tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/*
2 * xt_hashlimit - Netfilter module to limit the number of packets per time
3 * seperately for each hashbucket (sourceip/sourceport/dstip/dstport)
4 *
5 * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
6 * Copyright © CC Computer Consultants GmbH, 2007 - 2008
7 *
8 * Development of this code was funded by Astaro AG, http://www.astaro.com/
9 */
10#include <linux/module.h>
11#include <linux/spinlock.h>
12#include <linux/random.h>
13#include <linux/jhash.h>
14#include <linux/slab.h>
15#include <linux/vmalloc.h>
16#include <linux/proc_fs.h>
17#include <linux/seq_file.h>
18#include <linux/list.h>
19#include <linux/skbuff.h>
20#include <linux/mm.h>
21#include <linux/in.h>
22#include <linux/ip.h>
23#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
24#include <linux/ipv6.h>
25#include <net/ipv6.h>
26#endif
27
28#include <net/net_namespace.h>
29
30#include <linux/netfilter/x_tables.h>
31#include <linux/netfilter_ipv4/ip_tables.h>
32#include <linux/netfilter_ipv6/ip6_tables.h>
33#include <linux/netfilter/xt_hashlimit.h>
34#include <linux/mutex.h>
35
36MODULE_LICENSE("GPL");
37MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
38MODULE_AUTHOR("Jan Engelhardt <jengelh@computergmbh.de>");
39MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
40MODULE_ALIAS("ipt_hashlimit");
41MODULE_ALIAS("ip6t_hashlimit");
42
43/* need to declare this at the top */
44static struct proc_dir_entry *hashlimit_procdir4;
45static struct proc_dir_entry *hashlimit_procdir6;
46static const struct file_operations dl_file_ops;
47
48/* hash table crap */
49struct dsthash_dst {
50 union {
51 struct {
52 __be32 src;
53 __be32 dst;
54 } ip;
55#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
56 struct {
57 __be32 src[4];
58 __be32 dst[4];
59 } ip6;
60#endif
61 };
62 __be16 src_port;
63 __be16 dst_port;
64};
65
66struct dsthash_ent {
67 /* static / read-only parts in the beginning */
68 struct hlist_node node;
69 struct dsthash_dst dst;
70
71 /* modified structure members in the end */
72 unsigned long expires; /* precalculated expiry time */
73 struct {
74 unsigned long prev; /* last modification */
75 u_int32_t credit;
76 u_int32_t credit_cap, cost;
77 } rateinfo;
78};
79
80struct xt_hashlimit_htable {
81 struct hlist_node node; /* global list of all htables */
82 atomic_t use;
83 int family;
84
85 struct hashlimit_cfg1 cfg; /* config */
86
87 /* used internally */
88 spinlock_t lock; /* lock for list_head */
89 u_int32_t rnd; /* random seed for hash */
90 int rnd_initialized;
91 unsigned int count; /* number entries in table */
92 struct timer_list timer; /* timer for gc */
93
94 /* seq_file stuff */
95 struct proc_dir_entry *pde;
96
97 struct hlist_head hash[0]; /* hashtable itself */
98};
99
100static DEFINE_SPINLOCK(hashlimit_lock); /* protects htables list */
101static DEFINE_MUTEX(hlimit_mutex); /* additional checkentry protection */
102static HLIST_HEAD(hashlimit_htables);
103static struct kmem_cache *hashlimit_cachep __read_mostly;
104
105static inline bool dst_cmp(const struct dsthash_ent *ent,
106 const struct dsthash_dst *b)
107{
108 return !memcmp(&ent->dst, b, sizeof(ent->dst));
109}
110
111static u_int32_t
112hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
113{
114 u_int32_t hash = jhash2((const u32 *)dst,
115 sizeof(*dst)/sizeof(u32),
116 ht->rnd);
117 /*
118 * Instead of returning hash % ht->cfg.size (implying a divide)
119 * we return the high 32 bits of the (hash * ht->cfg.size) that will
120 * give results between [0 and cfg.size-1] and same hash distribution,
121 * but using a multiply, less expensive than a divide
122 */
123 return ((u64)hash * ht->cfg.size) >> 32;
124}
125
126static struct dsthash_ent *
127dsthash_find(const struct xt_hashlimit_htable *ht,
128 const struct dsthash_dst *dst)
129{
130 struct dsthash_ent *ent;
131 struct hlist_node *pos;
132 u_int32_t hash = hash_dst(ht, dst);
133
134 if (!hlist_empty(&ht->hash[hash])) {
135 hlist_for_each_entry(ent, pos, &ht->hash[hash], node)
136 if (dst_cmp(ent, dst))
137 return ent;
138 }
139 return NULL;
140}
141
142/* allocate dsthash_ent, initialize dst, put in htable and lock it */
143static struct dsthash_ent *
144dsthash_alloc_init(struct xt_hashlimit_htable *ht,
145 const struct dsthash_dst *dst)
146{
147 struct dsthash_ent *ent;
148
149 /* initialize hash with random val at the time we allocate
150 * the first hashtable entry */
151 if (!ht->rnd_initialized) {
152 get_random_bytes(&ht->rnd, 4);
153 ht->rnd_initialized = 1;
154 }
155
156 if (ht->cfg.max && ht->count >= ht->cfg.max) {
157 /* FIXME: do something. question is what.. */
158 if (net_ratelimit())
159 printk(KERN_WARNING
160 "xt_hashlimit: max count of %u reached\n",
161 ht->cfg.max);
162 return NULL;
163 }
164
165 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
166 if (!ent) {
167 if (net_ratelimit())
168 printk(KERN_ERR
169 "xt_hashlimit: can't allocate dsthash_ent\n");
170 return NULL;
171 }
172 memcpy(&ent->dst, dst, sizeof(ent->dst));
173
174 hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
175 ht->count++;
176 return ent;
177}
178
179static inline void
180dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
181{
182 hlist_del(&ent->node);
183 kmem_cache_free(hashlimit_cachep, ent);
184 ht->count--;
185}
186static void htable_gc(unsigned long htlong);
187
188static int htable_create_v0(struct xt_hashlimit_info *minfo, int family)
189{
190 struct xt_hashlimit_htable *hinfo;
191 unsigned int size;
192 unsigned int i;
193
194 if (minfo->cfg.size)
195 size = minfo->cfg.size;
196 else {
197 size = ((num_physpages << PAGE_SHIFT) / 16384) /
198 sizeof(struct list_head);
199 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
200 size = 8192;
201 if (size < 16)
202 size = 16;
203 }
204 /* FIXME: don't use vmalloc() here or anywhere else -HW */
205 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
206 sizeof(struct list_head) * size);
207 if (!hinfo) {
208 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
209 return -1;
210 }
211 minfo->hinfo = hinfo;
212
213 /* copy match config into hashtable config */
214 hinfo->cfg.mode = minfo->cfg.mode;
215 hinfo->cfg.avg = minfo->cfg.avg;
216 hinfo->cfg.burst = minfo->cfg.burst;
217 hinfo->cfg.max = minfo->cfg.max;
218 hinfo->cfg.gc_interval = minfo->cfg.gc_interval;
219 hinfo->cfg.expire = minfo->cfg.expire;
220
221 if (family == AF_INET)
222 hinfo->cfg.srcmask = hinfo->cfg.dstmask = 32;
223 else
224 hinfo->cfg.srcmask = hinfo->cfg.dstmask = 128;
225
226 hinfo->cfg.size = size;
227 if (!hinfo->cfg.max)
228 hinfo->cfg.max = 8 * hinfo->cfg.size;
229 else if (hinfo->cfg.max < hinfo->cfg.size)
230 hinfo->cfg.max = hinfo->cfg.size;
231
232 for (i = 0; i < hinfo->cfg.size; i++)
233 INIT_HLIST_HEAD(&hinfo->hash[i]);
234
235 atomic_set(&hinfo->use, 1);
236 hinfo->count = 0;
237 hinfo->family = family;
238 hinfo->rnd_initialized = 0;
239 spin_lock_init(&hinfo->lock);
240 hinfo->pde =
241 proc_create_data(minfo->name, 0,
242 family == AF_INET ? hashlimit_procdir4 :
243 hashlimit_procdir6,
244 &dl_file_ops, hinfo);
245 if (!hinfo->pde) {
246 vfree(hinfo);
247 return -1;
248 }
249
250 setup_timer(&hinfo->timer, htable_gc, (unsigned long )hinfo);
251 hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
252 add_timer(&hinfo->timer);
253
254 spin_lock_bh(&hashlimit_lock);
255 hlist_add_head(&hinfo->node, &hashlimit_htables);
256 spin_unlock_bh(&hashlimit_lock);
257
258 return 0;
259}
260
261static int htable_create(struct xt_hashlimit_mtinfo1 *minfo,
262 unsigned int family)
263{
264 struct xt_hashlimit_htable *hinfo;
265 unsigned int size;
266 unsigned int i;
267
268 if (minfo->cfg.size) {
269 size = minfo->cfg.size;
270 } else {
271 size = (num_physpages << PAGE_SHIFT) / 16384 /
272 sizeof(struct list_head);
273 if (num_physpages > 1024 * 1024 * 1024 / PAGE_SIZE)
274 size = 8192;
275 if (size < 16)
276 size = 16;
277 }
278 /* FIXME: don't use vmalloc() here or anywhere else -HW */
279 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
280 sizeof(struct list_head) * size);
281 if (hinfo == NULL) {
282 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
283 return -1;
284 }
285 minfo->hinfo = hinfo;
286
287 /* copy match config into hashtable config */
288 memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
289 hinfo->cfg.size = size;
290 if (hinfo->cfg.max == 0)
291 hinfo->cfg.max = 8 * hinfo->cfg.size;
292 else if (hinfo->cfg.max < hinfo->cfg.size)
293 hinfo->cfg.max = hinfo->cfg.size;
294
295 for (i = 0; i < hinfo->cfg.size; i++)
296 INIT_HLIST_HEAD(&hinfo->hash[i]);
297
298 atomic_set(&hinfo->use, 1);
299 hinfo->count = 0;
300 hinfo->family = family;
301 hinfo->rnd_initialized = 0;
302 spin_lock_init(&hinfo->lock);
303
304 hinfo->pde =
305 proc_create_data(minfo->name, 0,
306 family == AF_INET ? hashlimit_procdir4 :
307 hashlimit_procdir6,
308 &dl_file_ops, hinfo);
309 if (hinfo->pde == NULL) {
310 vfree(hinfo);
311 return -1;
312 }
313
314 setup_timer(&hinfo->timer, htable_gc, (unsigned long)hinfo);
315 hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
316 add_timer(&hinfo->timer);
317
318 spin_lock_bh(&hashlimit_lock);
319 hlist_add_head(&hinfo->node, &hashlimit_htables);
320 spin_unlock_bh(&hashlimit_lock);
321
322 return 0;
323}
324
325static bool select_all(const struct xt_hashlimit_htable *ht,
326 const struct dsthash_ent *he)
327{
328 return 1;
329}
330
331static bool select_gc(const struct xt_hashlimit_htable *ht,
332 const struct dsthash_ent *he)
333{
334 return time_after_eq(jiffies, he->expires);
335}
336
337static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
338 bool (*select)(const struct xt_hashlimit_htable *ht,
339 const struct dsthash_ent *he))
340{
341 unsigned int i;
342
343 /* lock hash table and iterate over it */
344 spin_lock_bh(&ht->lock);
345 for (i = 0; i < ht->cfg.size; i++) {
346 struct dsthash_ent *dh;
347 struct hlist_node *pos, *n;
348 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
349 if ((*select)(ht, dh))
350 dsthash_free(ht, dh);
351 }
352 }
353 spin_unlock_bh(&ht->lock);
354}
355
356/* hash table garbage collector, run by timer */
357static void htable_gc(unsigned long htlong)
358{
359 struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
360
361 htable_selective_cleanup(ht, select_gc);
362
363 /* re-add the timer accordingly */
364 ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
365 add_timer(&ht->timer);
366}
367
368static void htable_destroy(struct xt_hashlimit_htable *hinfo)
369{
370 /* remove timer, if it is pending */
371 if (timer_pending(&hinfo->timer))
372 del_timer(&hinfo->timer);
373
374 /* remove proc entry */
375 remove_proc_entry(hinfo->pde->name,
376 hinfo->family == AF_INET ? hashlimit_procdir4 :
377 hashlimit_procdir6);
378 htable_selective_cleanup(hinfo, select_all);
379 vfree(hinfo);
380}
381
382static struct xt_hashlimit_htable *htable_find_get(const char *name,
383 int family)
384{
385 struct xt_hashlimit_htable *hinfo;
386 struct hlist_node *pos;
387
388 spin_lock_bh(&hashlimit_lock);
389 hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) {
390 if (!strcmp(name, hinfo->pde->name) &&
391 hinfo->family == family) {
392 atomic_inc(&hinfo->use);
393 spin_unlock_bh(&hashlimit_lock);
394 return hinfo;
395 }
396 }
397 spin_unlock_bh(&hashlimit_lock);
398 return NULL;
399}
400
401static void htable_put(struct xt_hashlimit_htable *hinfo)
402{
403 if (atomic_dec_and_test(&hinfo->use)) {
404 spin_lock_bh(&hashlimit_lock);
405 hlist_del(&hinfo->node);
406 spin_unlock_bh(&hashlimit_lock);
407 htable_destroy(hinfo);
408 }
409}
410
411/* The algorithm used is the Simple Token Bucket Filter (TBF)
412 * see net/sched/sch_tbf.c in the linux source tree
413 */
414
415/* Rusty: This is my (non-mathematically-inclined) understanding of
416 this algorithm. The `average rate' in jiffies becomes your initial
417 amount of credit `credit' and the most credit you can ever have
418 `credit_cap'. The `peak rate' becomes the cost of passing the
419 test, `cost'.
420
421 `prev' tracks the last packet hit: you gain one credit per jiffy.
422 If you get credit balance more than this, the extra credit is
423 discarded. Every time the match passes, you lose `cost' credits;
424 if you don't have that many, the test fails.
425
426 See Alexey's formal explanation in net/sched/sch_tbf.c.
427
428 To get the maximum range, we multiply by this factor (ie. you get N
429 credits per jiffy). We want to allow a rate as low as 1 per day
430 (slowest userspace tool allows), which means
431 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
432*/
433#define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
434
435/* Repeated shift and or gives us all 1s, final shift and add 1 gives
436 * us the power of 2 below the theoretical max, so GCC simply does a
437 * shift. */
438#define _POW2_BELOW2(x) ((x)|((x)>>1))
439#define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
440#define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
441#define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
442#define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
443#define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
444
445#define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
446
447/* Precision saver. */
448static inline u_int32_t
449user2credits(u_int32_t user)
450{
451 /* If multiplying would overflow... */
452 if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
453 /* Divide first. */
454 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
455
456 return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
457}
458
459static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
460{
461 dh->rateinfo.credit += (now - dh->rateinfo.prev) * CREDITS_PER_JIFFY;
462 if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
463 dh->rateinfo.credit = dh->rateinfo.credit_cap;
464 dh->rateinfo.prev = now;
465}
466
467static inline __be32 maskl(__be32 a, unsigned int l)
468{
469 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
470}
471
472#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
473static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
474{
475 switch (p) {
476 case 0 ... 31:
477 i[0] = maskl(i[0], p);
478 i[1] = i[2] = i[3] = 0;
479 break;
480 case 32 ... 63:
481 i[1] = maskl(i[1], p - 32);
482 i[2] = i[3] = 0;
483 break;
484 case 64 ... 95:
485 i[2] = maskl(i[2], p - 64);
486 i[3] = 0;
487 case 96 ... 127:
488 i[3] = maskl(i[3], p - 96);
489 break;
490 case 128:
491 break;
492 }
493}
494#endif
495
496static int
497hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
498 struct dsthash_dst *dst,
499 const struct sk_buff *skb, unsigned int protoff)
500{
501 __be16 _ports[2], *ports;
502 u8 nexthdr;
503
504 memset(dst, 0, sizeof(*dst));
505
506 switch (hinfo->family) {
507 case AF_INET:
508 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
509 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
510 hinfo->cfg.dstmask);
511 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
512 dst->ip.src = maskl(ip_hdr(skb)->saddr,
513 hinfo->cfg.srcmask);
514
515 if (!(hinfo->cfg.mode &
516 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
517 return 0;
518 nexthdr = ip_hdr(skb)->protocol;
519 break;
520#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
521 case AF_INET6:
522 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
523 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
524 sizeof(dst->ip6.dst));
525 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
526 }
527 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
528 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
529 sizeof(dst->ip6.src));
530 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
531 }
532
533 if (!(hinfo->cfg.mode &
534 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
535 return 0;
536 nexthdr = ipv6_hdr(skb)->nexthdr;
537 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
538 if ((int)protoff < 0)
539 return -1;
540 break;
541#endif
542 default:
543 BUG();
544 return 0;
545 }
546
547 switch (nexthdr) {
548 case IPPROTO_TCP:
549 case IPPROTO_UDP:
550 case IPPROTO_UDPLITE:
551 case IPPROTO_SCTP:
552 case IPPROTO_DCCP:
553 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
554 &_ports);
555 break;
556 default:
557 _ports[0] = _ports[1] = 0;
558 ports = _ports;
559 break;
560 }
561 if (!ports)
562 return -1;
563 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
564 dst->src_port = ports[0];
565 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
566 dst->dst_port = ports[1];
567 return 0;
568}
569
570static bool
571hashlimit_mt_v0(const struct sk_buff *skb, const struct net_device *in,
572 const struct net_device *out, const struct xt_match *match,
573 const void *matchinfo, int offset, unsigned int protoff,
574 bool *hotdrop)
575{
576 const struct xt_hashlimit_info *r =
577 ((const struct xt_hashlimit_info *)matchinfo)->u.master;
578 struct xt_hashlimit_htable *hinfo = r->hinfo;
579 unsigned long now = jiffies;
580 struct dsthash_ent *dh;
581 struct dsthash_dst dst;
582
583 if (hashlimit_init_dst(hinfo, &dst, skb, protoff) < 0)
584 goto hotdrop;
585
586 spin_lock_bh(&hinfo->lock);
587 dh = dsthash_find(hinfo, &dst);
588 if (!dh) {
589 dh = dsthash_alloc_init(hinfo, &dst);
590 if (!dh) {
591 spin_unlock_bh(&hinfo->lock);
592 goto hotdrop;
593 }
594
595 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
596 dh->rateinfo.prev = jiffies;
597 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
598 hinfo->cfg.burst);
599 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
600 hinfo->cfg.burst);
601 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
602 } else {
603 /* update expiration timeout */
604 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
605 rateinfo_recalc(dh, now);
606 }
607
608 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
609 /* We're underlimit. */
610 dh->rateinfo.credit -= dh->rateinfo.cost;
611 spin_unlock_bh(&hinfo->lock);
612 return true;
613 }
614
615 spin_unlock_bh(&hinfo->lock);
616
617 /* default case: we're overlimit, thus don't match */
618 return false;
619
620hotdrop:
621 *hotdrop = true;
622 return false;
623}
624
625static bool
626hashlimit_mt(const struct sk_buff *skb, const struct net_device *in,
627 const struct net_device *out, const struct xt_match *match,
628 const void *matchinfo, int offset, unsigned int protoff,
629 bool *hotdrop)
630{
631 const struct xt_hashlimit_mtinfo1 *info = matchinfo;
632 struct xt_hashlimit_htable *hinfo = info->hinfo;
633 unsigned long now = jiffies;
634 struct dsthash_ent *dh;
635 struct dsthash_dst dst;
636
637 if (hashlimit_init_dst(hinfo, &dst, skb, protoff) < 0)
638 goto hotdrop;
639
640 spin_lock_bh(&hinfo->lock);
641 dh = dsthash_find(hinfo, &dst);
642 if (dh == NULL) {
643 dh = dsthash_alloc_init(hinfo, &dst);
644 if (dh == NULL) {
645 spin_unlock_bh(&hinfo->lock);
646 goto hotdrop;
647 }
648
649 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
650 dh->rateinfo.prev = jiffies;
651 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
652 hinfo->cfg.burst);
653 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
654 hinfo->cfg.burst);
655 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
656 } else {
657 /* update expiration timeout */
658 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
659 rateinfo_recalc(dh, now);
660 }
661
662 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
663 /* below the limit */
664 dh->rateinfo.credit -= dh->rateinfo.cost;
665 spin_unlock_bh(&hinfo->lock);
666 return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
667 }
668
669 spin_unlock_bh(&hinfo->lock);
670 /* default match is underlimit - so over the limit, we need to invert */
671 return info->cfg.mode & XT_HASHLIMIT_INVERT;
672
673 hotdrop:
674 *hotdrop = true;
675 return false;
676}
677
678static bool
679hashlimit_mt_check_v0(const char *tablename, const void *inf,
680 const struct xt_match *match, void *matchinfo,
681 unsigned int hook_mask)
682{
683 struct xt_hashlimit_info *r = matchinfo;
684
685 /* Check for overflow. */
686 if (r->cfg.burst == 0 ||
687 user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
688 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
689 r->cfg.avg, r->cfg.burst);
690 return false;
691 }
692 if (r->cfg.mode == 0 ||
693 r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
694 XT_HASHLIMIT_HASH_DIP |
695 XT_HASHLIMIT_HASH_SIP |
696 XT_HASHLIMIT_HASH_SPT))
697 return false;
698 if (!r->cfg.gc_interval)
699 return false;
700 if (!r->cfg.expire)
701 return false;
702 if (r->name[sizeof(r->name) - 1] != '\0')
703 return false;
704
705 /* This is the best we've got: We cannot release and re-grab lock,
706 * since checkentry() is called before x_tables.c grabs xt_mutex.
707 * We also cannot grab the hashtable spinlock, since htable_create will
708 * call vmalloc, and that can sleep. And we cannot just re-search
709 * the list of htable's in htable_create(), since then we would
710 * create duplicate proc files. -HW */
711 mutex_lock(&hlimit_mutex);
712 r->hinfo = htable_find_get(r->name, match->family);
713 if (!r->hinfo && htable_create_v0(r, match->family) != 0) {
714 mutex_unlock(&hlimit_mutex);
715 return false;
716 }
717 mutex_unlock(&hlimit_mutex);
718
719 /* Ugly hack: For SMP, we only want to use one set */
720 r->u.master = r;
721 return true;
722}
723
724static bool
725hashlimit_mt_check(const char *tablename, const void *inf,
726 const struct xt_match *match, void *matchinfo,
727 unsigned int hook_mask)
728{
729 struct xt_hashlimit_mtinfo1 *info = matchinfo;
730
731 /* Check for overflow. */
732 if (info->cfg.burst == 0 ||
733 user2credits(info->cfg.avg * info->cfg.burst) <
734 user2credits(info->cfg.avg)) {
735 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
736 info->cfg.avg, info->cfg.burst);
737 return false;
738 }
739 if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
740 return false;
741 if (info->name[sizeof(info->name)-1] != '\0')
742 return false;
743 if (match->family == AF_INET) {
744 if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
745 return false;
746 } else {
747 if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
748 return false;
749 }
750
751 /* This is the best we've got: We cannot release and re-grab lock,
752 * since checkentry() is called before x_tables.c grabs xt_mutex.
753 * We also cannot grab the hashtable spinlock, since htable_create will
754 * call vmalloc, and that can sleep. And we cannot just re-search
755 * the list of htable's in htable_create(), since then we would
756 * create duplicate proc files. -HW */
757 mutex_lock(&hlimit_mutex);
758 info->hinfo = htable_find_get(info->name, match->family);
759 if (!info->hinfo && htable_create(info, match->family) != 0) {
760 mutex_unlock(&hlimit_mutex);
761 return false;
762 }
763 mutex_unlock(&hlimit_mutex);
764 return true;
765}
766
767static void
768hashlimit_mt_destroy_v0(const struct xt_match *match, void *matchinfo)
769{
770 const struct xt_hashlimit_info *r = matchinfo;
771
772 htable_put(r->hinfo);
773}
774
775static void
776hashlimit_mt_destroy(const struct xt_match *match, void *matchinfo)
777{
778 const struct xt_hashlimit_mtinfo1 *info = matchinfo;
779
780 htable_put(info->hinfo);
781}
782
783#ifdef CONFIG_COMPAT
784struct compat_xt_hashlimit_info {
785 char name[IFNAMSIZ];
786 struct hashlimit_cfg cfg;
787 compat_uptr_t hinfo;
788 compat_uptr_t master;
789};
790
791static void hashlimit_mt_compat_from_user(void *dst, void *src)
792{
793 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
794
795 memcpy(dst, src, off);
796 memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
797}
798
799static int hashlimit_mt_compat_to_user(void __user *dst, void *src)
800{
801 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
802
803 return copy_to_user(dst, src, off) ? -EFAULT : 0;
804}
805#endif
806
807static struct xt_match hashlimit_mt_reg[] __read_mostly = {
808 {
809 .name = "hashlimit",
810 .revision = 0,
811 .family = AF_INET,
812 .match = hashlimit_mt_v0,
813 .matchsize = sizeof(struct xt_hashlimit_info),
814#ifdef CONFIG_COMPAT
815 .compatsize = sizeof(struct compat_xt_hashlimit_info),
816 .compat_from_user = hashlimit_mt_compat_from_user,
817 .compat_to_user = hashlimit_mt_compat_to_user,
818#endif
819 .checkentry = hashlimit_mt_check_v0,
820 .destroy = hashlimit_mt_destroy_v0,
821 .me = THIS_MODULE
822 },
823 {
824 .name = "hashlimit",
825 .revision = 1,
826 .family = AF_INET,
827 .match = hashlimit_mt,
828 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
829 .checkentry = hashlimit_mt_check,
830 .destroy = hashlimit_mt_destroy,
831 .me = THIS_MODULE,
832 },
833#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
834 {
835 .name = "hashlimit",
836 .family = AF_INET6,
837 .match = hashlimit_mt_v0,
838 .matchsize = sizeof(struct xt_hashlimit_info),
839#ifdef CONFIG_COMPAT
840 .compatsize = sizeof(struct compat_xt_hashlimit_info),
841 .compat_from_user = hashlimit_mt_compat_from_user,
842 .compat_to_user = hashlimit_mt_compat_to_user,
843#endif
844 .checkentry = hashlimit_mt_check_v0,
845 .destroy = hashlimit_mt_destroy_v0,
846 .me = THIS_MODULE
847 },
848 {
849 .name = "hashlimit",
850 .revision = 1,
851 .family = AF_INET6,
852 .match = hashlimit_mt,
853 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
854 .checkentry = hashlimit_mt_check,
855 .destroy = hashlimit_mt_destroy,
856 .me = THIS_MODULE,
857 },
858#endif
859};
860
861/* PROC stuff */
862static void *dl_seq_start(struct seq_file *s, loff_t *pos)
863 __acquires(htable->lock)
864{
865 struct proc_dir_entry *pde = s->private;
866 struct xt_hashlimit_htable *htable = pde->data;
867 unsigned int *bucket;
868
869 spin_lock_bh(&htable->lock);
870 if (*pos >= htable->cfg.size)
871 return NULL;
872
873 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
874 if (!bucket)
875 return ERR_PTR(-ENOMEM);
876
877 *bucket = *pos;
878 return bucket;
879}
880
881static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
882{
883 struct proc_dir_entry *pde = s->private;
884 struct xt_hashlimit_htable *htable = pde->data;
885 unsigned int *bucket = (unsigned int *)v;
886
887 *pos = ++(*bucket);
888 if (*pos >= htable->cfg.size) {
889 kfree(v);
890 return NULL;
891 }
892 return bucket;
893}
894
895static void dl_seq_stop(struct seq_file *s, void *v)
896 __releases(htable->lock)
897{
898 struct proc_dir_entry *pde = s->private;
899 struct xt_hashlimit_htable *htable = pde->data;
900 unsigned int *bucket = (unsigned int *)v;
901
902 kfree(bucket);
903 spin_unlock_bh(&htable->lock);
904}
905
906static int dl_seq_real_show(struct dsthash_ent *ent, int family,
907 struct seq_file *s)
908{
909 /* recalculate to show accurate numbers */
910 rateinfo_recalc(ent, jiffies);
911
912 switch (family) {
913 case AF_INET:
914 return seq_printf(s, "%ld %u.%u.%u.%u:%u->"
915 "%u.%u.%u.%u:%u %u %u %u\n",
916 (long)(ent->expires - jiffies)/HZ,
917 NIPQUAD(ent->dst.ip.src),
918 ntohs(ent->dst.src_port),
919 NIPQUAD(ent->dst.ip.dst),
920 ntohs(ent->dst.dst_port),
921 ent->rateinfo.credit, ent->rateinfo.credit_cap,
922 ent->rateinfo.cost);
923#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
924 case AF_INET6:
925 return seq_printf(s, "%ld " NIP6_FMT ":%u->"
926 NIP6_FMT ":%u %u %u %u\n",
927 (long)(ent->expires - jiffies)/HZ,
928 NIP6(*(struct in6_addr *)&ent->dst.ip6.src),
929 ntohs(ent->dst.src_port),
930 NIP6(*(struct in6_addr *)&ent->dst.ip6.dst),
931 ntohs(ent->dst.dst_port),
932 ent->rateinfo.credit, ent->rateinfo.credit_cap,
933 ent->rateinfo.cost);
934#endif
935 default:
936 BUG();
937 return 0;
938 }
939}
940
941static int dl_seq_show(struct seq_file *s, void *v)
942{
943 struct proc_dir_entry *pde = s->private;
944 struct xt_hashlimit_htable *htable = pde->data;
945 unsigned int *bucket = (unsigned int *)v;
946 struct dsthash_ent *ent;
947 struct hlist_node *pos;
948
949 if (!hlist_empty(&htable->hash[*bucket])) {
950 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
951 if (dl_seq_real_show(ent, htable->family, s))
952 return 1;
953 }
954 return 0;
955}
956
957static const struct seq_operations dl_seq_ops = {
958 .start = dl_seq_start,
959 .next = dl_seq_next,
960 .stop = dl_seq_stop,
961 .show = dl_seq_show
962};
963
964static int dl_proc_open(struct inode *inode, struct file *file)
965{
966 int ret = seq_open(file, &dl_seq_ops);
967
968 if (!ret) {
969 struct seq_file *sf = file->private_data;
970 sf->private = PDE(inode);
971 }
972 return ret;
973}
974
975static const struct file_operations dl_file_ops = {
976 .owner = THIS_MODULE,
977 .open = dl_proc_open,
978 .read = seq_read,
979 .llseek = seq_lseek,
980 .release = seq_release
981};
982
983static int __init hashlimit_mt_init(void)
984{
985 int err;
986
987 err = xt_register_matches(hashlimit_mt_reg,
988 ARRAY_SIZE(hashlimit_mt_reg));
989 if (err < 0)
990 goto err1;
991
992 err = -ENOMEM;
993 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
994 sizeof(struct dsthash_ent), 0, 0,
995 NULL);
996 if (!hashlimit_cachep) {
997 printk(KERN_ERR "xt_hashlimit: unable to create slab cache\n");
998 goto err2;
999 }
1000 hashlimit_procdir4 = proc_mkdir("ipt_hashlimit", init_net.proc_net);
1001 if (!hashlimit_procdir4) {
1002 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
1003 "entry\n");
1004 goto err3;
1005 }
1006 err = 0;
1007#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
1008 hashlimit_procdir6 = proc_mkdir("ip6t_hashlimit", init_net.proc_net);
1009 if (!hashlimit_procdir6) {
1010 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
1011 "entry\n");
1012 err = -ENOMEM;
1013 }
1014#endif
1015 if (!err)
1016 return 0;
1017 remove_proc_entry("ipt_hashlimit", init_net.proc_net);
1018err3:
1019 kmem_cache_destroy(hashlimit_cachep);
1020err2:
1021 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1022err1:
1023 return err;
1024
1025}
1026
1027static void __exit hashlimit_mt_exit(void)
1028{
1029 remove_proc_entry("ipt_hashlimit", init_net.proc_net);
1030#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
1031 remove_proc_entry("ip6t_hashlimit", init_net.proc_net);
1032#endif
1033 kmem_cache_destroy(hashlimit_cachep);
1034 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1035}
1036
1037module_init(hashlimit_mt_init);
1038module_exit(hashlimit_mt_exit);