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