?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[NETFILTER]: xt_hashlimit match, revision 1
[linux-2.6/libata-dev.git] / net / netfilter / xt_hashlimit.c
blob54aaf5bac2d72fd0edca94ed2c31a0b8793b49ac
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
36 MODULE_LICENSE("GPL");
37 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
38 MODULE_AUTHOR("Jan Engelhardt <jengelh@computergmbh.de>");
39 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
40 MODULE_ALIAS("ipt_hashlimit");
41 MODULE_ALIAS("ip6t_hashlimit");
42
43 /* need to declare this at the top */
44 static struct proc_dir_entry *hashlimit_procdir4;
45 static struct proc_dir_entry *hashlimit_procdir6;
46 static const struct file_operations dl_file_ops;
47
48 /* hash table crap */
49 struct 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
66 struct 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
80 struct 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
100 static DEFINE_SPINLOCK(hashlimit_lock); /* protects htables list */
101 static DEFINE_MUTEX(hlimit_mutex); /* additional checkentry protection */
102 static HLIST_HEAD(hashlimit_htables);
103 static struct kmem_cache *hashlimit_cachep __read_mostly;
104
105 static 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
111 static u_int32_t
112 hash_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
126 static struct dsthash_ent *
127 dsthash_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 */
143 static struct dsthash_ent *
144 dsthash_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
179 static inline void
180 dsthash_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 }
186 static void htable_gc(unsigned long htlong);
187
188 static 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 = create_proc_entry(minfo->name, 0,
241 family == AF_INET ? hashlimit_procdir4 :
242 hashlimit_procdir6);
243 if (!hinfo->pde) {
244 vfree(hinfo);
245 return -1;
246 }
247 hinfo->pde->proc_fops = &dl_file_ops;
248 hinfo->pde->data = hinfo;
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
261 static 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 = create_proc_entry(minfo->name, 0,
305 family == AF_INET ? hashlimit_procdir4 :
306 hashlimit_procdir6);
307 if (hinfo->pde == NULL) {
308 vfree(hinfo);
309 return -1;
310 }
311 hinfo->pde->proc_fops = &dl_file_ops;
312 hinfo->pde->data = hinfo;
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
325 static bool select_all(const struct xt_hashlimit_htable *ht,
326 const struct dsthash_ent *he)
327 {
328 return 1;
329 }
330
331 static 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
337 static 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 */
357 static 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
368 static 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
382 static 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
401 static 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. */
448 static inline u_int32_t
449 user2credits(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
459 static 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
467 static inline __be32 maskl(__be32 a, unsigned int l)
468 {
469 return htonl(ntohl(a) & ~(~(u_int32_t)0 >> l));
470 }
471
472 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
473 {
474 switch (p) {
475 case 0:
476 i[0] = i[1] = 0;
477 i[2] = i[3] = 0;
478 break;
479 case 1 ... 31:
480 i[0] = maskl(i[0], p);
481 i[1] = i[2] = i[3] = 0;
482 break;
483 case 32:
484 i[1] = i[2] = i[3] = 0;
485 break;
486 case 33 ... 63:
487 i[1] = maskl(i[1], p - 32);
488 i[2] = i[3] = 0;
489 break;
490 case 64:
491 i[2] = i[3] = 0;
492 break;
493 case 65 ... 95:
494 i[2] = maskl(i[2], p - 64);
495 i[3] = 0;
496 case 96:
497 i[3] = 0;
498 break;
499 case 97 ... 127:
500 i[3] = maskl(i[3], p - 96);
501 break;
502 case 128:
503 break;
504 }
505 }
506
507 static int
508 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
509 struct dsthash_dst *dst,
510 const struct sk_buff *skb, unsigned int protoff)
511 {
512 __be16 _ports[2], *ports;
513 u8 nexthdr;
514
515 memset(dst, 0, sizeof(*dst));
516
517 switch (hinfo->family) {
518 case AF_INET:
519 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
520 dst->ip.dst = maskl(ip_hdr(skb)->daddr,
521 hinfo->cfg.dstmask);
522 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
523 dst->ip.src = maskl(ip_hdr(skb)->saddr,
524 hinfo->cfg.srcmask);
525
526 if (!(hinfo->cfg.mode &
527 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
528 return 0;
529 nexthdr = ip_hdr(skb)->protocol;
530 break;
531 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
532 case AF_INET6:
533 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
534 memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
535 sizeof(dst->ip6.dst));
536 hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
537 }
538 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
539 memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
540 sizeof(dst->ip6.src));
541 hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
542 }
543
544 if (!(hinfo->cfg.mode &
545 (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
546 return 0;
547 nexthdr = ipv6_hdr(skb)->nexthdr;
548 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
549 if ((int)protoff < 0)
550 return -1;
551 break;
552 #endif
553 default:
554 BUG();
555 return 0;
556 }
557
558 switch (nexthdr) {
559 case IPPROTO_TCP:
560 case IPPROTO_UDP:
561 case IPPROTO_UDPLITE:
562 case IPPROTO_SCTP:
563 case IPPROTO_DCCP:
564 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
565 &_ports);
566 break;
567 default:
568 _ports[0] = _ports[1] = 0;
569 ports = _ports;
570 break;
571 }
572 if (!ports)
573 return -1;
574 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
575 dst->src_port = ports[0];
576 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
577 dst->dst_port = ports[1];
578 return 0;
579 }
580
581 static bool
582 hashlimit_mt_v0(const struct sk_buff *skb, const struct net_device *in,
583 const struct net_device *out, const struct xt_match *match,
584 const void *matchinfo, int offset, unsigned int protoff,
585 bool *hotdrop)
586 {
587 const struct xt_hashlimit_info *r =
588 ((const struct xt_hashlimit_info *)matchinfo)->u.master;
589 struct xt_hashlimit_htable *hinfo = r->hinfo;
590 unsigned long now = jiffies;
591 struct dsthash_ent *dh;
592 struct dsthash_dst dst;
593
594 if (hashlimit_init_dst(hinfo, &dst, skb, protoff) < 0)
595 goto hotdrop;
596
597 spin_lock_bh(&hinfo->lock);
598 dh = dsthash_find(hinfo, &dst);
599 if (!dh) {
600 dh = dsthash_alloc_init(hinfo, &dst);
601 if (!dh) {
602 spin_unlock_bh(&hinfo->lock);
603 goto hotdrop;
604 }
605
606 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
607 dh->rateinfo.prev = jiffies;
608 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
609 hinfo->cfg.burst);
610 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
611 hinfo->cfg.burst);
612 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
613 } else {
614 /* update expiration timeout */
615 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
616 rateinfo_recalc(dh, now);
617 }
618
619 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
620 /* We're underlimit. */
621 dh->rateinfo.credit -= dh->rateinfo.cost;
622 spin_unlock_bh(&hinfo->lock);
623 return true;
624 }
625
626 spin_unlock_bh(&hinfo->lock);
627
628 /* default case: we're overlimit, thus don't match */
629 return false;
630
631 hotdrop:
632 *hotdrop = true;
633 return false;
634 }
635
636 static bool
637 hashlimit_mt(const struct sk_buff *skb, const struct net_device *in,
638 const struct net_device *out, const struct xt_match *match,
639 const void *matchinfo, int offset, unsigned int protoff,
640 bool *hotdrop)
641 {
642 const struct xt_hashlimit_mtinfo1 *info = matchinfo;
643 struct xt_hashlimit_htable *hinfo = info->hinfo;
644 unsigned long now = jiffies;
645 struct dsthash_ent *dh;
646 struct dsthash_dst dst;
647
648 if (hashlimit_init_dst(hinfo, &dst, skb, protoff) < 0)
649 goto hotdrop;
650
651 spin_lock_bh(&hinfo->lock);
652 dh = dsthash_find(hinfo, &dst);
653 if (dh == NULL) {
654 dh = dsthash_alloc_init(hinfo, &dst);
655 if (dh == NULL) {
656 spin_unlock_bh(&hinfo->lock);
657 goto hotdrop;
658 }
659
660 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
661 dh->rateinfo.prev = jiffies;
662 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
663 hinfo->cfg.burst);
664 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
665 hinfo->cfg.burst);
666 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
667 } else {
668 /* update expiration timeout */
669 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
670 rateinfo_recalc(dh, now);
671 }
672
673 if (dh->rateinfo.credit >= dh->rateinfo.cost) {
674 /* below the limit */
675 dh->rateinfo.credit -= dh->rateinfo.cost;
676 spin_unlock_bh(&hinfo->lock);
677 return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
678 }
679
680 spin_unlock_bh(&hinfo->lock);
681 /* default match is underlimit - so over the limit, we need to invert */
682 return info->cfg.mode & XT_HASHLIMIT_INVERT;
683
684 hotdrop:
685 *hotdrop = true;
686 return false;
687 }
688
689 static bool
690 hashlimit_mt_check_v0(const char *tablename, const void *inf,
691 const struct xt_match *match, void *matchinfo,
692 unsigned int hook_mask)
693 {
694 struct xt_hashlimit_info *r = matchinfo;
695
696 /* Check for overflow. */
697 if (r->cfg.burst == 0 ||
698 user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
699 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
700 r->cfg.avg, r->cfg.burst);
701 return false;
702 }
703 if (r->cfg.mode == 0 ||
704 r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
705 XT_HASHLIMIT_HASH_DIP |
706 XT_HASHLIMIT_HASH_SIP |
707 XT_HASHLIMIT_HASH_SPT))
708 return false;
709 if (!r->cfg.gc_interval)
710 return false;
711 if (!r->cfg.expire)
712 return false;
713 if (r->name[sizeof(r->name) - 1] != '\0')
714 return false;
715
716 /* This is the best we've got: We cannot release and re-grab lock,
717 * since checkentry() is called before x_tables.c grabs xt_mutex.
718 * We also cannot grab the hashtable spinlock, since htable_create will
719 * call vmalloc, and that can sleep. And we cannot just re-search
720 * the list of htable's in htable_create(), since then we would
721 * create duplicate proc files. -HW */
722 mutex_lock(&hlimit_mutex);
723 r->hinfo = htable_find_get(r->name, match->family);
724 if (!r->hinfo && htable_create_v0(r, match->family) != 0) {
725 mutex_unlock(&hlimit_mutex);
726 return false;
727 }
728 mutex_unlock(&hlimit_mutex);
729
730 /* Ugly hack: For SMP, we only want to use one set */
731 r->u.master = r;
732 return true;
733 }
734
735 static bool
736 hashlimit_mt_check(const char *tablename, const void *inf,
737 const struct xt_match *match, void *matchinfo,
738 unsigned int hook_mask)
739 {
740 struct xt_hashlimit_mtinfo1 *info = matchinfo;
741
742 /* Check for overflow. */
743 if (info->cfg.burst == 0 ||
744 user2credits(info->cfg.avg * info->cfg.burst) <
745 user2credits(info->cfg.avg)) {
746 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
747 info->cfg.avg, info->cfg.burst);
748 return false;
749 }
750 if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
751 return false;
752 if (info->name[sizeof(info->name)-1] != '\0')
753 return false;
754 if (match->family == AF_INET) {
755 if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
756 return false;
757 } else {
758 if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
759 return false;
760 }
761
762 /* This is the best we've got: We cannot release and re-grab lock,
763 * since checkentry() is called before x_tables.c grabs xt_mutex.
764 * We also cannot grab the hashtable spinlock, since htable_create will
765 * call vmalloc, and that can sleep. And we cannot just re-search
766 * the list of htable's in htable_create(), since then we would
767 * create duplicate proc files. -HW */
768 mutex_lock(&hlimit_mutex);
769 info->hinfo = htable_find_get(info->name, match->family);
770 if (!info->hinfo && htable_create(info, match->family) != 0) {
771 mutex_unlock(&hlimit_mutex);
772 return false;
773 }
774 mutex_unlock(&hlimit_mutex);
775
776 /* Ugly hack: For SMP, we only want to use one set */
777 info->master = info;
778 return true;
779 }
780
781 static void
782 hashlimit_mt_destroy_v0(const struct xt_match *match, void *matchinfo)
783 {
784 const struct xt_hashlimit_info *r = matchinfo;
785
786 htable_put(r->hinfo);
787 }
788
789 static void
790 hashlimit_mt_destroy(const struct xt_match *match, void *matchinfo)
791 {
792 const struct xt_hashlimit_mtinfo1 *info = matchinfo;
793
794 htable_put(info->hinfo);
795 }
796
797 #ifdef CONFIG_COMPAT
798 struct compat_xt_hashlimit_info {
799 char name[IFNAMSIZ];
800 struct hashlimit_cfg cfg;
801 compat_uptr_t hinfo;
802 compat_uptr_t master;
803 };
804
805 static void hashlimit_mt_compat_from_user(void *dst, void *src)
806 {
807 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
808
809 memcpy(dst, src, off);
810 memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
811 }
812
813 static int hashlimit_mt_compat_to_user(void __user *dst, void *src)
814 {
815 int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
816
817 return copy_to_user(dst, src, off) ? -EFAULT : 0;
818 }
819 #endif
820
821 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
822 {
823 .name = "hashlimit",
824 .revision = 0,
825 .family = AF_INET,
826 .match = hashlimit_mt_v0,
827 .matchsize = sizeof(struct xt_hashlimit_info),
828 #ifdef CONFIG_COMPAT
829 .compatsize = sizeof(struct compat_xt_hashlimit_info),
830 .compat_from_user = hashlimit_mt_compat_from_user,
831 .compat_to_user = hashlimit_mt_compat_to_user,
832 #endif
833 .checkentry = hashlimit_mt_check_v0,
834 .destroy = hashlimit_mt_destroy_v0,
835 .me = THIS_MODULE
836 },
837 {
838 .name = "hashlimit",
839 .revision = 1,
840 .family = AF_INET,
841 .match = hashlimit_mt,
842 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
843 .checkentry = hashlimit_mt_check,
844 .destroy = hashlimit_mt_destroy,
845 .me = THIS_MODULE,
846 },
847 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
848 {
849 .name = "hashlimit",
850 .family = AF_INET6,
851 .match = hashlimit_mt_v0,
852 .matchsize = sizeof(struct xt_hashlimit_info),
853 #ifdef CONFIG_COMPAT
854 .compatsize = sizeof(struct compat_xt_hashlimit_info),
855 .compat_from_user = hashlimit_mt_compat_from_user,
856 .compat_to_user = hashlimit_mt_compat_to_user,
857 #endif
858 .checkentry = hashlimit_mt_check_v0,
859 .destroy = hashlimit_mt_destroy_v0,
860 .me = THIS_MODULE
861 },
862 {
863 .name = "hashlimit",
864 .revision = 1,
865 .family = AF_INET6,
866 .match = hashlimit_mt,
867 .matchsize = sizeof(struct xt_hashlimit_mtinfo1),
868 .checkentry = hashlimit_mt_check,
869 .destroy = hashlimit_mt_destroy,
870 .me = THIS_MODULE,
871 },
872 #endif
873 };
874
875 /* PROC stuff */
876 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
877 __acquires(htable->lock)
878 {
879 struct proc_dir_entry *pde = s->private;
880 struct xt_hashlimit_htable *htable = pde->data;
881 unsigned int *bucket;
882
883 spin_lock_bh(&htable->lock);
884 if (*pos >= htable->cfg.size)
885 return NULL;
886
887 bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
888 if (!bucket)
889 return ERR_PTR(-ENOMEM);
890
891 *bucket = *pos;
892 return bucket;
893 }
894
895 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
896 {
897 struct proc_dir_entry *pde = s->private;
898 struct xt_hashlimit_htable *htable = pde->data;
899 unsigned int *bucket = (unsigned int *)v;
900
901 *pos = ++(*bucket);
902 if (*pos >= htable->cfg.size) {
903 kfree(v);
904 return NULL;
905 }
906 return bucket;
907 }
908
909 static void dl_seq_stop(struct seq_file *s, void *v)
910 __releases(htable->lock)
911 {
912 struct proc_dir_entry *pde = s->private;
913 struct xt_hashlimit_htable *htable = pde->data;
914 unsigned int *bucket = (unsigned int *)v;
915
916 kfree(bucket);
917 spin_unlock_bh(&htable->lock);
918 }
919
920 static int dl_seq_real_show(struct dsthash_ent *ent, int family,
921 struct seq_file *s)
922 {
923 /* recalculate to show accurate numbers */
924 rateinfo_recalc(ent, jiffies);
925
926 switch (family) {
927 case AF_INET:
928 return seq_printf(s, "%ld %u.%u.%u.%u:%u->"
929 "%u.%u.%u.%u:%u %u %u %u\n",
930 (long)(ent->expires - jiffies)/HZ,
931 NIPQUAD(ent->dst.ip.src),
932 ntohs(ent->dst.src_port),
933 NIPQUAD(ent->dst.ip.dst),
934 ntohs(ent->dst.dst_port),
935 ent->rateinfo.credit, ent->rateinfo.credit_cap,
936 ent->rateinfo.cost);
937 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
938 case AF_INET6:
939 return seq_printf(s, "%ld " NIP6_FMT ":%u->"
940 NIP6_FMT ":%u %u %u %u\n",
941 (long)(ent->expires - jiffies)/HZ,
942 NIP6(*(struct in6_addr *)&ent->dst.ip6.src),
943 ntohs(ent->dst.src_port),
944 NIP6(*(struct in6_addr *)&ent->dst.ip6.dst),
945 ntohs(ent->dst.dst_port),
946 ent->rateinfo.credit, ent->rateinfo.credit_cap,
947 ent->rateinfo.cost);
948 #endif
949 default:
950 BUG();
951 return 0;
952 }
953 }
954
955 static int dl_seq_show(struct seq_file *s, void *v)
956 {
957 struct proc_dir_entry *pde = s->private;
958 struct xt_hashlimit_htable *htable = pde->data;
959 unsigned int *bucket = (unsigned int *)v;
960 struct dsthash_ent *ent;
961 struct hlist_node *pos;
962
963 if (!hlist_empty(&htable->hash[*bucket])) {
964 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
965 if (dl_seq_real_show(ent, htable->family, s))
966 return 1;
967 }
968 return 0;
969 }
970
971 static const struct seq_operations dl_seq_ops = {
972 .start = dl_seq_start,
973 .next = dl_seq_next,
974 .stop = dl_seq_stop,
975 .show = dl_seq_show
976 };
977
978 static int dl_proc_open(struct inode *inode, struct file *file)
979 {
980 int ret = seq_open(file, &dl_seq_ops);
981
982 if (!ret) {
983 struct seq_file *sf = file->private_data;
984 sf->private = PDE(inode);
985 }
986 return ret;
987 }
988
989 static const struct file_operations dl_file_ops = {
990 .owner = THIS_MODULE,
991 .open = dl_proc_open,
992 .read = seq_read,
993 .llseek = seq_lseek,
994 .release = seq_release
995 };
996
997 static int __init hashlimit_mt_init(void)
998 {
999 int err;
1000
1001 err = xt_register_matches(hashlimit_mt_reg,
1002 ARRAY_SIZE(hashlimit_mt_reg));
1003 if (err < 0)
1004 goto err1;
1005
1006 err = -ENOMEM;
1007 hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1008 sizeof(struct dsthash_ent), 0, 0,
1009 NULL);
1010 if (!hashlimit_cachep) {
1011 printk(KERN_ERR "xt_hashlimit: unable to create slab cache\n");
1012 goto err2;
1013 }
1014 hashlimit_procdir4 = proc_mkdir("ipt_hashlimit", init_net.proc_net);
1015 if (!hashlimit_procdir4) {
1016 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
1017 "entry\n");
1018 goto err3;
1019 }
1020 err = 0;
1021 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
1022 hashlimit_procdir6 = proc_mkdir("ip6t_hashlimit", init_net.proc_net);
1023 if (!hashlimit_procdir6) {
1024 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
1025 "entry\n");
1026 err = -ENOMEM;
1027 }
1028 #endif
1029 if (!err)
1030 return 0;
1031 remove_proc_entry("ipt_hashlimit", init_net.proc_net);
1032 err3:
1033 kmem_cache_destroy(hashlimit_cachep);
1034 err2:
1035 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1036 err1:
1037 return err;
1038
1039 }
1040
1041 static void __exit hashlimit_mt_exit(void)
1042 {
1043 remove_proc_entry("ipt_hashlimit", init_net.proc_net);
1044 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
1045 remove_proc_entry("ip6t_hashlimit", init_net.proc_net);
1046 #endif
1047 kmem_cache_destroy(hashlimit_cachep);
1048 xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1049 }
1050
1051 module_init(hashlimit_mt_init);
1052 module_exit(hashlimit_mt_exit);
Linux 2.6 libata-dev (mirror)