net: replace NIPQUAD() in net/netfilter/
[linux-2.6.git] / 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         u_int8_t 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, u_int8_t 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 == NFPROTO_IPV4)
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 = proc_create_data(minfo->name, 0,
241                 (family == NFPROTO_IPV4) ?
242                 hashlimit_procdir4 : hashlimit_procdir6,
243                 &dl_file_ops, hinfo);
244         if (!hinfo->pde) {
245                 vfree(hinfo);
246                 return -1;
247         }
248
249         setup_timer(&hinfo->timer, htable_gc, (unsigned long )hinfo);
250         hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
251         add_timer(&hinfo->timer);
252
253         spin_lock_bh(&hashlimit_lock);
254         hlist_add_head(&hinfo->node, &hashlimit_htables);
255         spin_unlock_bh(&hashlimit_lock);
256
257         return 0;
258 }
259
260 static int htable_create(struct xt_hashlimit_mtinfo1 *minfo, u_int8_t family)
261 {
262         struct xt_hashlimit_htable *hinfo;
263         unsigned int size;
264         unsigned int i;
265
266         if (minfo->cfg.size) {
267                 size = minfo->cfg.size;
268         } else {
269                 size = (num_physpages << PAGE_SHIFT) / 16384 /
270                        sizeof(struct list_head);
271                 if (num_physpages > 1024 * 1024 * 1024 / PAGE_SIZE)
272                         size = 8192;
273                 if (size < 16)
274                         size = 16;
275         }
276         /* FIXME: don't use vmalloc() here or anywhere else -HW */
277         hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
278                         sizeof(struct list_head) * size);
279         if (hinfo == NULL) {
280                 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
281                 return -1;
282         }
283         minfo->hinfo = hinfo;
284
285         /* copy match config into hashtable config */
286         memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
287         hinfo->cfg.size = size;
288         if (hinfo->cfg.max == 0)
289                 hinfo->cfg.max = 8 * hinfo->cfg.size;
290         else if (hinfo->cfg.max < hinfo->cfg.size)
291                 hinfo->cfg.max = hinfo->cfg.size;
292
293         for (i = 0; i < hinfo->cfg.size; i++)
294                 INIT_HLIST_HEAD(&hinfo->hash[i]);
295
296         atomic_set(&hinfo->use, 1);
297         hinfo->count = 0;
298         hinfo->family = family;
299         hinfo->rnd_initialized = 0;
300         spin_lock_init(&hinfo->lock);
301
302         hinfo->pde = proc_create_data(minfo->name, 0,
303                 (family == NFPROTO_IPV4) ?
304                 hashlimit_procdir4 : hashlimit_procdir6,
305                 &dl_file_ops, hinfo);
306         if (hinfo->pde == NULL) {
307                 vfree(hinfo);
308                 return -1;
309         }
310
311         setup_timer(&hinfo->timer, htable_gc, (unsigned long)hinfo);
312         hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
313         add_timer(&hinfo->timer);
314
315         spin_lock_bh(&hashlimit_lock);
316         hlist_add_head(&hinfo->node, &hashlimit_htables);
317         spin_unlock_bh(&hashlimit_lock);
318
319         return 0;
320 }
321
322 static bool select_all(const struct xt_hashlimit_htable *ht,
323                        const struct dsthash_ent *he)
324 {
325         return 1;
326 }
327
328 static bool select_gc(const struct xt_hashlimit_htable *ht,
329                       const struct dsthash_ent *he)
330 {
331         return time_after_eq(jiffies, he->expires);
332 }
333
334 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
335                         bool (*select)(const struct xt_hashlimit_htable *ht,
336                                       const struct dsthash_ent *he))
337 {
338         unsigned int i;
339
340         /* lock hash table and iterate over it */
341         spin_lock_bh(&ht->lock);
342         for (i = 0; i < ht->cfg.size; i++) {
343                 struct dsthash_ent *dh;
344                 struct hlist_node *pos, *n;
345                 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
346                         if ((*select)(ht, dh))
347                                 dsthash_free(ht, dh);
348                 }
349         }
350         spin_unlock_bh(&ht->lock);
351 }
352
353 /* hash table garbage collector, run by timer */
354 static void htable_gc(unsigned long htlong)
355 {
356         struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
357
358         htable_selective_cleanup(ht, select_gc);
359
360         /* re-add the timer accordingly */
361         ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
362         add_timer(&ht->timer);
363 }
364
365 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
366 {
367         del_timer_sync(&hinfo->timer);
368
369         /* remove proc entry */
370         remove_proc_entry(hinfo->pde->name,
371                           hinfo->family == NFPROTO_IPV4 ? hashlimit_procdir4 :
372                                                      hashlimit_procdir6);
373         htable_selective_cleanup(hinfo, select_all);
374         vfree(hinfo);
375 }
376
377 static struct xt_hashlimit_htable *htable_find_get(const char *name,
378                                                    u_int8_t family)
379 {
380         struct xt_hashlimit_htable *hinfo;
381         struct hlist_node *pos;
382
383         spin_lock_bh(&hashlimit_lock);
384         hlist_for_each_entry(hinfo, pos, &hashlimit_htables, node) {
385                 if (!strcmp(name, hinfo->pde->name) &&
386                     hinfo->family == family) {
387                         atomic_inc(&hinfo->use);
388                         spin_unlock_bh(&hashlimit_lock);
389                         return hinfo;
390                 }
391         }
392         spin_unlock_bh(&hashlimit_lock);
393         return NULL;
394 }
395
396 static void htable_put(struct xt_hashlimit_htable *hinfo)
397 {
398         if (atomic_dec_and_test(&hinfo->use)) {
399                 spin_lock_bh(&hashlimit_lock);
400                 hlist_del(&hinfo->node);
401                 spin_unlock_bh(&hashlimit_lock);
402                 htable_destroy(hinfo);
403         }
404 }
405
406 /* The algorithm used is the Simple Token Bucket Filter (TBF)
407  * see net/sched/sch_tbf.c in the linux source tree
408  */
409
410 /* Rusty: This is my (non-mathematically-inclined) understanding of
411    this algorithm.  The `average rate' in jiffies becomes your initial
412    amount of credit `credit' and the most credit you can ever have
413    `credit_cap'.  The `peak rate' becomes the cost of passing the
414    test, `cost'.
415
416    `prev' tracks the last packet hit: you gain one credit per jiffy.
417    If you get credit balance more than this, the extra credit is
418    discarded.  Every time the match passes, you lose `cost' credits;
419    if you don't have that many, the test fails.
420
421    See Alexey's formal explanation in net/sched/sch_tbf.c.
422
423    To get the maximum range, we multiply by this factor (ie. you get N
424    credits per jiffy).  We want to allow a rate as low as 1 per day
425    (slowest userspace tool allows), which means
426    CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
427 */
428 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
429
430 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
431  * us the power of 2 below the theoretical max, so GCC simply does a
432  * shift. */
433 #define _POW2_BELOW2(x) ((x)|((x)>>1))
434 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
435 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
436 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
437 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
438 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
439
440 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
441
442 /* Precision saver. */
443 static inline u_int32_t
444 user2credits(u_int32_t user)
445 {
446         /* If multiplying would overflow... */
447         if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
448                 /* Divide first. */
449                 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
450
451         return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
452 }
453
454 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
455 {
456         dh->rateinfo.credit += (now - dh->rateinfo.prev) * CREDITS_PER_JIFFY;
457         if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
458                 dh->rateinfo.credit = dh->rateinfo.credit_cap;
459         dh->rateinfo.prev = now;
460 }
461
462 static inline __be32 maskl(__be32 a, unsigned int l)
463 {
464         return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
465 }
466
467 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
468 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
469 {
470         switch (p) {
471         case 0 ... 31:
472                 i[0] = maskl(i[0], p);
473                 i[1] = i[2] = i[3] = 0;
474                 break;
475         case 32 ... 63:
476                 i[1] = maskl(i[1], p - 32);
477                 i[2] = i[3] = 0;
478                 break;
479         case 64 ... 95:
480                 i[2] = maskl(i[2], p - 64);
481                 i[3] = 0;
482         case 96 ... 127:
483                 i[3] = maskl(i[3], p - 96);
484                 break;
485         case 128:
486                 break;
487         }
488 }
489 #endif
490
491 static int
492 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
493                    struct dsthash_dst *dst,
494                    const struct sk_buff *skb, unsigned int protoff)
495 {
496         __be16 _ports[2], *ports;
497         u8 nexthdr;
498
499         memset(dst, 0, sizeof(*dst));
500
501         switch (hinfo->family) {
502         case NFPROTO_IPV4:
503                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
504                         dst->ip.dst = maskl(ip_hdr(skb)->daddr,
505                                       hinfo->cfg.dstmask);
506                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
507                         dst->ip.src = maskl(ip_hdr(skb)->saddr,
508                                       hinfo->cfg.srcmask);
509
510                 if (!(hinfo->cfg.mode &
511                       (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
512                         return 0;
513                 nexthdr = ip_hdr(skb)->protocol;
514                 break;
515 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
516         case NFPROTO_IPV6:
517                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
518                         memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
519                                sizeof(dst->ip6.dst));
520                         hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
521                 }
522                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
523                         memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
524                                sizeof(dst->ip6.src));
525                         hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
526                 }
527
528                 if (!(hinfo->cfg.mode &
529                       (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
530                         return 0;
531                 nexthdr = ipv6_hdr(skb)->nexthdr;
532                 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
533                 if ((int)protoff < 0)
534                         return -1;
535                 break;
536 #endif
537         default:
538                 BUG();
539                 return 0;
540         }
541
542         switch (nexthdr) {
543         case IPPROTO_TCP:
544         case IPPROTO_UDP:
545         case IPPROTO_UDPLITE:
546         case IPPROTO_SCTP:
547         case IPPROTO_DCCP:
548                 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
549                                            &_ports);
550                 break;
551         default:
552                 _ports[0] = _ports[1] = 0;
553                 ports = _ports;
554                 break;
555         }
556         if (!ports)
557                 return -1;
558         if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
559                 dst->src_port = ports[0];
560         if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
561                 dst->dst_port = ports[1];
562         return 0;
563 }
564
565 static bool
566 hashlimit_mt_v0(const struct sk_buff *skb, const struct xt_match_param *par)
567 {
568         const struct xt_hashlimit_info *r =
569                 ((const struct xt_hashlimit_info *)par->matchinfo)->u.master;
570         struct xt_hashlimit_htable *hinfo = r->hinfo;
571         unsigned long now = jiffies;
572         struct dsthash_ent *dh;
573         struct dsthash_dst dst;
574
575         if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
576                 goto hotdrop;
577
578         spin_lock_bh(&hinfo->lock);
579         dh = dsthash_find(hinfo, &dst);
580         if (!dh) {
581                 dh = dsthash_alloc_init(hinfo, &dst);
582                 if (!dh) {
583                         spin_unlock_bh(&hinfo->lock);
584                         goto hotdrop;
585                 }
586
587                 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
588                 dh->rateinfo.prev = jiffies;
589                 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
590                                                    hinfo->cfg.burst);
591                 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
592                                                        hinfo->cfg.burst);
593                 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
594         } else {
595                 /* update expiration timeout */
596                 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
597                 rateinfo_recalc(dh, now);
598         }
599
600         if (dh->rateinfo.credit >= dh->rateinfo.cost) {
601                 /* We're underlimit. */
602                 dh->rateinfo.credit -= dh->rateinfo.cost;
603                 spin_unlock_bh(&hinfo->lock);
604                 return true;
605         }
606
607         spin_unlock_bh(&hinfo->lock);
608
609         /* default case: we're overlimit, thus don't match */
610         return false;
611
612 hotdrop:
613         *par->hotdrop = true;
614         return false;
615 }
616
617 static bool
618 hashlimit_mt(const struct sk_buff *skb, const struct xt_match_param *par)
619 {
620         const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
621         struct xt_hashlimit_htable *hinfo = info->hinfo;
622         unsigned long now = jiffies;
623         struct dsthash_ent *dh;
624         struct dsthash_dst dst;
625
626         if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
627                 goto hotdrop;
628
629         spin_lock_bh(&hinfo->lock);
630         dh = dsthash_find(hinfo, &dst);
631         if (dh == NULL) {
632                 dh = dsthash_alloc_init(hinfo, &dst);
633                 if (dh == NULL) {
634                         spin_unlock_bh(&hinfo->lock);
635                         goto hotdrop;
636                 }
637
638                 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
639                 dh->rateinfo.prev = jiffies;
640                 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
641                                       hinfo->cfg.burst);
642                 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
643                                           hinfo->cfg.burst);
644                 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
645         } else {
646                 /* update expiration timeout */
647                 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
648                 rateinfo_recalc(dh, now);
649         }
650
651         if (dh->rateinfo.credit >= dh->rateinfo.cost) {
652                 /* below the limit */
653                 dh->rateinfo.credit -= dh->rateinfo.cost;
654                 spin_unlock_bh(&hinfo->lock);
655                 return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
656         }
657
658         spin_unlock_bh(&hinfo->lock);
659         /* default match is underlimit - so over the limit, we need to invert */
660         return info->cfg.mode & XT_HASHLIMIT_INVERT;
661
662  hotdrop:
663         *par->hotdrop = true;
664         return false;
665 }
666
667 static bool hashlimit_mt_check_v0(const struct xt_mtchk_param *par)
668 {
669         struct xt_hashlimit_info *r = par->matchinfo;
670
671         /* Check for overflow. */
672         if (r->cfg.burst == 0 ||
673             user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
674                 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
675                        r->cfg.avg, r->cfg.burst);
676                 return false;
677         }
678         if (r->cfg.mode == 0 ||
679             r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
680                            XT_HASHLIMIT_HASH_DIP |
681                            XT_HASHLIMIT_HASH_SIP |
682                            XT_HASHLIMIT_HASH_SPT))
683                 return false;
684         if (!r->cfg.gc_interval)
685                 return false;
686         if (!r->cfg.expire)
687                 return false;
688         if (r->name[sizeof(r->name) - 1] != '\0')
689                 return false;
690
691         /* This is the best we've got: We cannot release and re-grab lock,
692          * since checkentry() is called before x_tables.c grabs xt_mutex.
693          * We also cannot grab the hashtable spinlock, since htable_create will
694          * call vmalloc, and that can sleep.  And we cannot just re-search
695          * the list of htable's in htable_create(), since then we would
696          * create duplicate proc files. -HW */
697         mutex_lock(&hlimit_mutex);
698         r->hinfo = htable_find_get(r->name, par->match->family);
699         if (!r->hinfo && htable_create_v0(r, par->match->family) != 0) {
700                 mutex_unlock(&hlimit_mutex);
701                 return false;
702         }
703         mutex_unlock(&hlimit_mutex);
704
705         /* Ugly hack: For SMP, we only want to use one set */
706         r->u.master = r;
707         return true;
708 }
709
710 static bool hashlimit_mt_check(const struct xt_mtchk_param *par)
711 {
712         struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
713
714         /* Check for overflow. */
715         if (info->cfg.burst == 0 ||
716             user2credits(info->cfg.avg * info->cfg.burst) <
717             user2credits(info->cfg.avg)) {
718                 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
719                        info->cfg.avg, info->cfg.burst);
720                 return false;
721         }
722         if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
723                 return false;
724         if (info->name[sizeof(info->name)-1] != '\0')
725                 return false;
726         if (par->match->family == NFPROTO_IPV4) {
727                 if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
728                         return false;
729         } else {
730                 if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
731                         return false;
732         }
733
734         /* This is the best we've got: We cannot release and re-grab lock,
735          * since checkentry() is called before x_tables.c grabs xt_mutex.
736          * We also cannot grab the hashtable spinlock, since htable_create will
737          * call vmalloc, and that can sleep.  And we cannot just re-search
738          * the list of htable's in htable_create(), since then we would
739          * create duplicate proc files. -HW */
740         mutex_lock(&hlimit_mutex);
741         info->hinfo = htable_find_get(info->name, par->match->family);
742         if (!info->hinfo && htable_create(info, par->match->family) != 0) {
743                 mutex_unlock(&hlimit_mutex);
744                 return false;
745         }
746         mutex_unlock(&hlimit_mutex);
747         return true;
748 }
749
750 static void
751 hashlimit_mt_destroy_v0(const struct xt_mtdtor_param *par)
752 {
753         const struct xt_hashlimit_info *r = par->matchinfo;
754
755         htable_put(r->hinfo);
756 }
757
758 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
759 {
760         const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
761
762         htable_put(info->hinfo);
763 }
764
765 #ifdef CONFIG_COMPAT
766 struct compat_xt_hashlimit_info {
767         char name[IFNAMSIZ];
768         struct hashlimit_cfg cfg;
769         compat_uptr_t hinfo;
770         compat_uptr_t master;
771 };
772
773 static void hashlimit_mt_compat_from_user(void *dst, void *src)
774 {
775         int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
776
777         memcpy(dst, src, off);
778         memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
779 }
780
781 static int hashlimit_mt_compat_to_user(void __user *dst, void *src)
782 {
783         int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
784
785         return copy_to_user(dst, src, off) ? -EFAULT : 0;
786 }
787 #endif
788
789 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
790         {
791                 .name           = "hashlimit",
792                 .revision       = 0,
793                 .family         = NFPROTO_IPV4,
794                 .match          = hashlimit_mt_v0,
795                 .matchsize      = sizeof(struct xt_hashlimit_info),
796 #ifdef CONFIG_COMPAT
797                 .compatsize     = sizeof(struct compat_xt_hashlimit_info),
798                 .compat_from_user = hashlimit_mt_compat_from_user,
799                 .compat_to_user = hashlimit_mt_compat_to_user,
800 #endif
801                 .checkentry     = hashlimit_mt_check_v0,
802                 .destroy        = hashlimit_mt_destroy_v0,
803                 .me             = THIS_MODULE
804         },
805         {
806                 .name           = "hashlimit",
807                 .revision       = 1,
808                 .family         = NFPROTO_IPV4,
809                 .match          = hashlimit_mt,
810                 .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
811                 .checkentry     = hashlimit_mt_check,
812                 .destroy        = hashlimit_mt_destroy,
813                 .me             = THIS_MODULE,
814         },
815 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
816         {
817                 .name           = "hashlimit",
818                 .family         = NFPROTO_IPV6,
819                 .match          = hashlimit_mt_v0,
820                 .matchsize      = sizeof(struct xt_hashlimit_info),
821 #ifdef CONFIG_COMPAT
822                 .compatsize     = sizeof(struct compat_xt_hashlimit_info),
823                 .compat_from_user = hashlimit_mt_compat_from_user,
824                 .compat_to_user = hashlimit_mt_compat_to_user,
825 #endif
826                 .checkentry     = hashlimit_mt_check_v0,
827                 .destroy        = hashlimit_mt_destroy_v0,
828                 .me             = THIS_MODULE
829         },
830         {
831                 .name           = "hashlimit",
832                 .revision       = 1,
833                 .family         = NFPROTO_IPV6,
834                 .match          = hashlimit_mt,
835                 .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
836                 .checkentry     = hashlimit_mt_check,
837                 .destroy        = hashlimit_mt_destroy,
838                 .me             = THIS_MODULE,
839         },
840 #endif
841 };
842
843 /* PROC stuff */
844 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
845         __acquires(htable->lock)
846 {
847         struct proc_dir_entry *pde = s->private;
848         struct xt_hashlimit_htable *htable = pde->data;
849         unsigned int *bucket;
850
851         spin_lock_bh(&htable->lock);
852         if (*pos >= htable->cfg.size)
853                 return NULL;
854
855         bucket = kmalloc(sizeof(unsigned int), GFP_ATOMIC);
856         if (!bucket)
857                 return ERR_PTR(-ENOMEM);
858
859         *bucket = *pos;
860         return bucket;
861 }
862
863 static void *dl_seq_next(struct seq_file *s, void *v, loff_t *pos)
864 {
865         struct proc_dir_entry *pde = s->private;
866         struct xt_hashlimit_htable *htable = pde->data;
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 proc_dir_entry *pde = s->private;
881         struct xt_hashlimit_htable *htable = pde->data;
882         unsigned int *bucket = (unsigned int *)v;
883
884         kfree(bucket);
885         spin_unlock_bh(&htable->lock);
886 }
887
888 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
889                                    struct seq_file *s)
890 {
891         /* recalculate to show accurate numbers */
892         rateinfo_recalc(ent, jiffies);
893
894         switch (family) {
895         case NFPROTO_IPV4:
896                 return seq_printf(s, "%ld %pI4:%u->%pI4:%u %u %u %u\n",
897                                  (long)(ent->expires - jiffies)/HZ,
898                                  &ent->dst.ip.src,
899                                  ntohs(ent->dst.src_port),
900                                  &ent->dst.ip.dst,
901                                  ntohs(ent->dst.dst_port),
902                                  ent->rateinfo.credit, ent->rateinfo.credit_cap,
903                                  ent->rateinfo.cost);
904 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
905         case NFPROTO_IPV6:
906                 return seq_printf(s, "%ld %pI6:%u->%pI6:%u %u %u %u\n",
907                                  (long)(ent->expires - jiffies)/HZ,
908                                  &ent->dst.ip6.src,
909                                  ntohs(ent->dst.src_port),
910                                  &ent->dst.ip6.dst,
911                                  ntohs(ent->dst.dst_port),
912                                  ent->rateinfo.credit, ent->rateinfo.credit_cap,
913                                  ent->rateinfo.cost);
914 #endif
915         default:
916                 BUG();
917                 return 0;
918         }
919 }
920
921 static int dl_seq_show(struct seq_file *s, void *v)
922 {
923         struct proc_dir_entry *pde = s->private;
924         struct xt_hashlimit_htable *htable = pde->data;
925         unsigned int *bucket = (unsigned int *)v;
926         struct dsthash_ent *ent;
927         struct hlist_node *pos;
928
929         if (!hlist_empty(&htable->hash[*bucket])) {
930                 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
931                         if (dl_seq_real_show(ent, htable->family, s))
932                                 return 1;
933         }
934         return 0;
935 }
936
937 static const struct seq_operations dl_seq_ops = {
938         .start = dl_seq_start,
939         .next  = dl_seq_next,
940         .stop  = dl_seq_stop,
941         .show  = dl_seq_show
942 };
943
944 static int dl_proc_open(struct inode *inode, struct file *file)
945 {
946         int ret = seq_open(file, &dl_seq_ops);
947
948         if (!ret) {
949                 struct seq_file *sf = file->private_data;
950                 sf->private = PDE(inode);
951         }
952         return ret;
953 }
954
955 static const struct file_operations dl_file_ops = {
956         .owner   = THIS_MODULE,
957         .open    = dl_proc_open,
958         .read    = seq_read,
959         .llseek  = seq_lseek,
960         .release = seq_release
961 };
962
963 static int __init hashlimit_mt_init(void)
964 {
965         int err;
966
967         err = xt_register_matches(hashlimit_mt_reg,
968               ARRAY_SIZE(hashlimit_mt_reg));
969         if (err < 0)
970                 goto err1;
971
972         err = -ENOMEM;
973         hashlimit_cachep = kmem_cache_create("xt_hashlimit",
974                                             sizeof(struct dsthash_ent), 0, 0,
975                                             NULL);
976         if (!hashlimit_cachep) {
977                 printk(KERN_ERR "xt_hashlimit: unable to create slab cache\n");
978                 goto err2;
979         }
980         hashlimit_procdir4 = proc_mkdir("ipt_hashlimit", init_net.proc_net);
981         if (!hashlimit_procdir4) {
982                 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
983                                 "entry\n");
984                 goto err3;
985         }
986         err = 0;
987 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
988         hashlimit_procdir6 = proc_mkdir("ip6t_hashlimit", init_net.proc_net);
989         if (!hashlimit_procdir6) {
990                 printk(KERN_ERR "xt_hashlimit: unable to create proc dir "
991                                 "entry\n");
992                 err = -ENOMEM;
993         }
994 #endif
995         if (!err)
996                 return 0;
997         remove_proc_entry("ipt_hashlimit", init_net.proc_net);
998 err3:
999         kmem_cache_destroy(hashlimit_cachep);
1000 err2:
1001         xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1002 err1:
1003         return err;
1004
1005 }
1006
1007 static void __exit hashlimit_mt_exit(void)
1008 {
1009         remove_proc_entry("ipt_hashlimit", init_net.proc_net);
1010 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
1011         remove_proc_entry("ip6t_hashlimit", init_net.proc_net);
1012 #endif
1013         kmem_cache_destroy(hashlimit_cachep);
1014         xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1015 }
1016
1017 module_init(hashlimit_mt_init);
1018 module_exit(hashlimit_mt_exit);