&hashlimit_mutex);
1260 remove_proc_entry("ipt_hashlimit", net->proc_net);
1261 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1262 remove_proc_entry("ip6t_hashlimit", net->proc_net);
1263 #endif
1266 static int __net_init hashlimit_net_init(struct net *net)
1268 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1270 INIT_HLIST_HEAD(&hashlimit_net->htables);
1271 return hashlimit_proc_net_init(net);
1274 static void __net_exit hashlimit_net_exit(struct net *net)
1276 hashlimit_proc_net_exit(net);
1279 static struct pernet_operations hashlimit_net_ops = {
1280 .init = hashlimit_net_init,
1281 .exit = hashlimit_net_exit,
1282 .id = &hashlimit_net_id,
1283 .size = sizeof(struct hashlimit_net),
1286 static int __init hashlimit_mt_init(void)
1288 int err;
1290 err = register_pernet_subsys(&hashlimit_net_ops);
1291 if (err < 0)
1292 return err;
1293 err = xt_register_matches(hashlimit_mt_reg,
1294 ARRAY_SIZE(hashlimit_mt_reg));
1295 if (err < 0)
1296 goto err1;
1298 err = -ENOMEM;
1299 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1300 sizeof(struct dsthash_ent), 0, 0,
1301 NULL);
1302 if (!hashlimit_cachep) {
1303 pr_warn("unable to create slab cache\n");
1304 goto err2;
1306 return 0;
1308 err2:
1309 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1310 err1:
1311 unregister_pernet_subsys(&hashlimit_net_ops);
1312 return err;
1316 static void __exit hashlimit_mt_exit(void)
1318 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1319 unregister_pernet_subsys(&hashlimit_net_ops);
1321 rcu_barrier();
1322 kmem_cache_destroy(hashlimit_cachep);
1325 module_init(hashlimit_mt_init);
1326 module_exit(hashlimit_mt_exit);