Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * xt_hashlimit - Netfilter module to limit the number of packets per time
4 * separately for each hashbucket (sourceip/sourceport/dstip/dstport)
5 *
6 * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
7 * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
8 * Copyright © CC Computer Consultants GmbH, 2007 - 2008
9 *
10 * Development of this code was funded by Astaro AG, http://www.astaro.com/
11 */
12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13#include <linux/module.h>
14#include <linux/spinlock.h>
15#include <linux/random.h>
16#include <linux/jhash.h>
17#include <linux/slab.h>
18#include <linux/vmalloc.h>
19#include <linux/proc_fs.h>
20#include <linux/seq_file.h>
21#include <linux/list.h>
22#include <linux/skbuff.h>
23#include <linux/mm.h>
24#include <linux/in.h>
25#include <linux/ip.h>
26#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
27#include <linux/ipv6.h>
28#include <net/ipv6.h>
29#endif
30
31#include <net/net_namespace.h>
32#include <net/netns/generic.h>
33
34#include <linux/netfilter/x_tables.h>
35#include <linux/netfilter_ipv4/ip_tables.h>
36#include <linux/netfilter_ipv6/ip6_tables.h>
37#include <linux/mutex.h>
38#include <linux/kernel.h>
39#include <linux/refcount.h>
40#include <uapi/linux/netfilter/xt_hashlimit.h>
41
42#define XT_HASHLIMIT_ALL (XT_HASHLIMIT_HASH_DIP | XT_HASHLIMIT_HASH_DPT | \
43 XT_HASHLIMIT_HASH_SIP | XT_HASHLIMIT_HASH_SPT | \
44 XT_HASHLIMIT_INVERT | XT_HASHLIMIT_BYTES |\
45 XT_HASHLIMIT_RATE_MATCH)
46
47MODULE_LICENSE("GPL");
48MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
49MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
50MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
51MODULE_ALIAS("ipt_hashlimit");
52MODULE_ALIAS("ip6t_hashlimit");
53
54struct hashlimit_net {
55 struct hlist_head htables;
56 struct proc_dir_entry *ipt_hashlimit;
57 struct proc_dir_entry *ip6t_hashlimit;
58};
59
60static unsigned int hashlimit_net_id;
61static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
62{
63 return net_generic(net, hashlimit_net_id);
64}
65
66/* need to declare this at the top */
67static const struct seq_operations dl_seq_ops_v2;
68static const struct seq_operations dl_seq_ops_v1;
69static const struct seq_operations dl_seq_ops;
70
71/* hash table crap */
72struct dsthash_dst {
73 union {
74 struct {
75 __be32 src;
76 __be32 dst;
77 } ip;
78#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
79 struct {
80 __be32 src[4];
81 __be32 dst[4];
82 } ip6;
83#endif
84 };
85 __be16 src_port;
86 __be16 dst_port;
87};
88
89struct dsthash_ent {
90 /* static / read-only parts in the beginning */
91 struct hlist_node node;
92 struct dsthash_dst dst;
93
94 /* modified structure members in the end */
95 spinlock_t lock;
96 unsigned long expires; /* precalculated expiry time */
97 struct {
98 unsigned long prev; /* last modification */
99 union {
100 struct {
101 u_int64_t credit;
102 u_int64_t credit_cap;
103 u_int64_t cost;
104 };
105 struct {
106 u_int32_t interval, prev_window;
107 u_int64_t current_rate;
108 u_int64_t rate;
109 int64_t burst;
110 };
111 };
112 } rateinfo;
113 struct rcu_head rcu;
114};
115
116struct xt_hashlimit_htable {
117 struct hlist_node node; /* global list of all htables */
118 refcount_t use;
119 u_int8_t family;
120 bool rnd_initialized;
121
122 struct hashlimit_cfg3 cfg; /* config */
123
124 /* used internally */
125 spinlock_t lock; /* lock for list_head */
126 u_int32_t rnd; /* random seed for hash */
127 unsigned int count; /* number entries in table */
128 struct delayed_work gc_work;
129
130 /* seq_file stuff */
131 struct proc_dir_entry *pde;
132 const char *name;
133 struct net *net;
134
135 struct hlist_head hash[]; /* hashtable itself */
136};
137
138static int
139cfg_copy(struct hashlimit_cfg3 *to, const void *from, int revision)
140{
141 if (revision == 1) {
142 struct hashlimit_cfg1 *cfg = (struct hashlimit_cfg1 *)from;
143
144 to->mode = cfg->mode;
145 to->avg = cfg->avg;
146 to->burst = cfg->burst;
147 to->size = cfg->size;
148 to->max = cfg->max;
149 to->gc_interval = cfg->gc_interval;
150 to->expire = cfg->expire;
151 to->srcmask = cfg->srcmask;
152 to->dstmask = cfg->dstmask;
153 } else if (revision == 2) {
154 struct hashlimit_cfg2 *cfg = (struct hashlimit_cfg2 *)from;
155
156 to->mode = cfg->mode;
157 to->avg = cfg->avg;
158 to->burst = cfg->burst;
159 to->size = cfg->size;
160 to->max = cfg->max;
161 to->gc_interval = cfg->gc_interval;
162 to->expire = cfg->expire;
163 to->srcmask = cfg->srcmask;
164 to->dstmask = cfg->dstmask;
165 } else if (revision == 3) {
166 memcpy(to, from, sizeof(struct hashlimit_cfg3));
167 } else {
168 return -EINVAL;
169 }
170
171 return 0;
172}
173
174static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */
175static struct kmem_cache *hashlimit_cachep __read_mostly;
176
177static inline bool dst_cmp(const struct dsthash_ent *ent,
178 const struct dsthash_dst *b)
179{
180 return !memcmp(&ent->dst, b, sizeof(ent->dst));
181}
182
183static u_int32_t
184hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
185{
186 u_int32_t hash = jhash2((const u32 *)dst,
187 sizeof(*dst)/sizeof(u32),
188 ht->rnd);
189 /*
190 * Instead of returning hash % ht->cfg.size (implying a divide)
191 * we return the high 32 bits of the (hash * ht->cfg.size) that will
192 * give results between [0 and cfg.size-1] and same hash distribution,
193 * but using a multiply, less expensive than a divide
194 */
195 return reciprocal_scale(hash, ht->cfg.size);
196}
197
198static struct dsthash_ent *
199dsthash_find(const struct xt_hashlimit_htable *ht,
200 const struct dsthash_dst *dst)
201{
202 struct dsthash_ent *ent;
203 u_int32_t hash = hash_dst(ht, dst);
204
205 if (!hlist_empty(&ht->hash[hash])) {
206 hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
207 if (dst_cmp(ent, dst)) {
208 spin_lock(&ent->lock);
209 return ent;
210 }
211 }
212 return NULL;
213}
214
215/* allocate dsthash_ent, initialize dst, put in htable and lock it */
216static struct dsthash_ent *
217dsthash_alloc_init(struct xt_hashlimit_htable *ht,
218 const struct dsthash_dst *dst, bool *race)
219{
220 struct dsthash_ent *ent;
221
222 spin_lock(&ht->lock);
223
224 /* Two or more packets may race to create the same entry in the
225 * hashtable, double check if this packet lost race.
226 */
227 ent = dsthash_find(ht, dst);
228 if (ent != NULL) {
229 spin_unlock(&ht->lock);
230 *race = true;
231 return ent;
232 }
233
234 /* initialize hash with random val at the time we allocate
235 * the first hashtable entry */
236 if (unlikely(!ht->rnd_initialized)) {
237 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
238 ht->rnd_initialized = true;
239 }
240
241 if (ht->cfg.max && ht->count >= ht->cfg.max) {
242 /* FIXME: do something. question is what.. */
243 net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
244 ent = NULL;
245 } else
246 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
247 if (ent) {
248 memcpy(&ent->dst, dst, sizeof(ent->dst));
249 spin_lock_init(&ent->lock);
250
251 spin_lock(&ent->lock);
252 hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
253 ht->count++;
254 }
255 spin_unlock(&ht->lock);
256 return ent;
257}
258
259static void dsthash_free_rcu(struct rcu_head *head)
260{
261 struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
262
263 kmem_cache_free(hashlimit_cachep, ent);
264}
265
266static inline void
267dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
268{
269 hlist_del_rcu(&ent->node);
270 call_rcu(&ent->rcu, dsthash_free_rcu);
271 ht->count--;
272}
273static void htable_gc(struct work_struct *work);
274
275static int htable_create(struct net *net, struct hashlimit_cfg3 *cfg,
276 const char *name, u_int8_t family,
277 struct xt_hashlimit_htable **out_hinfo,
278 int revision)
279{
280 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
281 struct xt_hashlimit_htable *hinfo;
282 const struct seq_operations *ops;
283 unsigned int size, i;
284 unsigned long nr_pages = totalram_pages();
285 int ret;
286
287 if (cfg->size) {
288 size = cfg->size;
289 } else {
290 size = (nr_pages << PAGE_SHIFT) / 16384 /
291 sizeof(struct hlist_head);
292 if (nr_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
293 size = 8192;
294 if (size < 16)
295 size = 16;
296 }
297 /* FIXME: don't use vmalloc() here or anywhere else -HW */
298 hinfo = vmalloc(struct_size(hinfo, hash, size));
299 if (hinfo == NULL)
300 return -ENOMEM;
301 *out_hinfo = hinfo;
302
303 /* copy match config into hashtable config */
304 ret = cfg_copy(&hinfo->cfg, (void *)cfg, 3);
305 if (ret) {
306 vfree(hinfo);
307 return ret;
308 }
309
310 hinfo->cfg.size = size;
311 if (hinfo->cfg.max == 0)
312 hinfo->cfg.max = 8 * hinfo->cfg.size;
313 else if (hinfo->cfg.max < hinfo->cfg.size)
314 hinfo->cfg.max = hinfo->cfg.size;
315
316 for (i = 0; i < hinfo->cfg.size; i++)
317 INIT_HLIST_HEAD(&hinfo->hash[i]);
318
319 refcount_set(&hinfo->use, 1);
320 hinfo->count = 0;
321 hinfo->family = family;
322 hinfo->rnd_initialized = false;
323 hinfo->name = kstrdup(name, GFP_KERNEL);
324 if (!hinfo->name) {
325 vfree(hinfo);
326 return -ENOMEM;
327 }
328 spin_lock_init(&hinfo->lock);
329
330 switch (revision) {
331 case 1:
332 ops = &dl_seq_ops_v1;
333 break;
334 case 2:
335 ops = &dl_seq_ops_v2;
336 break;
337 default:
338 ops = &dl_seq_ops;
339 }
340
341 hinfo->pde = proc_create_seq_data(name, 0,
342 (family == NFPROTO_IPV4) ?
343 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
344 ops, hinfo);
345 if (hinfo->pde == NULL) {
346 kfree(hinfo->name);
347 vfree(hinfo);
348 return -ENOMEM;
349 }
350 hinfo->net = net;
351
352 INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
353 queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
354 msecs_to_jiffies(hinfo->cfg.gc_interval));
355
356 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
357
358 return 0;
359}
360
361static void htable_selective_cleanup(struct xt_hashlimit_htable *ht, bool select_all)
362{
363 unsigned int i;
364
365 for (i = 0; i < ht->cfg.size; i++) {
366 struct dsthash_ent *dh;
367 struct hlist_node *n;
368
369 spin_lock_bh(&ht->lock);
370 hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
371 if (time_after_eq(jiffies, dh->expires) || select_all)
372 dsthash_free(ht, dh);
373 }
374 spin_unlock_bh(&ht->lock);
375 cond_resched();
376 }
377}
378
379static void htable_gc(struct work_struct *work)
380{
381 struct xt_hashlimit_htable *ht;
382
383 ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
384
385 htable_selective_cleanup(ht, false);
386
387 queue_delayed_work(system_power_efficient_wq,
388 &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
389}
390
391static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
392{
393 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
394 struct proc_dir_entry *parent;
395
396 if (hinfo->family == NFPROTO_IPV4)
397 parent = hashlimit_net->ipt_hashlimit;
398 else
399 parent = hashlimit_net->ip6t_hashlimit;
400
401 if (parent != NULL)
402 remove_proc_entry(hinfo->name, parent);
403}
404
405static struct xt_hashlimit_htable *htable_find_get(struct net *net,
406 const char *name,
407 u_int8_t family)
408{
409 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
410 struct xt_hashlimit_htable *hinfo;
411
412 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
413 if (!strcmp(name, hinfo->name) &&
414 hinfo->family == family) {
415 refcount_inc(&hinfo->use);
416 return hinfo;
417 }
418 }
419 return NULL;
420}
421
422static void htable_put(struct xt_hashlimit_htable *hinfo)
423{
424 if (refcount_dec_and_mutex_lock(&hinfo->use, &hashlimit_mutex)) {
425 hlist_del(&hinfo->node);
426 htable_remove_proc_entry(hinfo);
427 mutex_unlock(&hashlimit_mutex);
428
429 cancel_delayed_work_sync(&hinfo->gc_work);
430 htable_selective_cleanup(hinfo, true);
431 kfree(hinfo->name);
432 vfree(hinfo);
433 }
434}
435
436/* The algorithm used is the Simple Token Bucket Filter (TBF)
437 * see net/sched/sch_tbf.c in the linux source tree
438 */
439
440/* Rusty: This is my (non-mathematically-inclined) understanding of
441 this algorithm. The `average rate' in jiffies becomes your initial
442 amount of credit `credit' and the most credit you can ever have
443 `credit_cap'. The `peak rate' becomes the cost of passing the
444 test, `cost'.
445
446 `prev' tracks the last packet hit: you gain one credit per jiffy.
447 If you get credit balance more than this, the extra credit is
448 discarded. Every time the match passes, you lose `cost' credits;
449 if you don't have that many, the test fails.
450
451 See Alexey's formal explanation in net/sched/sch_tbf.c.
452
453 To get the maximum range, we multiply by this factor (ie. you get N
454 credits per jiffy). We want to allow a rate as low as 1 per day
455 (slowest userspace tool allows), which means
456 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
457*/
458#define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
459#define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
460
461/* Repeated shift and or gives us all 1s, final shift and add 1 gives
462 * us the power of 2 below the theoretical max, so GCC simply does a
463 * shift. */
464#define _POW2_BELOW2(x) ((x)|((x)>>1))
465#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
466#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
467#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
468#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
469#define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
470#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
471#define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
472
473#define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
474#define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
475
476/* in byte mode, the lowest possible rate is one packet/second.
477 * credit_cap is used as a counter that tells us how many times we can
478 * refill the "credits available" counter when it becomes empty.
479 */
480#define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
481#define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
482
483static u32 xt_hashlimit_len_to_chunks(u32 len)
484{
485 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
486}
487
488/* Precision saver. */
489static u64 user2credits(u64 user, int revision)
490{
491 u64 scale = (revision == 1) ?
492 XT_HASHLIMIT_SCALE : XT_HASHLIMIT_SCALE_v2;
493 u64 cpj = (revision == 1) ?
494 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
495
496 /* Avoid overflow: divide the constant operands first */
497 if (scale >= HZ * cpj)
498 return div64_u64(user, div64_u64(scale, HZ * cpj));
499
500 return user * div64_u64(HZ * cpj, scale);
501}
502
503static u32 user2credits_byte(u32 user)
504{
505 u64 us = user;
506 us *= HZ * CREDITS_PER_JIFFY_BYTES;
507 return (u32) (us >> 32);
508}
509
510static u64 user2rate(u64 user)
511{
512 if (user != 0) {
513 return div64_u64(XT_HASHLIMIT_SCALE_v2, user);
514 } else {
515 pr_info_ratelimited("invalid rate from userspace: %llu\n",
516 user);
517 return 0;
518 }
519}
520
521static u64 user2rate_bytes(u32 user)
522{
523 u64 r;
524
525 r = user ? U32_MAX / user : U32_MAX;
526 return (r - 1) << XT_HASHLIMIT_BYTE_SHIFT;
527}
528
529static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now,
530 u32 mode, int revision)
531{
532 unsigned long delta = now - dh->rateinfo.prev;
533 u64 cap, cpj;
534
535 if (delta == 0)
536 return;
537
538 if (revision >= 3 && mode & XT_HASHLIMIT_RATE_MATCH) {
539 u64 interval = dh->rateinfo.interval * HZ;
540
541 if (delta < interval)
542 return;
543
544 dh->rateinfo.prev = now;
545 dh->rateinfo.prev_window =
546 ((dh->rateinfo.current_rate * interval) >
547 (delta * dh->rateinfo.rate));
548 dh->rateinfo.current_rate = 0;
549
550 return;
551 }
552
553 dh->rateinfo.prev = now;
554
555 if (mode & XT_HASHLIMIT_BYTES) {
556 u64 tmp = dh->rateinfo.credit;
557 dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
558 cap = CREDITS_PER_JIFFY_BYTES * HZ;
559 if (tmp >= dh->rateinfo.credit) {/* overflow */
560 dh->rateinfo.credit = cap;
561 return;
562 }
563 } else {
564 cpj = (revision == 1) ?
565 CREDITS_PER_JIFFY_v1 : CREDITS_PER_JIFFY;
566 dh->rateinfo.credit += delta * cpj;
567 cap = dh->rateinfo.credit_cap;
568 }
569 if (dh->rateinfo.credit > cap)
570 dh->rateinfo.credit = cap;
571}
572
573static void rateinfo_init(struct dsthash_ent *dh,
574 struct xt_hashlimit_htable *hinfo, int revision)
575{
576 dh->rateinfo.prev = jiffies;
577 if (revision >= 3 && hinfo->cfg.mode & XT_HASHLIMIT_RATE_MATCH) {
578 dh->rateinfo.prev_window = 0;
579 dh->rateinfo.current_rate = 0;
580 if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
581 dh->rateinfo.rate =
582 user2rate_bytes((u32)hinfo->cfg.avg);
583 if (hinfo->cfg.burst)
584 dh->rateinfo.burst =
585 hinfo->cfg.burst * dh->rateinfo.rate;
586 else
587 dh->rateinfo.burst = dh->rateinfo.rate;
588 } else {
589 dh->rateinfo.rate = user2rate(hinfo->cfg.avg);
590 dh->rateinfo.burst =
591 hinfo->cfg.burst + dh->rateinfo.rate;
592 }
593 dh->rateinfo.interval = hinfo->cfg.interval;
594 } else if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
595 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
596 dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
597 dh->rateinfo.credit_cap = hinfo->cfg.burst;
598 } else {
599 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
600 hinfo->cfg.burst, revision);
601 dh->rateinfo.cost = user2credits(hinfo->cfg.avg, revision);
602 dh->rateinfo.credit_cap = dh->rateinfo.credit;
603 }
604}
605
606static inline __be32 maskl(__be32 a, unsigned int l)
607{
608 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
609}
610
611#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
612static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
613{
614 switch (p) {
615 case 0 ... 31:
616 i[0] = maskl(i[0], p);
617 i[1] = i[2] = i[3] = 0;
618 break;
619 case 32 ... 63:
620 i[1] = maskl(i[1], p - 32);
621 i[2] = i[3] = 0;
622 break;
623 case 64 ... 95:
624 i[2] = maskl(i[2], p - 64);
625 i[3] = 0;
626 break;
627 case 96 ... 127:
628 i[3] = maskl(i[3], p - 96);
629 break;
630 case 128:
631 break;
632 }
633}
634#endif
635
636static int
637hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
638 struct dsthash_dst *dst,
639 const struct sk_buff *skb, unsigned int protoff)
640{
641 __be16 _ports[2], *ports;
642 u8 nexthdr;
643 int poff;
644
645 memset(dst, 0, sizeof(*dst));
646
647 switch (hinfo->family) {
648 case NFPROTO_IPV4:
649 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
650 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
651 hinfo->cfg.dstmask);
652 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
653 dst->ip.src = maskl(ip_hdr(skb)->saddr,
654 hinfo->cfg.srcmask);
655
656 if (!(hinfo->cfg.mode &
657 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
658 return 0;
659 nexthdr = ip_hdr(skb)->protocol;
660 break;
661#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
662 case NFPROTO_IPV6:
663 {
664 __be16 frag_off;
665
666 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
667 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
668 sizeof(dst->ip6.dst));
669 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
670 }
671 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
672 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
673 sizeof(dst->ip6.src));
674 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
675 }
676
677 if (!(hinfo->cfg.mode &
678 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
679 return 0;
680 nexthdr = ipv6_hdr(skb)->nexthdr;
681 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
682 if ((int)protoff < 0)
683 return -1;
684 break;
685 }
686#endif
687 default:
688 BUG();
689 return 0;
690 }
691
692 poff = proto_ports_offset(nexthdr);
693 if (poff >= 0) {
694 ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
695 &_ports);
696 } else {
697 _ports[0] = _ports[1] = 0;
698 ports = _ports;
699 }
700 if (!ports)
701 return -1;
702 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
703 dst->src_port = ports[0];
704 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
705 dst->dst_port = ports[1];
706 return 0;
707}
708
709static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
710{
711 u64 tmp = xt_hashlimit_len_to_chunks(len);
712 tmp = tmp * dh->rateinfo.cost;
713
714 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
715 tmp = CREDITS_PER_JIFFY_BYTES * HZ;
716
717 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
718 dh->rateinfo.credit_cap--;
719 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
720 }
721 return (u32) tmp;
722}
723
724static bool
725hashlimit_mt_common(const struct sk_buff *skb, struct xt_action_param *par,
726 struct xt_hashlimit_htable *hinfo,
727 const struct hashlimit_cfg3 *cfg, int revision)
728{
729 unsigned long now = jiffies;
730 struct dsthash_ent *dh;
731 struct dsthash_dst dst;
732 bool race = false;
733 u64 cost;
734
735 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
736 goto hotdrop;
737
738 local_bh_disable();
739 dh = dsthash_find(hinfo, &dst);
740 if (dh == NULL) {
741 dh = dsthash_alloc_init(hinfo, &dst, &race);
742 if (dh == NULL) {
743 local_bh_enable();
744 goto hotdrop;
745 } else if (race) {
746 /* Already got an entry, update expiration timeout */
747 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
748 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
749 } else {
750 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
751 rateinfo_init(dh, hinfo, revision);
752 }
753 } else {
754 /* update expiration timeout */
755 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
756 rateinfo_recalc(dh, now, hinfo->cfg.mode, revision);
757 }
758
759 if (cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
760 cost = (cfg->mode & XT_HASHLIMIT_BYTES) ? skb->len : 1;
761 dh->rateinfo.current_rate += cost;
762
763 if (!dh->rateinfo.prev_window &&
764 (dh->rateinfo.current_rate <= dh->rateinfo.burst)) {
765 spin_unlock(&dh->lock);
766 local_bh_enable();
767 return !(cfg->mode & XT_HASHLIMIT_INVERT);
768 } else {
769 goto overlimit;
770 }
771 }
772
773 if (cfg->mode & XT_HASHLIMIT_BYTES)
774 cost = hashlimit_byte_cost(skb->len, dh);
775 else
776 cost = dh->rateinfo.cost;
777
778 if (dh->rateinfo.credit >= cost) {
779 /* below the limit */
780 dh->rateinfo.credit -= cost;
781 spin_unlock(&dh->lock);
782 local_bh_enable();
783 return !(cfg->mode & XT_HASHLIMIT_INVERT);
784 }
785
786overlimit:
787 spin_unlock(&dh->lock);
788 local_bh_enable();
789 /* default match is underlimit - so over the limit, we need to invert */
790 return cfg->mode & XT_HASHLIMIT_INVERT;
791
792 hotdrop:
793 par->hotdrop = true;
794 return false;
795}
796
797static bool
798hashlimit_mt_v1(const struct sk_buff *skb, struct xt_action_param *par)
799{
800 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
801 struct xt_hashlimit_htable *hinfo = info->hinfo;
802 struct hashlimit_cfg3 cfg = {};
803 int ret;
804
805 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
806 if (ret)
807 return ret;
808
809 return hashlimit_mt_common(skb, par, hinfo, &cfg, 1);
810}
811
812static bool
813hashlimit_mt_v2(const struct sk_buff *skb, struct xt_action_param *par)
814{
815 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
816 struct xt_hashlimit_htable *hinfo = info->hinfo;
817 struct hashlimit_cfg3 cfg = {};
818 int ret;
819
820 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
821 if (ret)
822 return ret;
823
824 return hashlimit_mt_common(skb, par, hinfo, &cfg, 2);
825}
826
827static bool
828hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
829{
830 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
831 struct xt_hashlimit_htable *hinfo = info->hinfo;
832
833 return hashlimit_mt_common(skb, par, hinfo, &info->cfg, 3);
834}
835
836#define HASHLIMIT_MAX_SIZE 1048576
837
838static int hashlimit_mt_check_common(const struct xt_mtchk_param *par,
839 struct xt_hashlimit_htable **hinfo,
840 struct hashlimit_cfg3 *cfg,
841 const char *name, int revision)
842{
843 struct net *net = par->net;
844 int ret;
845
846 if (cfg->gc_interval == 0 || cfg->expire == 0)
847 return -EINVAL;
848 if (cfg->size > HASHLIMIT_MAX_SIZE) {
849 cfg->size = HASHLIMIT_MAX_SIZE;
850 pr_info_ratelimited("size too large, truncated to %u\n", cfg->size);
851 }
852 if (cfg->max > HASHLIMIT_MAX_SIZE) {
853 cfg->max = HASHLIMIT_MAX_SIZE;
854 pr_info_ratelimited("max too large, truncated to %u\n", cfg->max);
855 }
856 if (par->family == NFPROTO_IPV4) {
857 if (cfg->srcmask > 32 || cfg->dstmask > 32)
858 return -EINVAL;
859 } else {
860 if (cfg->srcmask > 128 || cfg->dstmask > 128)
861 return -EINVAL;
862 }
863
864 if (cfg->mode & ~XT_HASHLIMIT_ALL) {
865 pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
866 cfg->mode);
867 return -EINVAL;
868 }
869
870 /* Check for overflow. */
871 if (revision >= 3 && cfg->mode & XT_HASHLIMIT_RATE_MATCH) {
872 if (cfg->avg == 0 || cfg->avg > U32_MAX) {
873 pr_info_ratelimited("invalid rate\n");
874 return -ERANGE;
875 }
876
877 if (cfg->interval == 0) {
878 pr_info_ratelimited("invalid interval\n");
879 return -EINVAL;
880 }
881 } else if (cfg->mode & XT_HASHLIMIT_BYTES) {
882 if (user2credits_byte(cfg->avg) == 0) {
883 pr_info_ratelimited("overflow, rate too high: %llu\n",
884 cfg->avg);
885 return -EINVAL;
886 }
887 } else if (cfg->burst == 0 ||
888 user2credits(cfg->avg * cfg->burst, revision) <
889 user2credits(cfg->avg, revision)) {
890 pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
891 cfg->avg, cfg->burst);
892 return -ERANGE;
893 }
894
895 mutex_lock(&hashlimit_mutex);
896 *hinfo = htable_find_get(net, name, par->family);
897 if (*hinfo == NULL) {
898 ret = htable_create(net, cfg, name, par->family,
899 hinfo, revision);
900 if (ret < 0) {
901 mutex_unlock(&hashlimit_mutex);
902 return ret;
903 }
904 }
905 mutex_unlock(&hashlimit_mutex);
906
907 return 0;
908}
909
910static int hashlimit_mt_check_v1(const struct xt_mtchk_param *par)
911{
912 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
913 struct hashlimit_cfg3 cfg = {};
914 int ret;
915
916 ret = xt_check_proc_name(info->name, sizeof(info->name));
917 if (ret)
918 return ret;
919
920 ret = cfg_copy(&cfg, (void *)&info->cfg, 1);
921 if (ret)
922 return ret;
923
924 return hashlimit_mt_check_common(par, &info->hinfo,
925 &cfg, info->name, 1);
926}
927
928static int hashlimit_mt_check_v2(const struct xt_mtchk_param *par)
929{
930 struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
931 struct hashlimit_cfg3 cfg = {};
932 int ret;
933
934 ret = xt_check_proc_name(info->name, sizeof(info->name));
935 if (ret)
936 return ret;
937
938 ret = cfg_copy(&cfg, (void *)&info->cfg, 2);
939 if (ret)
940 return ret;
941
942 return hashlimit_mt_check_common(par, &info->hinfo,
943 &cfg, info->name, 2);
944}
945
946static int hashlimit_mt_check(const struct xt_mtchk_param *par)
947{
948 struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
949 int ret;
950
951 ret = xt_check_proc_name(info->name, sizeof(info->name));
952 if (ret)
953 return ret;
954
955 return hashlimit_mt_check_common(par, &info->hinfo, &info->cfg,
956 info->name, 3);
957}
958
959static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param *par)
960{
961 const struct xt_hashlimit_mtinfo2 *info = par->matchinfo;
962
963 htable_put(info->hinfo);
964}
965
966static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param *par)
967{
968 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
969
970 htable_put(info->hinfo);
971}
972
973static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
974{
975 const struct xt_hashlimit_mtinfo3 *info = par->matchinfo;
976
977 htable_put(info->hinfo);
978}
979
980static struct xt_match hashlimit_mt_reg[] __read_mostly = {
981 {
982 .name = "hashlimit",
983 .revision = 1,
984 .family = NFPROTO_IPV4,
985 .match = hashlimit_mt_v1,
986 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
987 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
988 .checkentry = hashlimit_mt_check_v1,
989 .destroy = hashlimit_mt_destroy_v1,
990 .me = THIS_MODULE,
991 },
992 {
993 .name = "hashlimit",
994 .revision = 2,
995 .family = NFPROTO_IPV4,
996 .match = hashlimit_mt_v2,
997 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
998 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
999 .checkentry = hashlimit_mt_check_v2,
1000 .destroy = hashlimit_mt_destroy_v2,
1001 .me = THIS_MODULE,
1002 },
1003 {
1004 .name = "hashlimit",
1005 .revision = 3,
1006 .family = NFPROTO_IPV4,
1007 .match = hashlimit_mt,
1008 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1009 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1010 .checkentry = hashlimit_mt_check,
1011 .destroy = hashlimit_mt_destroy,
1012 .me = THIS_MODULE,
1013 },
1014#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1015 {
1016 .name = "hashlimit",
1017 .revision = 1,
1018 .family = NFPROTO_IPV6,
1019 .match = hashlimit_mt_v1,
1020 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
1021 .usersize = offsetof(struct xt_hashlimit_mtinfo1, hinfo),
1022 .checkentry = hashlimit_mt_check_v1,
1023 .destroy = hashlimit_mt_destroy_v1,
1024 .me = THIS_MODULE,
1025 },
1026 {
1027 .name = "hashlimit",
1028 .revision = 2,
1029 .family = NFPROTO_IPV6,
1030 .match = hashlimit_mt_v2,
1031 .matchsize = sizeof(struct xt_hashlimit_mtinfo2),
1032 .usersize = offsetof(struct xt_hashlimit_mtinfo2, hinfo),
1033 .checkentry = hashlimit_mt_check_v2,
1034 .destroy = hashlimit_mt_destroy_v2,
1035 .me = THIS_MODULE,
1036 },
1037 {
1038 .name = "hashlimit",
1039 .revision = 3,
1040 .family = NFPROTO_IPV6,
1041 .match = hashlimit_mt,
1042 .matchsize = sizeof(struct xt_hashlimit_mtinfo3),
1043 .usersize = offsetof(struct xt_hashlimit_mtinfo3, hinfo),
1044 .checkentry = hashlimit_mt_check,
1045 .destroy = hashlimit_mt_destroy,
1046 .me = THIS_MODULE,
1047 },
1048#endif
1049};
1050
1051/* PROC stuff */
1052static void *dl_seq_start(struct seq_file *s, loff_t *pos)
1053 __acquires(htable->lock)
1054{
1055 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1056 unsigned int *bucket;
1057
1058 spin_lock_bh(&htable->lock);
1059 if (*pos >= htable->cfg.size)
1060 return NULL;
1061
1062 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
1063 if (!bucket)
1064 return ERR_PTR(-ENOMEM);
1065
1066 *bucket = *pos;
1067 return bucket;
1068}
1069
1070static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
1071{
1072 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1073 unsigned int *bucket = v;
1074
1075 *pos = ++(*bucket);
1076 if (*pos >= htable->cfg.size) {
1077 kfree(v);
1078 return NULL;
1079 }
1080 return bucket;
1081}
1082
1083static void dl_seq_stop(struct seq_file *s, void *v)
1084 __releases(htable->lock)
1085{
1086 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1087 unsigned int *bucket = v;
1088
1089 if (!IS_ERR(bucket))
1090 kfree(bucket);
1091 spin_unlock_bh(&htable->lock);
1092}
1093
1094static void dl_seq_print(struct dsthash_ent *ent, u_int8_t family,
1095 struct seq_file *s)
1096{
1097 switch (family) {
1098 case NFPROTO_IPV4:
1099 seq_printf(s, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1100 (long)(ent->expires - jiffies)/HZ,
1101 &ent->dst.ip.src,
1102 ntohs(ent->dst.src_port),
1103 &ent->dst.ip.dst,
1104 ntohs(ent->dst.dst_port),
1105 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1106 ent->rateinfo.cost);
1107 break;
1108#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1109 case NFPROTO_IPV6:
1110 seq_printf(s, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1111 (long)(ent->expires - jiffies)/HZ,
1112 &ent->dst.ip6.src,
1113 ntohs(ent->dst.src_port),
1114 &ent->dst.ip6.dst,
1115 ntohs(ent->dst.dst_port),
1116 ent->rateinfo.credit, ent->rateinfo.credit_cap,
1117 ent->rateinfo.cost);
1118 break;
1119#endif
1120 default:
1121 BUG();
1122 }
1123}
1124
1125static int dl_seq_real_show_v2(struct dsthash_ent *ent, u_int8_t family,
1126 struct seq_file *s)
1127{
1128 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1129
1130 spin_lock(&ent->lock);
1131 /* recalculate to show accurate numbers */
1132 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 2);
1133
1134 dl_seq_print(ent, family, s);
1135
1136 spin_unlock(&ent->lock);
1137 return seq_has_overflowed(s);
1138}
1139
1140static int dl_seq_real_show_v1(struct dsthash_ent *ent, u_int8_t family,
1141 struct seq_file *s)
1142{
1143 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1144
1145 spin_lock(&ent->lock);
1146 /* recalculate to show accurate numbers */
1147 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 1);
1148
1149 dl_seq_print(ent, family, s);
1150
1151 spin_unlock(&ent->lock);
1152 return seq_has_overflowed(s);
1153}
1154
1155static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
1156 struct seq_file *s)
1157{
1158 struct xt_hashlimit_htable *ht = PDE_DATA(file_inode(s->file));
1159
1160 spin_lock(&ent->lock);
1161 /* recalculate to show accurate numbers */
1162 rateinfo_recalc(ent, jiffies, ht->cfg.mode, 3);
1163
1164 dl_seq_print(ent, family, s);
1165
1166 spin_unlock(&ent->lock);
1167 return seq_has_overflowed(s);
1168}
1169
1170static int dl_seq_show_v2(struct seq_file *s, void *v)
1171{
1172 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1173 unsigned int *bucket = (unsigned int *)v;
1174 struct dsthash_ent *ent;
1175
1176 if (!hlist_empty(&htable->hash[*bucket])) {
1177 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1178 if (dl_seq_real_show_v2(ent, htable->family, s))
1179 return -1;
1180 }
1181 return 0;
1182}
1183
1184static int dl_seq_show_v1(struct seq_file *s, void *v)
1185{
1186 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1187 unsigned int *bucket = v;
1188 struct dsthash_ent *ent;
1189
1190 if (!hlist_empty(&htable->hash[*bucket])) {
1191 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1192 if (dl_seq_real_show_v1(ent, htable->family, s))
1193 return -1;
1194 }
1195 return 0;
1196}
1197
1198static int dl_seq_show(struct seq_file *s, void *v)
1199{
1200 struct xt_hashlimit_htable *htable = PDE_DATA(file_inode(s->file));
1201 unsigned int *bucket = v;
1202 struct dsthash_ent *ent;
1203
1204 if (!hlist_empty(&htable->hash[*bucket])) {
1205 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
1206 if (dl_seq_real_show(ent, htable->family, s))
1207 return -1;
1208 }
1209 return 0;
1210}
1211
1212static const struct seq_operations dl_seq_ops_v1 = {
1213 .start = dl_seq_start,
1214 .next = dl_seq_next,
1215 .stop = dl_seq_stop,
1216 .show = dl_seq_show_v1
1217};
1218
1219static const struct seq_operations dl_seq_ops_v2 = {
1220 .start = dl_seq_start,
1221 .next = dl_seq_next,
1222 .stop = dl_seq_stop,
1223 .show = dl_seq_show_v2
1224};
1225
1226static const struct seq_operations dl_seq_ops = {
1227 .start = dl_seq_start,
1228 .next = dl_seq_next,
1229 .stop = dl_seq_stop,
1230 .show = dl_seq_show
1231};
1232
1233static int __net_init hashlimit_proc_net_init(struct net *net)
1234{
1235 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1236
1237 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
1238 if (!hashlimit_net->ipt_hashlimit)
1239 return -ENOMEM;
1240#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1241 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
1242 if (!hashlimit_net->ip6t_hashlimit) {
1243 remove_proc_entry("ipt_hashlimit", net->proc_net);
1244 return -ENOMEM;
1245 }
1246#endif
1247 return 0;
1248}
1249
1250static void __net_exit hashlimit_proc_net_exit(struct net *net)
1251{
1252 struct xt_hashlimit_htable *hinfo;
1253 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1254
1255 /* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1256 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1257 * entries is empty before trying to remove it.
1258 */
1259 mutex_lock(&hashlimit_mutex);
1260 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
1261 htable_remove_proc_entry(hinfo);
1262 hashlimit_net->ipt_hashlimit = NULL;
1263 hashlimit_net->ip6t_hashlimit = NULL;
1264 mutex_unlock(&hashlimit_mutex);
1265
1266 remove_proc_entry("ipt_hashlimit", net->proc_net);
1267#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1268 remove_proc_entry("ip6t_hashlimit", net->proc_net);
1269#endif
1270}
1271
1272static int __net_init hashlimit_net_init(struct net *net)
1273{
1274 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
1275
1276 INIT_HLIST_HEAD(&hashlimit_net->htables);
1277 return hashlimit_proc_net_init(net);
1278}
1279
1280static void __net_exit hashlimit_net_exit(struct net *net)
1281{
1282 hashlimit_proc_net_exit(net);
1283}
1284
1285static struct pernet_operations hashlimit_net_ops = {
1286 .init = hashlimit_net_init,
1287 .exit = hashlimit_net_exit,
1288 .id = &hashlimit_net_id,
1289 .size = sizeof(struct hashlimit_net),
1290};
1291
1292static int __init hashlimit_mt_init(void)
1293{
1294 int err;
1295
1296 err = register_pernet_subsys(&hashlimit_net_ops);
1297 if (err < 0)
1298 return err;
1299 err = xt_register_matches(hashlimit_mt_reg,
1300 ARRAY_SIZE(hashlimit_mt_reg));
1301 if (err < 0)
1302 goto err1;
1303
1304 err = -ENOMEM;
1305 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1306 sizeof(struct dsthash_ent), 0, 0,
1307 NULL);
1308 if (!hashlimit_cachep) {
1309 pr_warn("unable to create slab cache\n");
1310 goto err2;
1311 }
1312 return 0;
1313
1314err2:
1315 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1316err1:
1317 unregister_pernet_subsys(&hashlimit_net_ops);
1318 return err;
1319
1320}
1321
1322static void __exit hashlimit_mt_exit(void)
1323{
1324 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1325 unregister_pernet_subsys(&hashlimit_net_ops);
1326
1327 rcu_barrier();
1328 kmem_cache_destroy(hashlimit_cachep);
1329}
1330
1331module_init(hashlimit_mt_init);
1332module_exit(hashlimit_mt_exit);
1/*
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 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;
60
61/* hash table crap */
62struct dsthash_dst {
63 union {
64 struct {
65 __be32 src;
66 __be32 dst;
67 } ip;
68#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
69 struct {
70 __be32 src[4];
71 __be32 dst[4];
72 } ip6;
73#endif
74 };
75 __be16 src_port;
76 __be16 dst_port;
77};
78
79struct dsthash_ent {
80 /* static / read-only parts in the beginning */
81 struct hlist_node node;
82 struct dsthash_dst dst;
83
84 /* modified structure members in the end */
85 spinlock_t lock;
86 unsigned long expires; /* precalculated expiry time */
87 struct {
88 unsigned long prev; /* last modification */
89 u_int32_t credit;
90 u_int32_t credit_cap, cost;
91 } rateinfo;
92 struct rcu_head rcu;
93};
94
95struct xt_hashlimit_htable {
96 struct hlist_node node; /* global list of all htables */
97 int use;
98 u_int8_t family;
99 bool rnd_initialized;
100
101 struct hashlimit_cfg1 cfg; /* config */
102
103 /* used internally */
104 spinlock_t lock; /* lock for list_head */
105 u_int32_t rnd; /* random seed for hash */
106 unsigned int count; /* number entries in table */
107 struct delayed_work gc_work;
108
109 /* seq_file stuff */
110 struct proc_dir_entry *pde;
111 const char *name;
112 struct net *net;
113
114 struct hlist_head hash[0]; /* hashtable itself */
115};
116
117static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */
118static struct kmem_cache *hashlimit_cachep __read_mostly;
119
120static inline bool dst_cmp(const struct dsthash_ent *ent,
121 const struct dsthash_dst *b)
122{
123 return !memcmp(&ent->dst, b, sizeof(ent->dst));
124}
125
126static u_int32_t
127hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
128{
129 u_int32_t hash = jhash2((const u32 *)dst,
130 sizeof(*dst)/sizeof(u32),
131 ht->rnd);
132 /*
133 * Instead of returning hash % ht->cfg.size (implying a divide)
134 * we return the high 32 bits of the (hash * ht->cfg.size) that will
135 * give results between [0 and cfg.size-1] and same hash distribution,
136 * but using a multiply, less expensive than a divide
137 */
138 return reciprocal_scale(hash, ht->cfg.size);
139}
140
141static struct dsthash_ent *
142dsthash_find(const struct xt_hashlimit_htable *ht,
143 const struct dsthash_dst *dst)
144{
145 struct dsthash_ent *ent;
146 u_int32_t hash = hash_dst(ht, dst);
147
148 if (!hlist_empty(&ht->hash[hash])) {
149 hlist_for_each_entry_rcu(ent, &ht->hash[hash], node)
150 if (dst_cmp(ent, dst)) {
151 spin_lock(&ent->lock);
152 return ent;
153 }
154 }
155 return NULL;
156}
157
158/* allocate dsthash_ent, initialize dst, put in htable and lock it */
159static struct dsthash_ent *
160dsthash_alloc_init(struct xt_hashlimit_htable *ht,
161 const struct dsthash_dst *dst, bool *race)
162{
163 struct dsthash_ent *ent;
164
165 spin_lock(&ht->lock);
166
167 /* Two or more packets may race to create the same entry in the
168 * hashtable, double check if this packet lost race.
169 */
170 ent = dsthash_find(ht, dst);
171 if (ent != NULL) {
172 spin_unlock(&ht->lock);
173 *race = true;
174 return ent;
175 }
176
177 /* initialize hash with random val at the time we allocate
178 * the first hashtable entry */
179 if (unlikely(!ht->rnd_initialized)) {
180 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
181 ht->rnd_initialized = true;
182 }
183
184 if (ht->cfg.max && ht->count >= ht->cfg.max) {
185 /* FIXME: do something. question is what.. */
186 net_err_ratelimited("max count of %u reached\n", ht->cfg.max);
187 ent = NULL;
188 } else
189 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
190 if (ent) {
191 memcpy(&ent->dst, dst, sizeof(ent->dst));
192 spin_lock_init(&ent->lock);
193
194 spin_lock(&ent->lock);
195 hlist_add_head_rcu(&ent->node, &ht->hash[hash_dst(ht, dst)]);
196 ht->count++;
197 }
198 spin_unlock(&ht->lock);
199 return ent;
200}
201
202static void dsthash_free_rcu(struct rcu_head *head)
203{
204 struct dsthash_ent *ent = container_of(head, struct dsthash_ent, rcu);
205
206 kmem_cache_free(hashlimit_cachep, ent);
207}
208
209static inline void
210dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
211{
212 hlist_del_rcu(&ent->node);
213 call_rcu_bh(&ent->rcu, dsthash_free_rcu);
214 ht->count--;
215}
216static void htable_gc(struct work_struct *work);
217
218static int htable_create(struct net *net, struct xt_hashlimit_mtinfo1 *minfo,
219 u_int8_t family)
220{
221 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
222 struct xt_hashlimit_htable *hinfo;
223 unsigned int size;
224 unsigned int i;
225
226 if (minfo->cfg.size) {
227 size = minfo->cfg.size;
228 } else {
229 size = (totalram_pages << PAGE_SHIFT) / 16384 /
230 sizeof(struct list_head);
231 if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
232 size = 8192;
233 if (size < 16)
234 size = 16;
235 }
236 /* FIXME: don't use vmalloc() here or anywhere else -HW */
237 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
238 sizeof(struct list_head) * size);
239 if (hinfo == NULL)
240 return -ENOMEM;
241 minfo->hinfo = hinfo;
242
243 /* copy match config into hashtable config */
244 memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
245 hinfo->cfg.size = size;
246 if (hinfo->cfg.max == 0)
247 hinfo->cfg.max = 8 * hinfo->cfg.size;
248 else if (hinfo->cfg.max < hinfo->cfg.size)
249 hinfo->cfg.max = hinfo->cfg.size;
250
251 for (i = 0; i < hinfo->cfg.size; i++)
252 INIT_HLIST_HEAD(&hinfo->hash[i]);
253
254 hinfo->use = 1;
255 hinfo->count = 0;
256 hinfo->family = family;
257 hinfo->rnd_initialized = false;
258 hinfo->name = kstrdup(minfo->name, GFP_KERNEL);
259 if (!hinfo->name) {
260 vfree(hinfo);
261 return -ENOMEM;
262 }
263 spin_lock_init(&hinfo->lock);
264
265 hinfo->pde = proc_create_data(minfo->name, 0,
266 (family == NFPROTO_IPV4) ?
267 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
268 &dl_file_ops, hinfo);
269 if (hinfo->pde == NULL) {
270 kfree(hinfo->name);
271 vfree(hinfo);
272 return -ENOMEM;
273 }
274 hinfo->net = net;
275
276 INIT_DEFERRABLE_WORK(&hinfo->gc_work, htable_gc);
277 queue_delayed_work(system_power_efficient_wq, &hinfo->gc_work,
278 msecs_to_jiffies(hinfo->cfg.gc_interval));
279
280 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
281
282 return 0;
283}
284
285static bool select_all(const struct xt_hashlimit_htable *ht,
286 const struct dsthash_ent *he)
287{
288 return 1;
289}
290
291static bool select_gc(const struct xt_hashlimit_htable *ht,
292 const struct dsthash_ent *he)
293{
294 return time_after_eq(jiffies, he->expires);
295}
296
297static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
298 bool (*select)(const struct xt_hashlimit_htable *ht,
299 const struct dsthash_ent *he))
300{
301 unsigned int i;
302
303 for (i = 0; i < ht->cfg.size; i++) {
304 struct dsthash_ent *dh;
305 struct hlist_node *n;
306
307 spin_lock_bh(&ht->lock);
308 hlist_for_each_entry_safe(dh, n, &ht->hash[i], node) {
309 if ((*select)(ht, dh))
310 dsthash_free(ht, dh);
311 }
312 spin_unlock_bh(&ht->lock);
313 cond_resched();
314 }
315}
316
317static void htable_gc(struct work_struct *work)
318{
319 struct xt_hashlimit_htable *ht;
320
321 ht = container_of(work, struct xt_hashlimit_htable, gc_work.work);
322
323 htable_selective_cleanup(ht, select_gc);
324
325 queue_delayed_work(system_power_efficient_wq,
326 &ht->gc_work, msecs_to_jiffies(ht->cfg.gc_interval));
327}
328
329static void htable_remove_proc_entry(struct xt_hashlimit_htable *hinfo)
330{
331 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
332 struct proc_dir_entry *parent;
333
334 if (hinfo->family == NFPROTO_IPV4)
335 parent = hashlimit_net->ipt_hashlimit;
336 else
337 parent = hashlimit_net->ip6t_hashlimit;
338
339 if (parent != NULL)
340 remove_proc_entry(hinfo->name, parent);
341}
342
343static void htable_destroy(struct xt_hashlimit_htable *hinfo)
344{
345 cancel_delayed_work_sync(&hinfo->gc_work);
346 htable_remove_proc_entry(hinfo);
347 htable_selective_cleanup(hinfo, select_all);
348 kfree(hinfo->name);
349 vfree(hinfo);
350}
351
352static struct xt_hashlimit_htable *htable_find_get(struct net *net,
353 const char *name,
354 u_int8_t family)
355{
356 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
357 struct xt_hashlimit_htable *hinfo;
358
359 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node) {
360 if (!strcmp(name, hinfo->name) &&
361 hinfo->family == family) {
362 hinfo->use++;
363 return hinfo;
364 }
365 }
366 return NULL;
367}
368
369static void htable_put(struct xt_hashlimit_htable *hinfo)
370{
371 mutex_lock(&hashlimit_mutex);
372 if (--hinfo->use == 0) {
373 hlist_del(&hinfo->node);
374 htable_destroy(hinfo);
375 }
376 mutex_unlock(&hashlimit_mutex);
377}
378
379/* The algorithm used is the Simple Token Bucket Filter (TBF)
380 * see net/sched/sch_tbf.c in the linux source tree
381 */
382
383/* Rusty: This is my (non-mathematically-inclined) understanding of
384 this algorithm. The `average rate' in jiffies becomes your initial
385 amount of credit `credit' and the most credit you can ever have
386 `credit_cap'. The `peak rate' becomes the cost of passing the
387 test, `cost'.
388
389 `prev' tracks the last packet hit: you gain one credit per jiffy.
390 If you get credit balance more than this, the extra credit is
391 discarded. Every time the match passes, you lose `cost' credits;
392 if you don't have that many, the test fails.
393
394 See Alexey's formal explanation in net/sched/sch_tbf.c.
395
396 To get the maximum range, we multiply by this factor (ie. you get N
397 credits per jiffy). We want to allow a rate as low as 1 per day
398 (slowest userspace tool allows), which means
399 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
400*/
401#define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
402
403/* Repeated shift and or gives us all 1s, final shift and add 1 gives
404 * us the power of 2 below the theoretical max, so GCC simply does a
405 * shift. */
406#define _POW2_BELOW2(x) ((x)|((x)>>1))
407#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
408#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
409#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
410#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
411#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
412
413#define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
414
415/* in byte mode, the lowest possible rate is one packet/second.
416 * credit_cap is used as a counter that tells us how many times we can
417 * refill the "credits available" counter when it becomes empty.
418 */
419#define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
420#define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
421
422static u32 xt_hashlimit_len_to_chunks(u32 len)
423{
424 return (len >> XT_HASHLIMIT_BYTE_SHIFT) + 1;
425}
426
427/* Precision saver. */
428static u32 user2credits(u32 user)
429{
430 /* If multiplying would overflow... */
431 if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
432 /* Divide first. */
433 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
434
435 return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
436}
437
438static u32 user2credits_byte(u32 user)
439{
440 u64 us = user;
441 us *= HZ * CREDITS_PER_JIFFY_BYTES;
442 return (u32) (us >> 32);
443}
444
445static void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now, u32 mode)
446{
447 unsigned long delta = now - dh->rateinfo.prev;
448 u32 cap;
449
450 if (delta == 0)
451 return;
452
453 dh->rateinfo.prev = now;
454
455 if (mode & XT_HASHLIMIT_BYTES) {
456 u32 tmp = dh->rateinfo.credit;
457 dh->rateinfo.credit += CREDITS_PER_JIFFY_BYTES * delta;
458 cap = CREDITS_PER_JIFFY_BYTES * HZ;
459 if (tmp >= dh->rateinfo.credit) {/* overflow */
460 dh->rateinfo.credit = cap;
461 return;
462 }
463 } else {
464 dh->rateinfo.credit += delta * CREDITS_PER_JIFFY;
465 cap = dh->rateinfo.credit_cap;
466 }
467 if (dh->rateinfo.credit > cap)
468 dh->rateinfo.credit = cap;
469}
470
471static void rateinfo_init(struct dsthash_ent *dh,
472 struct xt_hashlimit_htable *hinfo)
473{
474 dh->rateinfo.prev = jiffies;
475 if (hinfo->cfg.mode & XT_HASHLIMIT_BYTES) {
476 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
477 dh->rateinfo.cost = user2credits_byte(hinfo->cfg.avg);
478 dh->rateinfo.credit_cap = hinfo->cfg.burst;
479 } else {
480 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
481 hinfo->cfg.burst);
482 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
483 dh->rateinfo.credit_cap = dh->rateinfo.credit;
484 }
485}
486
487static inline __be32 maskl(__be32 a, unsigned int l)
488{
489 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
490}
491
492#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
493static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
494{
495 switch (p) {
496 case 0 ... 31:
497 i[0] = maskl(i[0], p);
498 i[1] = i[2] = i[3] = 0;
499 break;
500 case 32 ... 63:
501 i[1] = maskl(i[1], p - 32);
502 i[2] = i[3] = 0;
503 break;
504 case 64 ... 95:
505 i[2] = maskl(i[2], p - 64);
506 i[3] = 0;
507 break;
508 case 96 ... 127:
509 i[3] = maskl(i[3], p - 96);
510 break;
511 case 128:
512 break;
513 }
514}
515#endif
516
517static int
518hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
519 struct dsthash_dst *dst,
520 const struct sk_buff *skb, unsigned int protoff)
521{
522 __be16 _ports[2], *ports;
523 u8 nexthdr;
524 int poff;
525
526 memset(dst, 0, sizeof(*dst));
527
528 switch (hinfo->family) {
529 case NFPROTO_IPV4:
530 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
531 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
532 hinfo->cfg.dstmask);
533 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
534 dst->ip.src = maskl(ip_hdr(skb)->saddr,
535 hinfo->cfg.srcmask);
536
537 if (!(hinfo->cfg.mode &
538 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
539 return 0;
540 nexthdr = ip_hdr(skb)->protocol;
541 break;
542#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
543 case NFPROTO_IPV6:
544 {
545 __be16 frag_off;
546
547 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
548 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
549 sizeof(dst->ip6.dst));
550 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
551 }
552 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
553 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
554 sizeof(dst->ip6.src));
555 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
556 }
557
558 if (!(hinfo->cfg.mode &
559 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
560 return 0;
561 nexthdr = ipv6_hdr(skb)->nexthdr;
562 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, &frag_off);
563 if ((int)protoff < 0)
564 return -1;
565 break;
566 }
567#endif
568 default:
569 BUG();
570 return 0;
571 }
572
573 poff = proto_ports_offset(nexthdr);
574 if (poff >= 0) {
575 ports = skb_header_pointer(skb, protoff + poff, sizeof(_ports),
576 &_ports);
577 } else {
578 _ports[0] = _ports[1] = 0;
579 ports = _ports;
580 }
581 if (!ports)
582 return -1;
583 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
584 dst->src_port = ports[0];
585 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
586 dst->dst_port = ports[1];
587 return 0;
588}
589
590static u32 hashlimit_byte_cost(unsigned int len, struct dsthash_ent *dh)
591{
592 u64 tmp = xt_hashlimit_len_to_chunks(len);
593 tmp = tmp * dh->rateinfo.cost;
594
595 if (unlikely(tmp > CREDITS_PER_JIFFY_BYTES * HZ))
596 tmp = CREDITS_PER_JIFFY_BYTES * HZ;
597
598 if (dh->rateinfo.credit < tmp && dh->rateinfo.credit_cap) {
599 dh->rateinfo.credit_cap--;
600 dh->rateinfo.credit = CREDITS_PER_JIFFY_BYTES * HZ;
601 }
602 return (u32) tmp;
603}
604
605static bool
606hashlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
607{
608 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
609 struct xt_hashlimit_htable *hinfo = info->hinfo;
610 unsigned long now = jiffies;
611 struct dsthash_ent *dh;
612 struct dsthash_dst dst;
613 bool race = false;
614 u32 cost;
615
616 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
617 goto hotdrop;
618
619 rcu_read_lock_bh();
620 dh = dsthash_find(hinfo, &dst);
621 if (dh == NULL) {
622 dh = dsthash_alloc_init(hinfo, &dst, &race);
623 if (dh == NULL) {
624 rcu_read_unlock_bh();
625 goto hotdrop;
626 } else if (race) {
627 /* Already got an entry, update expiration timeout */
628 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
629 rateinfo_recalc(dh, now, hinfo->cfg.mode);
630 } else {
631 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
632 rateinfo_init(dh, hinfo);
633 }
634 } else {
635 /* update expiration timeout */
636 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
637 rateinfo_recalc(dh, now, hinfo->cfg.mode);
638 }
639
640 if (info->cfg.mode & XT_HASHLIMIT_BYTES)
641 cost = hashlimit_byte_cost(skb->len, dh);
642 else
643 cost = dh->rateinfo.cost;
644
645 if (dh->rateinfo.credit >= cost) {
646 /* below the limit */
647 dh->rateinfo.credit -= cost;
648 spin_unlock(&dh->lock);
649 rcu_read_unlock_bh();
650 return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
651 }
652
653 spin_unlock(&dh->lock);
654 rcu_read_unlock_bh();
655 /* default match is underlimit - so over the limit, we need to invert */
656 return info->cfg.mode & XT_HASHLIMIT_INVERT;
657
658 hotdrop:
659 par->hotdrop = true;
660 return false;
661}
662
663static int hashlimit_mt_check(const struct xt_mtchk_param *par)
664{
665 struct net *net = par->net;
666 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
667 int ret;
668
669 if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
670 return -EINVAL;
671 if (info->name[sizeof(info->name)-1] != '\0')
672 return -EINVAL;
673 if (par->family == NFPROTO_IPV4) {
674 if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
675 return -EINVAL;
676 } else {
677 if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
678 return -EINVAL;
679 }
680
681 if (info->cfg.mode & ~XT_HASHLIMIT_ALL) {
682 pr_info("Unknown mode mask %X, kernel too old?\n",
683 info->cfg.mode);
684 return -EINVAL;
685 }
686
687 /* Check for overflow. */
688 if (info->cfg.mode & XT_HASHLIMIT_BYTES) {
689 if (user2credits_byte(info->cfg.avg) == 0) {
690 pr_info("overflow, rate too high: %u\n", info->cfg.avg);
691 return -EINVAL;
692 }
693 } else if (info->cfg.burst == 0 ||
694 user2credits(info->cfg.avg * info->cfg.burst) <
695 user2credits(info->cfg.avg)) {
696 pr_info("overflow, try lower: %u/%u\n",
697 info->cfg.avg, info->cfg.burst);
698 return -ERANGE;
699 }
700
701 mutex_lock(&hashlimit_mutex);
702 info->hinfo = htable_find_get(net, info->name, par->family);
703 if (info->hinfo == NULL) {
704 ret = htable_create(net, info, par->family);
705 if (ret < 0) {
706 mutex_unlock(&hashlimit_mutex);
707 return ret;
708 }
709 }
710 mutex_unlock(&hashlimit_mutex);
711 return 0;
712}
713
714static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
715{
716 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
717
718 htable_put(info->hinfo);
719}
720
721static struct xt_match hashlimit_mt_reg[] __read_mostly = {
722 {
723 .name = "hashlimit",
724 .revision = 1,
725 .family = NFPROTO_IPV4,
726 .match = hashlimit_mt,
727 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
728 .checkentry = hashlimit_mt_check,
729 .destroy = hashlimit_mt_destroy,
730 .me = THIS_MODULE,
731 },
732#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
733 {
734 .name = "hashlimit",
735 .revision = 1,
736 .family = NFPROTO_IPV6,
737 .match = hashlimit_mt,
738 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
739 .checkentry = hashlimit_mt_check,
740 .destroy = hashlimit_mt_destroy,
741 .me = THIS_MODULE,
742 },
743#endif
744};
745
746/* PROC stuff */
747static void *dl_seq_start(struct seq_file *s, loff_t *pos)
748 __acquires(htable->lock)
749{
750 struct xt_hashlimit_htable *htable = s->private;
751 unsigned int *bucket;
752
753 spin_lock_bh(&htable->lock);
754 if (*pos >= htable->cfg.size)
755 return NULL;
756
757 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
758 if (!bucket)
759 return ERR_PTR(-ENOMEM);
760
761 *bucket = *pos;
762 return bucket;
763}
764
765static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
766{
767 struct xt_hashlimit_htable *htable = s->private;
768 unsigned int *bucket = (unsigned int *)v;
769
770 *pos = ++(*bucket);
771 if (*pos >= htable->cfg.size) {
772 kfree(v);
773 return NULL;
774 }
775 return bucket;
776}
777
778static void dl_seq_stop(struct seq_file *s, void *v)
779 __releases(htable->lock)
780{
781 struct xt_hashlimit_htable *htable = s->private;
782 unsigned int *bucket = (unsigned int *)v;
783
784 if (!IS_ERR(bucket))
785 kfree(bucket);
786 spin_unlock_bh(&htable->lock);
787}
788
789static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
790 struct seq_file *s)
791{
792 const struct xt_hashlimit_htable *ht = s->private;
793
794 spin_lock(&ent->lock);
795 /* recalculate to show accurate numbers */
796 rateinfo_recalc(ent, jiffies, ht->cfg.mode);
797
798 switch (family) {
799 case NFPROTO_IPV4:
800 seq_printf(s, "%ld %pI4:%u->%pI4:%u %u %u %u\n",
801 (long)(ent->expires - jiffies)/HZ,
802 &ent->dst.ip.src,
803 ntohs(ent->dst.src_port),
804 &ent->dst.ip.dst,
805 ntohs(ent->dst.dst_port),
806 ent->rateinfo.credit, ent->rateinfo.credit_cap,
807 ent->rateinfo.cost);
808 break;
809#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
810 case NFPROTO_IPV6:
811 seq_printf(s, "%ld %pI6:%u->%pI6:%u %u %u %u\n",
812 (long)(ent->expires - jiffies)/HZ,
813 &ent->dst.ip6.src,
814 ntohs(ent->dst.src_port),
815 &ent->dst.ip6.dst,
816 ntohs(ent->dst.dst_port),
817 ent->rateinfo.credit, ent->rateinfo.credit_cap,
818 ent->rateinfo.cost);
819 break;
820#endif
821 default:
822 BUG();
823 }
824 spin_unlock(&ent->lock);
825 return seq_has_overflowed(s);
826}
827
828static int dl_seq_show(struct seq_file *s, void *v)
829{
830 struct xt_hashlimit_htable *htable = s->private;
831 unsigned int *bucket = (unsigned int *)v;
832 struct dsthash_ent *ent;
833
834 if (!hlist_empty(&htable->hash[*bucket])) {
835 hlist_for_each_entry(ent, &htable->hash[*bucket], node)
836 if (dl_seq_real_show(ent, htable->family, s))
837 return -1;
838 }
839 return 0;
840}
841
842static const struct seq_operations dl_seq_ops = {
843 .start = dl_seq_start,
844 .next = dl_seq_next,
845 .stop = dl_seq_stop,
846 .show = dl_seq_show
847};
848
849static int dl_proc_open(struct inode *inode, struct file *file)
850{
851 int ret = seq_open(file, &dl_seq_ops);
852
853 if (!ret) {
854 struct seq_file *sf = file->private_data;
855 sf->private = PDE_DATA(inode);
856 }
857 return ret;
858}
859
860static const struct file_operations dl_file_ops = {
861 .owner = THIS_MODULE,
862 .open = dl_proc_open,
863 .read = seq_read,
864 .llseek = seq_lseek,
865 .release = seq_release
866};
867
868static int __net_init hashlimit_proc_net_init(struct net *net)
869{
870 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
871
872 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
873 if (!hashlimit_net->ipt_hashlimit)
874 return -ENOMEM;
875#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
876 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
877 if (!hashlimit_net->ip6t_hashlimit) {
878 remove_proc_entry("ipt_hashlimit", net->proc_net);
879 return -ENOMEM;
880 }
881#endif
882 return 0;
883}
884
885static void __net_exit hashlimit_proc_net_exit(struct net *net)
886{
887 struct xt_hashlimit_htable *hinfo;
888 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
889
890 /* hashlimit_net_exit() is called before hashlimit_mt_destroy().
891 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
892 * entries is empty before trying to remove it.
893 */
894 mutex_lock(&hashlimit_mutex);
895 hlist_for_each_entry(hinfo, &hashlimit_net->htables, node)
896 htable_remove_proc_entry(hinfo);
897 hashlimit_net->ipt_hashlimit = NULL;
898 hashlimit_net->ip6t_hashlimit = NULL;
899 mutex_unlock(&hashlimit_mutex);
900
901 remove_proc_entry("ipt_hashlimit", net->proc_net);
902#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
903 remove_proc_entry("ip6t_hashlimit", net->proc_net);
904#endif
905}
906
907static int __net_init hashlimit_net_init(struct net *net)
908{
909 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
910
911 INIT_HLIST_HEAD(&hashlimit_net->htables);
912 return hashlimit_proc_net_init(net);
913}
914
915static void __net_exit hashlimit_net_exit(struct net *net)
916{
917 hashlimit_proc_net_exit(net);
918}
919
920static struct pernet_operations hashlimit_net_ops = {
921 .init = hashlimit_net_init,
922 .exit = hashlimit_net_exit,
923 .id = &hashlimit_net_id,
924 .size = sizeof(struct hashlimit_net),
925};
926
927static int __init hashlimit_mt_init(void)
928{
929 int err;
930
931 err = register_pernet_subsys(&hashlimit_net_ops);
932 if (err < 0)
933 return err;
934 err = xt_register_matches(hashlimit_mt_reg,
935 ARRAY_SIZE(hashlimit_mt_reg));
936 if (err < 0)
937 goto err1;
938
939 err = -ENOMEM;
940 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
941 sizeof(struct dsthash_ent), 0, 0,
942 NULL);
943 if (!hashlimit_cachep) {
944 pr_warn("unable to create slab cache\n");
945 goto err2;
946 }
947 return 0;
948
949err2:
950 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
951err1:
952 unregister_pernet_subsys(&hashlimit_net_ops);
953 return err;
954
955}
956
957static void __exit hashlimit_mt_exit(void)
958{
959 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
960 unregister_pernet_subsys(&hashlimit_net_ops);
961
962 rcu_barrier_bh();
963 kmem_cache_destroy(hashlimit_cachep);
964}
965
966module_init(hashlimit_mt_init);
967module_exit(hashlimit_mt_exit);