?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
netfilter: xtables: slightly better error reporting
[pohmelfs.git] / net / netfilter / xt_hashlimit.c
blobc89fde7d12343f68a8a2c4d9d6092340508808cc
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 * Copyright © CC Computer Consultants GmbH, 2007 - 2008
7 *
8 * Development of this code was funded by Astaro AG, http://www.astaro.com/
9 */
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/module.h>
12 #include <linux/spinlock.h>
13 #include <linux/random.h>
14 #include <linux/jhash.h>
15 #include <linux/slab.h>
16 #include <linux/vmalloc.h>
17 #include <linux/proc_fs.h>
18 #include <linux/seq_file.h>
19 #include <linux/list.h>
20 #include <linux/skbuff.h>
21 #include <linux/mm.h>
22 #include <linux/in.h>
23 #include <linux/ip.h>
24 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
25 #include <linux/ipv6.h>
26 #include <net/ipv6.h>
27 #endif
28
29 #include <net/net_namespace.h>
30 #include <net/netns/generic.h>
31
32 #include <linux/netfilter/x_tables.h>
33 #include <linux/netfilter_ipv4/ip_tables.h>
34 #include <linux/netfilter_ipv6/ip6_tables.h>
35 #include <linux/netfilter/xt_hashlimit.h>
36 #include <linux/mutex.h>
37
38 MODULE_LICENSE("GPL");
39 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
40 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
41 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
42 MODULE_ALIAS("ipt_hashlimit");
43 MODULE_ALIAS("ip6t_hashlimit");
44
45 struct hashlimit_net {
46 struct hlist_head htables;
47 struct proc_dir_entry *ipt_hashlimit;
48 struct proc_dir_entry *ip6t_hashlimit;
49 };
50
51 static int hashlimit_net_id;
52 static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
53 {
54 return net_generic(net, hashlimit_net_id);
55 }
56
57 /* need to declare this at the top */
58 static const struct file_operations dl_file_ops;
59
60 /* hash table crap */
61 struct dsthash_dst {
62 union {
63 struct {
64 __be32 src;
65 __be32 dst;
66 } ip;
67 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
68 struct {
69 __be32 src[4];
70 __be32 dst[4];
71 } ip6;
72 #endif
73 };
74 __be16 src_port;
75 __be16 dst_port;
76 };
77
78 struct dsthash_ent {
79 /* static / read-only parts in the beginning */
80 struct hlist_node node;
81 struct dsthash_dst dst;
82
83 /* modified structure members in the end */
84 unsigned long expires; /* precalculated expiry time */
85 struct {
86 unsigned long prev; /* last modification */
87 u_int32_t credit;
88 u_int32_t credit_cap, cost;
89 } rateinfo;
90 };
91
92 struct xt_hashlimit_htable {
93 struct hlist_node node; /* global list of all htables */
94 int use;
95 u_int8_t family;
96 bool rnd_initialized;
97
98 struct hashlimit_cfg1 cfg; /* config */
99
100 /* used internally */
101 spinlock_t lock; /* lock for list_head */
102 u_int32_t rnd; /* random seed for hash */
103 unsigned int count; /* number entries in table */
104 struct timer_list timer; /* timer for gc */
105
106 /* seq_file stuff */
107 struct proc_dir_entry *pde;
108 struct net *net;
109
110 struct hlist_head hash[0]; /* hashtable itself */
111 };
112
113 static DEFINE_MUTEX(hashlimit_mutex); /* protects htables list */
114 static struct kmem_cache *hashlimit_cachep __read_mostly;
115
116 static inline bool dst_cmp(const struct dsthash_ent *ent,
117 const struct dsthash_dst *b)
118 {
119 return !memcmp(&ent->dst, b, sizeof(ent->dst));
120 }
121
122 static u_int32_t
123 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
124 {
125 u_int32_t hash = jhash2((const u32 *)dst,
126 sizeof(*dst)/sizeof(u32),
127 ht->rnd);
128 /*
129 * Instead of returning hash % ht->cfg.size (implying a divide)
130 * we return the high 32 bits of the (hash * ht->cfg.size) that will
131 * give results between [0 and cfg.size-1] and same hash distribution,
132 * but using a multiply, less expensive than a divide
133 */
134 return ((u64)hash * ht->cfg.size) >> 32;
135 }
136
137 static struct dsthash_ent *
138 dsthash_find(const struct xt_hashlimit_htable *ht,
139 const struct dsthash_dst *dst)
140 {
141 struct dsthash_ent *ent;
142 struct hlist_node *pos;
143 u_int32_t hash = hash_dst(ht, dst);
144
145 if (!hlist_empty(&ht->hash[hash])) {
146 hlist_for_each_entry(ent, pos, &ht->hash[hash], node)
147 if (dst_cmp(ent, dst))
148 return ent;
149 }
150 return NULL;
151 }
152
153 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
154 static struct dsthash_ent *
155 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
156 const struct dsthash_dst *dst)
157 {
158 struct dsthash_ent *ent;
159
160 /* initialize hash with random val at the time we allocate
161 * the first hashtable entry */
162 if (!ht->rnd_initialized) {
163 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
164 ht->rnd_initialized = true;
165 }
166
167 if (ht->cfg.max && ht->count >= ht->cfg.max) {
168 /* FIXME: do something. question is what.. */
169 if (net_ratelimit())
170 pr_err("max count of %u reached\n", ht->cfg.max);
171 return NULL;
172 }
173
174 ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
175 if (!ent) {
176 if (net_ratelimit())
177 pr_err("cannot allocate dsthash_ent\n");
178 return NULL;
179 }
180 memcpy(&ent->dst, dst, sizeof(ent->dst));
181
182 hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
183 ht->count++;
184 return ent;
185 }
186
187 static inline void
188 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
189 {
190 hlist_del(&ent->node);
191 kmem_cache_free(hashlimit_cachep, ent);
192 ht->count--;
193 }
194 static void htable_gc(unsigned long htlong);
195
196 static int htable_create_v0(struct net *net, struct xt_hashlimit_info *minfo, u_int8_t family)
197 {
198 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
199 struct xt_hashlimit_htable *hinfo;
200 unsigned int size;
201 unsigned int i;
202
203 if (minfo->cfg.size)
204 size = minfo->cfg.size;
205 else {
206 size = ((totalram_pages << PAGE_SHIFT) / 16384) /
207 sizeof(struct list_head);
208 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
209 size = 8192;
210 if (size < 16)
211 size = 16;
212 }
213 /* FIXME: don't use vmalloc() here or anywhere else -HW */
214 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
215 sizeof(struct list_head) * size);
216 if (!hinfo)
217 return -ENOMEM;
218 minfo->hinfo = hinfo;
219
220 /* copy match config into hashtable config */
221 hinfo->cfg.mode = minfo->cfg.mode;
222 hinfo->cfg.avg = minfo->cfg.avg;
223 hinfo->cfg.burst = minfo->cfg.burst;
224 hinfo->cfg.max = minfo->cfg.max;
225 hinfo->cfg.gc_interval = minfo->cfg.gc_interval;
226 hinfo->cfg.expire = minfo->cfg.expire;
227
228 if (family == NFPROTO_IPV4)
229 hinfo->cfg.srcmask = hinfo->cfg.dstmask = 32;
230 else
231 hinfo->cfg.srcmask = hinfo->cfg.dstmask = 128;
232
233 hinfo->cfg.size = size;
234 if (!hinfo->cfg.max)
235 hinfo->cfg.max = 8 * hinfo->cfg.size;
236 else if (hinfo->cfg.max < hinfo->cfg.size)
237 hinfo->cfg.max = hinfo->cfg.size;
238
239 for (i = 0; i < hinfo->cfg.size; i++)
240 INIT_HLIST_HEAD(&hinfo->hash[i]);
241
242 hinfo->use = 1;
243 hinfo->count = 0;
244 hinfo->family = family;
245 hinfo->rnd_initialized = false;
246 spin_lock_init(&hinfo->lock);
247 hinfo->pde = proc_create_data(minfo->name, 0,
248 (family == NFPROTO_IPV4) ?
249 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
250 &dl_file_ops, hinfo);
251 if (!hinfo->pde) {
252 vfree(hinfo);
253 return -ENOMEM;
254 }
255 hinfo->net = net;
256
257 setup_timer(&hinfo->timer, htable_gc, (unsigned long )hinfo);
258 hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
259 add_timer(&hinfo->timer);
260
261 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
262
263 return 0;
264 }
265
266 static int htable_create(struct net *net, struct xt_hashlimit_mtinfo1 *minfo,
267 u_int8_t family)
268 {
269 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
270 struct xt_hashlimit_htable *hinfo;
271 unsigned int size;
272 unsigned int i;
273
274 if (minfo->cfg.size) {
275 size = minfo->cfg.size;
276 } else {
277 size = (totalram_pages << PAGE_SHIFT) / 16384 /
278 sizeof(struct list_head);
279 if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
280 size = 8192;
281 if (size < 16)
282 size = 16;
283 }
284 /* FIXME: don't use vmalloc() here or anywhere else -HW */
285 hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
286 sizeof(struct list_head) * size);
287 if (hinfo == NULL)
288 return -ENOMEM;
289 minfo->hinfo = hinfo;
290
291 /* copy match config into hashtable config */
292 memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
293 hinfo->cfg.size = size;
294 if (hinfo->cfg.max == 0)
295 hinfo->cfg.max = 8 * hinfo->cfg.size;
296 else if (hinfo->cfg.max < hinfo->cfg.size)
297 hinfo->cfg.max = hinfo->cfg.size;
298
299 for (i = 0; i < hinfo->cfg.size; i++)
300 INIT_HLIST_HEAD(&hinfo->hash[i]);
301
302 hinfo->use = 1;
303 hinfo->count = 0;
304 hinfo->family = family;
305 hinfo->rnd_initialized = false;
306 spin_lock_init(&hinfo->lock);
307
308 hinfo->pde = proc_create_data(minfo->name, 0,
309 (family == NFPROTO_IPV4) ?
310 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
311 &dl_file_ops, hinfo);
312 if (hinfo->pde == NULL) {
313 vfree(hinfo);
314 return -ENOMEM;
315 }
316 hinfo->net = net;
317
318 setup_timer(&hinfo->timer, htable_gc, (unsigned long)hinfo);
319 hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
320 add_timer(&hinfo->timer);
321
322 hlist_add_head(&hinfo->node, &hashlimit_net->htables);
323
324 return 0;
325 }
326
327 static bool select_all(const struct xt_hashlimit_htable *ht,
328 const struct dsthash_ent *he)
329 {
330 return 1;
331 }
332
333 static bool select_gc(const struct xt_hashlimit_htable *ht,
334 const struct dsthash_ent *he)
335 {
336 return time_after_eq(jiffies, he->expires);
337 }
338
339 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
340 bool (*select)(const struct xt_hashlimit_htable *ht,
341 const struct dsthash_ent *he))
342 {
343 unsigned int i;
344
345 /* lock hash table and iterate over it */
346 spin_lock_bh(&ht->lock);
347 for (i = 0; i < ht->cfg.size; i++) {
348 struct dsthash_ent *dh;
349 struct hlist_node *pos, *n;
350 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
351 if ((*select)(ht, dh))
352 dsthash_free(ht, dh);
353 }
354 }
355 spin_unlock_bh(&ht->lock);
356 }
357
358 /* hash table garbage collector, run by timer */
359 static void htable_gc(unsigned long htlong)
360 {
361 struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
362
363 htable_selective_cleanup(ht, select_gc);
364
365 /* re-add the timer accordingly */
366 ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
367 add_timer(&ht->timer);
368 }
369
370 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
371 {
372 struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
373 struct proc_dir_entry *parent;
374
375 del_timer_sync(&hinfo->timer);
376
377 if (hinfo->family == NFPROTO_IPV4)
378 parent = hashlimit_net->ipt_hashlimit;
379 else
380 parent = hashlimit_net->ip6t_hashlimit;
381 remove_proc_entry(hinfo->pde->name, parent);
382 htable_selective_cleanup(hinfo, select_all);
383 vfree(hinfo);
384 }
385
386 static struct xt_hashlimit_htable *htable_find_get(struct net *net,
387 const char *name,
388 u_int8_t family)
389 {
390 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
391 struct xt_hashlimit_htable *hinfo;
392 struct hlist_node *pos;
393
394 hlist_for_each_entry(hinfo, pos, &hashlimit_net->htables, node) {
395 if (!strcmp(name, hinfo->pde->name) &&
396 hinfo->family == family) {
397 hinfo->use++;
398 return hinfo;
399 }
400 }
401 return NULL;
402 }
403
404 static void htable_put(struct xt_hashlimit_htable *hinfo)
405 {
406 mutex_lock(&hashlimit_mutex);
407 if (--hinfo->use == 0) {
408 hlist_del(&hinfo->node);
409 htable_destroy(hinfo);
410 }
411 mutex_unlock(&hashlimit_mutex);
412 }
413
414 /* The algorithm used is the Simple Token Bucket Filter (TBF)
415 * see net/sched/sch_tbf.c in the linux source tree
416 */
417
418 /* Rusty: This is my (non-mathematically-inclined) understanding of
419 this algorithm. The `average rate' in jiffies becomes your initial
420 amount of credit `credit' and the most credit you can ever have
421 `credit_cap'. The `peak rate' becomes the cost of passing the
422 test, `cost'.
423
424 `prev' tracks the last packet hit: you gain one credit per jiffy.
425 If you get credit balance more than this, the extra credit is
426 discarded. Every time the match passes, you lose `cost' credits;
427 if you don't have that many, the test fails.
428
429 See Alexey's formal explanation in net/sched/sch_tbf.c.
430
431 To get the maximum range, we multiply by this factor (ie. you get N
432 credits per jiffy). We want to allow a rate as low as 1 per day
433 (slowest userspace tool allows), which means
434 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
435 */
436 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
437
438 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
439 * us the power of 2 below the theoretical max, so GCC simply does a
440 * shift. */
441 #define _POW2_BELOW2(x) ((x)|((x)>>1))
442 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
443 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
444 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
445 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
446 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
447
448 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
449
450 /* Precision saver. */
451 static inline u_int32_t
452 user2credits(u_int32_t user)
453 {
454 /* If multiplying would overflow... */
455 if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
456 /* Divide first. */
457 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
458
459 return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
460 }
461
462 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
463 {
464 dh->rateinfo.credit += (now - dh->rateinfo.prev) * CREDITS_PER_JIFFY;
465 if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
466 dh->rateinfo.credit = dh->rateinfo.credit_cap;
467 dh->rateinfo.prev = now;
468 }
469
470 static inline __be32 maskl(__be32 a, unsigned int l)
471 {
472 return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
473 }
474
475 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
476 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
477 {
478 switch (p) {
479 case 0 ... 31:
480 i[0] = maskl(i[0], p);
481 i[1] = i[2] = i[3] = 0;
482 break;
483 case 32 ... 63:
484 i[1] = maskl(i[1], p - 32);
485 i[2] = i[3] = 0;
486 break;
487 case 64 ... 95:
488 i[2] = maskl(i[2], p - 64);
489 i[3] = 0;
490 case 96 ... 127:
491 i[3] = maskl(i[3], p - 96);
492 break;
493 case 128:
494 break;
495 }
496 }
497 #endif
498
499 static int
500 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
501 struct dsthash_dst *dst,
502 const struct sk_buff *skb, unsigned int protoff)
503 {
504 __be16 _ports[2], *ports;
505 u8 nexthdr;
506
507 memset(dst, 0, sizeof(*dst));
508
509 switch (hinfo->family) {
510 case NFPROTO_IPV4:
511 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
512 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
513 hinfo->cfg.dstmask);
514 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
515 dst->ip.src = maskl(ip_hdr(skb)->saddr,
516 hinfo->cfg.srcmask);
517
518 if (!(hinfo->cfg.mode &
519 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
520 return 0;
521 nexthdr = ip_hdr(skb)->protocol;
522 break;
523 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
524 case NFPROTO_IPV6:
525 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
526 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
527 sizeof(dst->ip6.dst));
528 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
529 }
530 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
531 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
532 sizeof(dst->ip6.src));
533 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
534 }
535
536 if (!(hinfo->cfg.mode &
537 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
538 return 0;
539 nexthdr = ipv6_hdr(skb)->nexthdr;
540 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
541 if ((int)protoff < 0)
542 return -1;
543 break;
544 #endif
545 default:
546 BUG();
547 return 0;
548 }
549
550 switch (nexthdr) {
551 case IPPROTO_TCP:
552 case IPPROTO_UDP:
553 case IPPROTO_UDPLITE:
554 case IPPROTO_SCTP:
555 case IPPROTO_DCCP:
556 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
557 &_ports);
558 break;
559 default:
560 _ports[0] = _ports[1] = 0;
561 ports = _ports;
562 break;
563 }
564 if (!ports)
565 return -1;
566 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
567 dst->src_port = ports[0];
568 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
569 dst->dst_port = ports[1];
570 return 0;
571 }
572
573 static bool
574 hashlimit_mt_v0(const struct sk_buff *skb, const struct xt_match_param *par)
575 {
576 const struct xt_hashlimit_info *r = par->matchinfo;
577 struct xt_hashlimit_htable *hinfo = r->hinfo;
578 unsigned long now = jiffies;
579 struct dsthash_ent *dh;
580 struct dsthash_dst dst;
581
582 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
583 goto hotdrop;
584
585 spin_lock_bh(&hinfo->lock);
586 dh = dsthash_find(hinfo, &dst);
587 if (!dh) {
588 dh = dsthash_alloc_init(hinfo, &dst);
589 if (!dh) {
590 spin_unlock_bh(&hinfo->lock);
591 goto hotdrop;
592 }
593
594 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
595 dh->rateinfo.prev = jiffies;
596 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
597 hinfo->cfg.burst);
598 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
599 hinfo->cfg.burst);
600 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
601 } else {
602 /* update expiration timeout */
603 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
604 rateinfo_recalc(dh, now);
605 }
606
607 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
608 /* We're underlimit. */
609 dh->rateinfo.credit -= dh->rateinfo.cost;
610 spin_unlock_bh(&hinfo->lock);
611 return true;
612 }
613
614 spin_unlock_bh(&hinfo->lock);
615
616 /* default case: we're overlimit, thus don't match */
617 return false;
618
619 hotdrop:
620 *par->hotdrop = true;
621 return false;
622 }
623
624 static bool
625 hashlimit_mt(const struct sk_buff *skb, const struct xt_match_param *par)
626 {
627 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
628 struct xt_hashlimit_htable *hinfo = info->hinfo;
629 unsigned long now = jiffies;
630 struct dsthash_ent *dh;
631 struct dsthash_dst dst;
632
633 if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
634 goto hotdrop;
635
636 spin_lock_bh(&hinfo->lock);
637 dh = dsthash_find(hinfo, &dst);
638 if (dh == NULL) {
639 dh = dsthash_alloc_init(hinfo, &dst);
640 if (dh == NULL) {
641 spin_unlock_bh(&hinfo->lock);
642 goto hotdrop;
643 }
644
645 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
646 dh->rateinfo.prev = jiffies;
647 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
648 hinfo->cfg.burst);
649 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
650 hinfo->cfg.burst);
651 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
652 } else {
653 /* update expiration timeout */
654 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
655 rateinfo_recalc(dh, now);
656 }
657
658 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
659 /* below the limit */
660 dh->rateinfo.credit -= dh->rateinfo.cost;
661 spin_unlock_bh(&hinfo->lock);
662 return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
663 }
664
665 spin_unlock_bh(&hinfo->lock);
666 /* default match is underlimit - so over the limit, we need to invert */
667 return info->cfg.mode & XT_HASHLIMIT_INVERT;
668
669 hotdrop:
670 *par->hotdrop = true;
671 return false;
672 }
673
674 static int hashlimit_mt_check_v0(const struct xt_mtchk_param *par)
675 {
676 struct net *net = par->net;
677 struct xt_hashlimit_info *r = par->matchinfo;
678 int ret;
679
680 /* Check for overflow. */
681 if (r->cfg.burst == 0 ||
682 user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
683 pr_info("overflow, try lower: %u/%u\n",
684 r->cfg.avg, r->cfg.burst);
685 return -ERANGE;
686 }
687 if (r->cfg.mode == 0 ||
688 r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
689 XT_HASHLIMIT_HASH_DIP |
690 XT_HASHLIMIT_HASH_SIP |
691 XT_HASHLIMIT_HASH_SPT))
692 return -EINVAL;
693 if (!r->cfg.gc_interval)
694 return -EINVAL;
695 if (!r->cfg.expire)
696 return -EINVAL;
697 if (r->name[sizeof(r->name) - 1] != '\0')
698 return -EINVAL;
699
700 mutex_lock(&hashlimit_mutex);
701 r->hinfo = htable_find_get(net, r->name, par->family);
702 if (r->hinfo == NULL) {
703 ret = htable_create_v0(net, r, par->family);
704 if (ret < 0) {
705 mutex_unlock(&hashlimit_mutex);
706 return ret;
707 }
708 }
709 mutex_unlock(&hashlimit_mutex);
710 return 0;
711 }
712
713 static int hashlimit_mt_check(const struct xt_mtchk_param *par)
714 {
715 struct net *net = par->net;
716 struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
717 int ret;
718
719 /* Check for overflow. */
720 if (info->cfg.burst == 0 ||
721 user2credits(info->cfg.avg * info->cfg.burst) <
722 user2credits(info->cfg.avg)) {
723 pr_info("overflow, try lower: %u/%u\n",
724 info->cfg.avg, info->cfg.burst);
725 return -ERANGE;
726 }
727 if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
728 return -EINVAL;
729 if (info->name[sizeof(info->name)-1] != '\0')
730 return -EINVAL;
731 if (par->family == NFPROTO_IPV4) {
732 if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
733 return -EINVAL;
734 } else {
735 if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
736 return -EINVAL;
737 }
738
739 mutex_lock(&hashlimit_mutex);
740 info->hinfo = htable_find_get(net, info->name, par->family);
741 if (info->hinfo == NULL) {
742 ret = htable_create(net, info, par->family);
743 if (ret < 0) {
744 mutex_unlock(&hashlimit_mutex);
745 return ret;
746 }
747 }
748 mutex_unlock(&hashlimit_mutex);
749 return 0;
750 }
751
752 static void
753 hashlimit_mt_destroy_v0(const struct xt_mtdtor_param *par)
754 {
755 const struct xt_hashlimit_info *r = par->matchinfo;
756
757 htable_put(r->hinfo);
758 }
759
760 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
761 {
762 const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
763
764 htable_put(info->hinfo);
765 }
766
767 #ifdef CONFIG_COMPAT
768 struct compat_xt_hashlimit_info {
769 char name[IFNAMSIZ];
770 struct hashlimit_cfg cfg;
771 compat_uptr_t hinfo;
772 compat_uptr_t master;
773 };
774
775 static void hashlimit_mt_compat_from_user(void *dst, const void *src)
776 {
777 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
778
779 memcpy(dst, src, off);
780 memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
781 }
782
783 static int hashlimit_mt_compat_to_user(void __user *dst, const void *src)
784 {
785 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
786
787 return copy_to_user(dst, src, off) ? -EFAULT : 0;
788 }
789 #endif
790
791 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
792 {
793 .name = "hashlimit",
794 .revision = 0,
795 .family = NFPROTO_IPV4,
796 .match = hashlimit_mt_v0,
797 .matchsize = sizeof(struct xt_hashlimit_info),
798 #ifdef CONFIG_COMPAT
799 .compatsize = sizeof(struct compat_xt_hashlimit_info),
800 .compat_from_user = hashlimit_mt_compat_from_user,
801 .compat_to_user = hashlimit_mt_compat_to_user,
802 #endif
803 .checkentry = hashlimit_mt_check_v0,
804 .destroy = hashlimit_mt_destroy_v0,
805 .me = THIS_MODULE
806 },
807 {
808 .name = "hashlimit",
809 .revision = 1,
810 .family = NFPROTO_IPV4,
811 .match = hashlimit_mt,
812 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
813 .checkentry = hashlimit_mt_check,
814 .destroy = hashlimit_mt_destroy,
815 .me = THIS_MODULE,
816 },
817 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
818 {
819 .name = "hashlimit",
820 .family = NFPROTO_IPV6,
821 .match = hashlimit_mt_v0,
822 .matchsize = sizeof(struct xt_hashlimit_info),
823 #ifdef CONFIG_COMPAT
824 .compatsize = sizeof(struct compat_xt_hashlimit_info),
825 .compat_from_user = hashlimit_mt_compat_from_user,
826 .compat_to_user = hashlimit_mt_compat_to_user,
827 #endif
828 .checkentry = hashlimit_mt_check_v0,
829 .destroy = hashlimit_mt_destroy_v0,
830 .me = THIS_MODULE
831 },
832 {
833 .name = "hashlimit",
834 .revision = 1,
835 .family = NFPROTO_IPV6,
836 .match = hashlimit_mt,
837 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
838 .checkentry = hashlimit_mt_check,
839 .destroy = hashlimit_mt_destroy,
840 .me = THIS_MODULE,
841 },
842 #endif
843 };
844
845 /* PROC stuff */
846 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
847 __acquires(htable->lock)
848 {
849 struct xt_hashlimit_htable *htable = s->private;
850 unsigned int *bucket;
851
852 spin_lock_bh(&htable->lock);
853 if (*pos >= htable->cfg.size)
854 return NULL;
855
856 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
857 if (!bucket)
858 return ERR_PTR(-ENOMEM);
859
860 *bucket = *pos;
861 return bucket;
862 }
863
864 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
865 {
866 struct xt_hashlimit_htable *htable = s->private;
867 unsigned int *bucket = (unsigned int *)v;
868
869 *pos = ++(*bucket);
870 if (*pos >= htable->cfg.size) {
871 kfree(v);
872 return NULL;
873 }
874 return bucket;
875 }
876
877 static void dl_seq_stop(struct seq_file *s, void *v)
878 __releases(htable->lock)
879 {
880 struct xt_hashlimit_htable *htable = s->private;
881 unsigned int *bucket = (unsigned int *)v;
882
883 kfree(bucket);
884 spin_unlock_bh(&htable->lock);
885 }
886
887 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
888 struct seq_file *s)
889 {
890 /* recalculate to show accurate numbers */
891 rateinfo_recalc(ent, jiffies);
892
893 switch (family) {
894 case NFPROTO_IPV4:
895 return seq_printf(s, "%ld %pI4:%u->%pI4:%u %u %u %u\n",
896 (long)(ent->expires - jiffies)/HZ,
897 &ent->dst.ip.src,
898 ntohs(ent->dst.src_port),
899 &ent->dst.ip.dst,
900 ntohs(ent->dst.dst_port),
901 ent->rateinfo.credit, ent->rateinfo.credit_cap,
902 ent->rateinfo.cost);
903 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
904 case NFPROTO_IPV6:
905 return seq_printf(s, "%ld %pI6:%u->%pI6:%u %u %u %u\n",
906 (long)(ent->expires - jiffies)/HZ,
907 &ent->dst.ip6.src,
908 ntohs(ent->dst.src_port),
909 &ent->dst.ip6.dst,
910 ntohs(ent->dst.dst_port),
911 ent->rateinfo.credit, ent->rateinfo.credit_cap,
912 ent->rateinfo.cost);
913 #endif
914 default:
915 BUG();
916 return 0;
917 }
918 }
919
920 static int dl_seq_show(struct seq_file *s, void *v)
921 {
922 struct xt_hashlimit_htable *htable = s->private;
923 unsigned int *bucket = (unsigned int *)v;
924 struct dsthash_ent *ent;
925 struct hlist_node *pos;
926
927 if (!hlist_empty(&htable->hash[*bucket])) {
928 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
929 if (dl_seq_real_show(ent, htable->family, s))
930 return -1;
931 }
932 return 0;
933 }
934
935 static const struct seq_operations dl_seq_ops = {
936 .start = dl_seq_start,
937 .next = dl_seq_next,
938 .stop = dl_seq_stop,
939 .show = dl_seq_show
940 };
941
942 static int dl_proc_open(struct inode *inode, struct file *file)
943 {
944 int ret = seq_open(file, &dl_seq_ops);
945
946 if (!ret) {
947 struct seq_file *sf = file->private_data;
948 sf->private = PDE(inode)->data;
949 }
950 return ret;
951 }
952
953 static const struct file_operations dl_file_ops = {
954 .owner = THIS_MODULE,
955 .open = dl_proc_open,
956 .read = seq_read,
957 .llseek = seq_lseek,
958 .release = seq_release
959 };
960
961 static int __net_init hashlimit_proc_net_init(struct net *net)
962 {
963 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
964
965 hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
966 if (!hashlimit_net->ipt_hashlimit)
967 return -ENOMEM;
968 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
969 hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
970 if (!hashlimit_net->ip6t_hashlimit) {
971 proc_net_remove(net, "ipt_hashlimit");
972 return -ENOMEM;
973 }
974 #endif
975 return 0;
976 }
977
978 static void __net_exit hashlimit_proc_net_exit(struct net *net)
979 {
980 proc_net_remove(net, "ipt_hashlimit");
981 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
982 proc_net_remove(net, "ip6t_hashlimit");
983 #endif
984 }
985
986 static int __net_init hashlimit_net_init(struct net *net)
987 {
988 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
989
990 INIT_HLIST_HEAD(&hashlimit_net->htables);
991 return hashlimit_proc_net_init(net);
992 }
993
994 static void __net_exit hashlimit_net_exit(struct net *net)
995 {
996 struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
997
998 BUG_ON(!hlist_empty(&hashlimit_net->htables));
999 hashlimit_proc_net_exit(net);
1000 }
1001
1002 static struct pernet_operations hashlimit_net_ops = {
1003 .init = hashlimit_net_init,
1004 .exit = hashlimit_net_exit,
1005 .id = &hashlimit_net_id,
1006 .size = sizeof(struct hashlimit_net),
1007 };
1008
1009 static int __init hashlimit_mt_init(void)
1010 {
1011 int err;
1012
1013 err = register_pernet_subsys(&hashlimit_net_ops);
1014 if (err < 0)
1015 return err;
1016 err = xt_register_matches(hashlimit_mt_reg,
1017 ARRAY_SIZE(hashlimit_mt_reg));
1018 if (err < 0)
1019 goto err1;
1020
1021 err = -ENOMEM;
1022 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1023 sizeof(struct dsthash_ent), 0, 0,
1024 NULL);
1025 if (!hashlimit_cachep) {
1026 pr_warning("unable to create slab cache\n");
1027 goto err2;
1028 }
1029 return 0;
1030
1031 err2:
1032 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1033 err1:
1034 unregister_pernet_subsys(&hashlimit_net_ops);
1035 return err;
1036
1037 }
1038
1039 static void __exit hashlimit_mt_exit(void)
1040 {
1041 kmem_cache_destroy(hashlimit_cachep);
1042 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1043 unregister_pernet_subsys(&hashlimit_net_ops);
1044 }
1045
1046 module_init(hashlimit_mt_init);
1047 module_exit(hashlimit_mt_exit);