IPv6: fix race between cleanup and add/delete address
[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 #include <net/netns/generic.h>
30
31 #include <linux/netfilter/x_tables.h>
32 #include <linux/netfilter_ipv4/ip_tables.h>
33 #include <linux/netfilter_ipv6/ip6_tables.h>
34 #include <linux/netfilter/xt_hashlimit.h>
35 #include <linux/mutex.h>
36
37 MODULE_LICENSE("GPL");
38 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
39 MODULE_AUTHOR("Jan Engelhardt <jengelh@computergmbh.de>");
40 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
41 MODULE_ALIAS("ipt_hashlimit");
42 MODULE_ALIAS("ip6t_hashlimit");
43
44 struct hashlimit_net {
45         struct hlist_head       htables;
46         struct proc_dir_entry   *ipt_hashlimit;
47         struct proc_dir_entry   *ip6t_hashlimit;
48 };
49
50 static int hashlimit_net_id;
51 static inline struct hashlimit_net *hashlimit_pernet(struct net *net)
52 {
53         return net_generic(net, hashlimit_net_id);
54 }
55
56 /* need to declare this at the top */
57 static const struct file_operations dl_file_ops;
58
59 /* hash table crap */
60 struct dsthash_dst {
61         union {
62                 struct {
63                         __be32 src;
64                         __be32 dst;
65                 } ip;
66 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
67                 struct {
68                         __be32 src[4];
69                         __be32 dst[4];
70                 } ip6;
71 #endif
72         };
73         __be16 src_port;
74         __be16 dst_port;
75 };
76
77 struct dsthash_ent {
78         /* static / read-only parts in the beginning */
79         struct hlist_node node;
80         struct dsthash_dst dst;
81
82         /* modified structure members in the end */
83         unsigned long expires;          /* precalculated expiry time */
84         struct {
85                 unsigned long prev;     /* last modification */
86                 u_int32_t credit;
87                 u_int32_t credit_cap, cost;
88         } rateinfo;
89 };
90
91 struct xt_hashlimit_htable {
92         struct hlist_node node;         /* global list of all htables */
93         int use;
94         u_int8_t family;
95         bool rnd_initialized;
96
97         struct hashlimit_cfg1 cfg;      /* config */
98
99         /* used internally */
100         spinlock_t lock;                /* lock for list_head */
101         u_int32_t rnd;                  /* random seed for hash */
102         unsigned int count;             /* number entries in table */
103         struct timer_list timer;        /* timer for gc */
104
105         /* seq_file stuff */
106         struct proc_dir_entry *pde;
107         struct net *net;
108
109         struct hlist_head hash[0];      /* hashtable itself */
110 };
111
112 static DEFINE_MUTEX(hashlimit_mutex);   /* protects htables list */
113 static struct kmem_cache *hashlimit_cachep __read_mostly;
114
115 static inline bool dst_cmp(const struct dsthash_ent *ent,
116                            const struct dsthash_dst *b)
117 {
118         return !memcmp(&ent->dst, b, sizeof(ent->dst));
119 }
120
121 static u_int32_t
122 hash_dst(const struct xt_hashlimit_htable *ht, const struct dsthash_dst *dst)
123 {
124         u_int32_t hash = jhash2((const u32 *)dst,
125                                 sizeof(*dst)/sizeof(u32),
126                                 ht->rnd);
127         /*
128          * Instead of returning hash % ht->cfg.size (implying a divide)
129          * we return the high 32 bits of the (hash * ht->cfg.size) that will
130          * give results between [0 and cfg.size-1] and same hash distribution,
131          * but using a multiply, less expensive than a divide
132          */
133         return ((u64)hash * ht->cfg.size) >> 32;
134 }
135
136 static struct dsthash_ent *
137 dsthash_find(const struct xt_hashlimit_htable *ht,
138              const struct dsthash_dst *dst)
139 {
140         struct dsthash_ent *ent;
141         struct hlist_node *pos;
142         u_int32_t hash = hash_dst(ht, dst);
143
144         if (!hlist_empty(&ht->hash[hash])) {
145                 hlist_for_each_entry(ent, pos, &ht->hash[hash], node)
146                         if (dst_cmp(ent, dst))
147                                 return ent;
148         }
149         return NULL;
150 }
151
152 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
153 static struct dsthash_ent *
154 dsthash_alloc_init(struct xt_hashlimit_htable *ht,
155                    const struct dsthash_dst *dst)
156 {
157         struct dsthash_ent *ent;
158
159         /* initialize hash with random val at the time we allocate
160          * the first hashtable entry */
161         if (!ht->rnd_initialized) {
162                 get_random_bytes(&ht->rnd, sizeof(ht->rnd));
163                 ht->rnd_initialized = true;
164         }
165
166         if (ht->cfg.max && ht->count >= ht->cfg.max) {
167                 /* FIXME: do something. question is what.. */
168                 if (net_ratelimit())
169                         printk(KERN_WARNING
170                                 "xt_hashlimit: max count of %u reached\n",
171                                 ht->cfg.max);
172                 return NULL;
173         }
174
175         ent = kmem_cache_alloc(hashlimit_cachep, GFP_ATOMIC);
176         if (!ent) {
177                 if (net_ratelimit())
178                         printk(KERN_ERR
179                                 "xt_hashlimit: can't allocate dsthash_ent\n");
180                 return NULL;
181         }
182         memcpy(&ent->dst, dst, sizeof(ent->dst));
183
184         hlist_add_head(&ent->node, &ht->hash[hash_dst(ht, dst)]);
185         ht->count++;
186         return ent;
187 }
188
189 static inline void
190 dsthash_free(struct xt_hashlimit_htable *ht, struct dsthash_ent *ent)
191 {
192         hlist_del(&ent->node);
193         kmem_cache_free(hashlimit_cachep, ent);
194         ht->count--;
195 }
196 static void htable_gc(unsigned long htlong);
197
198 static int htable_create_v0(struct net *net, struct xt_hashlimit_info *minfo, u_int8_t family)
199 {
200         struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
201         struct xt_hashlimit_htable *hinfo;
202         unsigned int size;
203         unsigned int i;
204
205         if (minfo->cfg.size)
206                 size = minfo->cfg.size;
207         else {
208                 size = ((totalram_pages << PAGE_SHIFT) / 16384) /
209                        sizeof(struct list_head);
210                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
211                         size = 8192;
212                 if (size < 16)
213                         size = 16;
214         }
215         /* FIXME: don't use vmalloc() here or anywhere else -HW */
216         hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
217                         sizeof(struct list_head) * size);
218         if (!hinfo) {
219                 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
220                 return -1;
221         }
222         minfo->hinfo = hinfo;
223
224         /* copy match config into hashtable config */
225         hinfo->cfg.mode        = minfo->cfg.mode;
226         hinfo->cfg.avg         = minfo->cfg.avg;
227         hinfo->cfg.burst       = minfo->cfg.burst;
228         hinfo->cfg.max         = minfo->cfg.max;
229         hinfo->cfg.gc_interval = minfo->cfg.gc_interval;
230         hinfo->cfg.expire      = minfo->cfg.expire;
231
232         if (family == NFPROTO_IPV4)
233                 hinfo->cfg.srcmask = hinfo->cfg.dstmask = 32;
234         else
235                 hinfo->cfg.srcmask = hinfo->cfg.dstmask = 128;
236
237         hinfo->cfg.size = size;
238         if (!hinfo->cfg.max)
239                 hinfo->cfg.max = 8 * hinfo->cfg.size;
240         else if (hinfo->cfg.max < hinfo->cfg.size)
241                 hinfo->cfg.max = hinfo->cfg.size;
242
243         for (i = 0; i < hinfo->cfg.size; i++)
244                 INIT_HLIST_HEAD(&hinfo->hash[i]);
245
246         hinfo->use = 1;
247         hinfo->count = 0;
248         hinfo->family = family;
249         hinfo->rnd_initialized = false;
250         spin_lock_init(&hinfo->lock);
251         hinfo->pde = proc_create_data(minfo->name, 0,
252                 (family == NFPROTO_IPV4) ?
253                 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
254                 &dl_file_ops, hinfo);
255         if (!hinfo->pde) {
256                 vfree(hinfo);
257                 return -1;
258         }
259         hinfo->net = net;
260
261         setup_timer(&hinfo->timer, htable_gc, (unsigned long )hinfo);
262         hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
263         add_timer(&hinfo->timer);
264
265         hlist_add_head(&hinfo->node, &hashlimit_net->htables);
266
267         return 0;
268 }
269
270 static int htable_create(struct net *net, struct xt_hashlimit_mtinfo1 *minfo,
271                          u_int8_t family)
272 {
273         struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
274         struct xt_hashlimit_htable *hinfo;
275         unsigned int size;
276         unsigned int i;
277
278         if (minfo->cfg.size) {
279                 size = minfo->cfg.size;
280         } else {
281                 size = (totalram_pages << PAGE_SHIFT) / 16384 /
282                        sizeof(struct list_head);
283                 if (totalram_pages > 1024 * 1024 * 1024 / PAGE_SIZE)
284                         size = 8192;
285                 if (size < 16)
286                         size = 16;
287         }
288         /* FIXME: don't use vmalloc() here or anywhere else -HW */
289         hinfo = vmalloc(sizeof(struct xt_hashlimit_htable) +
290                         sizeof(struct list_head) * size);
291         if (hinfo == NULL) {
292                 printk(KERN_ERR "xt_hashlimit: unable to create hashtable\n");
293                 return -1;
294         }
295         minfo->hinfo = hinfo;
296
297         /* copy match config into hashtable config */
298         memcpy(&hinfo->cfg, &minfo->cfg, sizeof(hinfo->cfg));
299         hinfo->cfg.size = size;
300         if (hinfo->cfg.max == 0)
301                 hinfo->cfg.max = 8 * hinfo->cfg.size;
302         else if (hinfo->cfg.max < hinfo->cfg.size)
303                 hinfo->cfg.max = hinfo->cfg.size;
304
305         for (i = 0; i < hinfo->cfg.size; i++)
306                 INIT_HLIST_HEAD(&hinfo->hash[i]);
307
308         hinfo->use = 1;
309         hinfo->count = 0;
310         hinfo->family = family;
311         hinfo->rnd_initialized = false;
312         spin_lock_init(&hinfo->lock);
313
314         hinfo->pde = proc_create_data(minfo->name, 0,
315                 (family == NFPROTO_IPV4) ?
316                 hashlimit_net->ipt_hashlimit : hashlimit_net->ip6t_hashlimit,
317                 &dl_file_ops, hinfo);
318         if (hinfo->pde == NULL) {
319                 vfree(hinfo);
320                 return -1;
321         }
322         hinfo->net = net;
323
324         setup_timer(&hinfo->timer, htable_gc, (unsigned long)hinfo);
325         hinfo->timer.expires = jiffies + msecs_to_jiffies(hinfo->cfg.gc_interval);
326         add_timer(&hinfo->timer);
327
328         hlist_add_head(&hinfo->node, &hashlimit_net->htables);
329
330         return 0;
331 }
332
333 static bool select_all(const struct xt_hashlimit_htable *ht,
334                        const struct dsthash_ent *he)
335 {
336         return 1;
337 }
338
339 static bool select_gc(const struct xt_hashlimit_htable *ht,
340                       const struct dsthash_ent *he)
341 {
342         return time_after_eq(jiffies, he->expires);
343 }
344
345 static void htable_selective_cleanup(struct xt_hashlimit_htable *ht,
346                         bool (*select)(const struct xt_hashlimit_htable *ht,
347                                       const struct dsthash_ent *he))
348 {
349         unsigned int i;
350
351         /* lock hash table and iterate over it */
352         spin_lock_bh(&ht->lock);
353         for (i = 0; i < ht->cfg.size; i++) {
354                 struct dsthash_ent *dh;
355                 struct hlist_node *pos, *n;
356                 hlist_for_each_entry_safe(dh, pos, n, &ht->hash[i], node) {
357                         if ((*select)(ht, dh))
358                                 dsthash_free(ht, dh);
359                 }
360         }
361         spin_unlock_bh(&ht->lock);
362 }
363
364 /* hash table garbage collector, run by timer */
365 static void htable_gc(unsigned long htlong)
366 {
367         struct xt_hashlimit_htable *ht = (struct xt_hashlimit_htable *)htlong;
368
369         htable_selective_cleanup(ht, select_gc);
370
371         /* re-add the timer accordingly */
372         ht->timer.expires = jiffies + msecs_to_jiffies(ht->cfg.gc_interval);
373         add_timer(&ht->timer);
374 }
375
376 static void htable_destroy(struct xt_hashlimit_htable *hinfo)
377 {
378         struct hashlimit_net *hashlimit_net = hashlimit_pernet(hinfo->net);
379         struct proc_dir_entry *parent;
380
381         del_timer_sync(&hinfo->timer);
382
383         if (hinfo->family == NFPROTO_IPV4)
384                 parent = hashlimit_net->ipt_hashlimit;
385         else
386                 parent = hashlimit_net->ip6t_hashlimit;
387         remove_proc_entry(hinfo->pde->name, parent);
388         htable_selective_cleanup(hinfo, select_all);
389         vfree(hinfo);
390 }
391
392 static struct xt_hashlimit_htable *htable_find_get(struct net *net,
393                                                    const char *name,
394                                                    u_int8_t family)
395 {
396         struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
397         struct xt_hashlimit_htable *hinfo;
398         struct hlist_node *pos;
399
400         hlist_for_each_entry(hinfo, pos, &hashlimit_net->htables, node) {
401                 if (!strcmp(name, hinfo->pde->name) &&
402                     hinfo->family == family) {
403                         hinfo->use++;
404                         return hinfo;
405                 }
406         }
407         return NULL;
408 }
409
410 static void htable_put(struct xt_hashlimit_htable *hinfo)
411 {
412         mutex_lock(&hashlimit_mutex);
413         if (--hinfo->use == 0) {
414                 hlist_del(&hinfo->node);
415                 htable_destroy(hinfo);
416         }
417         mutex_unlock(&hashlimit_mutex);
418 }
419
420 /* The algorithm used is the Simple Token Bucket Filter (TBF)
421  * see net/sched/sch_tbf.c in the linux source tree
422  */
423
424 /* Rusty: This is my (non-mathematically-inclined) understanding of
425    this algorithm.  The `average rate' in jiffies becomes your initial
426    amount of credit `credit' and the most credit you can ever have
427    `credit_cap'.  The `peak rate' becomes the cost of passing the
428    test, `cost'.
429
430    `prev' tracks the last packet hit: you gain one credit per jiffy.
431    If you get credit balance more than this, the extra credit is
432    discarded.  Every time the match passes, you lose `cost' credits;
433    if you don't have that many, the test fails.
434
435    See Alexey's formal explanation in net/sched/sch_tbf.c.
436
437    To get the maximum range, we multiply by this factor (ie. you get N
438    credits per jiffy).  We want to allow a rate as low as 1 per day
439    (slowest userspace tool allows), which means
440    CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
441 */
442 #define MAX_CPJ (0xFFFFFFFF / (HZ*60*60*24))
443
444 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
445  * us the power of 2 below the theoretical max, so GCC simply does a
446  * shift. */
447 #define _POW2_BELOW2(x) ((x)|((x)>>1))
448 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
449 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
450 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
451 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
452 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
453
454 #define CREDITS_PER_JIFFY POW2_BELOW32(MAX_CPJ)
455
456 /* Precision saver. */
457 static inline u_int32_t
458 user2credits(u_int32_t user)
459 {
460         /* If multiplying would overflow... */
461         if (user > 0xFFFFFFFF / (HZ*CREDITS_PER_JIFFY))
462                 /* Divide first. */
463                 return (user / XT_HASHLIMIT_SCALE) * HZ * CREDITS_PER_JIFFY;
464
465         return (user * HZ * CREDITS_PER_JIFFY) / XT_HASHLIMIT_SCALE;
466 }
467
468 static inline void rateinfo_recalc(struct dsthash_ent *dh, unsigned long now)
469 {
470         dh->rateinfo.credit += (now - dh->rateinfo.prev) * CREDITS_PER_JIFFY;
471         if (dh->rateinfo.credit > dh->rateinfo.credit_cap)
472                 dh->rateinfo.credit = dh->rateinfo.credit_cap;
473         dh->rateinfo.prev = now;
474 }
475
476 static inline __be32 maskl(__be32 a, unsigned int l)
477 {
478         return l ? htonl(ntohl(a) & ~0 << (32 - l)) : 0;
479 }
480
481 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
482 static void hashlimit_ipv6_mask(__be32 *i, unsigned int p)
483 {
484         switch (p) {
485         case 0 ... 31:
486                 i[0] = maskl(i[0], p);
487                 i[1] = i[2] = i[3] = 0;
488                 break;
489         case 32 ... 63:
490                 i[1] = maskl(i[1], p - 32);
491                 i[2] = i[3] = 0;
492                 break;
493         case 64 ... 95:
494                 i[2] = maskl(i[2], p - 64);
495                 i[3] = 0;
496         case 96 ... 127:
497                 i[3] = maskl(i[3], p - 96);
498                 break;
499         case 128:
500                 break;
501         }
502 }
503 #endif
504
505 static int
506 hashlimit_init_dst(const struct xt_hashlimit_htable *hinfo,
507                    struct dsthash_dst *dst,
508                    const struct sk_buff *skb, unsigned int protoff)
509 {
510         __be16 _ports[2], *ports;
511         u8 nexthdr;
512
513         memset(dst, 0, sizeof(*dst));
514
515         switch (hinfo->family) {
516         case NFPROTO_IPV4:
517                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP)
518                         dst->ip.dst = maskl(ip_hdr(skb)->daddr,
519                                       hinfo->cfg.dstmask);
520                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP)
521                         dst->ip.src = maskl(ip_hdr(skb)->saddr,
522                                       hinfo->cfg.srcmask);
523
524                 if (!(hinfo->cfg.mode &
525                       (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
526                         return 0;
527                 nexthdr = ip_hdr(skb)->protocol;
528                 break;
529 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
530         case NFPROTO_IPV6:
531                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DIP) {
532                         memcpy(&dst->ip6.dst, &ipv6_hdr(skb)->daddr,
533                                sizeof(dst->ip6.dst));
534                         hashlimit_ipv6_mask(dst->ip6.dst, hinfo->cfg.dstmask);
535                 }
536                 if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SIP) {
537                         memcpy(&dst->ip6.src, &ipv6_hdr(skb)->saddr,
538                                sizeof(dst->ip6.src));
539                         hashlimit_ipv6_mask(dst->ip6.src, hinfo->cfg.srcmask);
540                 }
541
542                 if (!(hinfo->cfg.mode &
543                       (XT_HASHLIMIT_HASH_DPT | XT_HASHLIMIT_HASH_SPT)))
544                         return 0;
545                 nexthdr = ipv6_hdr(skb)->nexthdr;
546                 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
547                 if ((int)protoff < 0)
548                         return -1;
549                 break;
550 #endif
551         default:
552                 BUG();
553                 return 0;
554         }
555
556         switch (nexthdr) {
557         case IPPROTO_TCP:
558         case IPPROTO_UDP:
559         case IPPROTO_UDPLITE:
560         case IPPROTO_SCTP:
561         case IPPROTO_DCCP:
562                 ports = skb_header_pointer(skb, protoff, sizeof(_ports),
563                                            &_ports);
564                 break;
565         default:
566                 _ports[0] = _ports[1] = 0;
567                 ports = _ports;
568                 break;
569         }
570         if (!ports)
571                 return -1;
572         if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_SPT)
573                 dst->src_port = ports[0];
574         if (hinfo->cfg.mode & XT_HASHLIMIT_HASH_DPT)
575                 dst->dst_port = ports[1];
576         return 0;
577 }
578
579 static bool
580 hashlimit_mt_v0(const struct sk_buff *skb, const struct xt_match_param *par)
581 {
582         const struct xt_hashlimit_info *r = par->matchinfo;
583         struct xt_hashlimit_htable *hinfo = r->hinfo;
584         unsigned long now = jiffies;
585         struct dsthash_ent *dh;
586         struct dsthash_dst dst;
587
588         if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
589                 goto hotdrop;
590
591         spin_lock_bh(&hinfo->lock);
592         dh = dsthash_find(hinfo, &dst);
593         if (!dh) {
594                 dh = dsthash_alloc_init(hinfo, &dst);
595                 if (!dh) {
596                         spin_unlock_bh(&hinfo->lock);
597                         goto hotdrop;
598                 }
599
600                 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
601                 dh->rateinfo.prev = jiffies;
602                 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
603                                                    hinfo->cfg.burst);
604                 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
605                                                        hinfo->cfg.burst);
606                 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
607         } else {
608                 /* update expiration timeout */
609                 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
610                 rateinfo_recalc(dh, now);
611         }
612
613         if (dh->rateinfo.credit >= dh->rateinfo.cost) {
614                 /* We're underlimit. */
615                 dh->rateinfo.credit -= dh->rateinfo.cost;
616                 spin_unlock_bh(&hinfo->lock);
617                 return true;
618         }
619
620         spin_unlock_bh(&hinfo->lock);
621
622         /* default case: we're overlimit, thus don't match */
623         return false;
624
625 hotdrop:
626         *par->hotdrop = true;
627         return false;
628 }
629
630 static bool
631 hashlimit_mt(const struct sk_buff *skb, const struct xt_match_param *par)
632 {
633         const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
634         struct xt_hashlimit_htable *hinfo = info->hinfo;
635         unsigned long now = jiffies;
636         struct dsthash_ent *dh;
637         struct dsthash_dst dst;
638
639         if (hashlimit_init_dst(hinfo, &dst, skb, par->thoff) < 0)
640                 goto hotdrop;
641
642         spin_lock_bh(&hinfo->lock);
643         dh = dsthash_find(hinfo, &dst);
644         if (dh == NULL) {
645                 dh = dsthash_alloc_init(hinfo, &dst);
646                 if (dh == NULL) {
647                         spin_unlock_bh(&hinfo->lock);
648                         goto hotdrop;
649                 }
650
651                 dh->expires = jiffies + msecs_to_jiffies(hinfo->cfg.expire);
652                 dh->rateinfo.prev = jiffies;
653                 dh->rateinfo.credit = user2credits(hinfo->cfg.avg *
654                                       hinfo->cfg.burst);
655                 dh->rateinfo.credit_cap = user2credits(hinfo->cfg.avg *
656                                           hinfo->cfg.burst);
657                 dh->rateinfo.cost = user2credits(hinfo->cfg.avg);
658         } else {
659                 /* update expiration timeout */
660                 dh->expires = now + msecs_to_jiffies(hinfo->cfg.expire);
661                 rateinfo_recalc(dh, now);
662         }
663
664         if (dh->rateinfo.credit >= dh->rateinfo.cost) {
665                 /* below the limit */
666                 dh->rateinfo.credit -= dh->rateinfo.cost;
667                 spin_unlock_bh(&hinfo->lock);
668                 return !(info->cfg.mode & XT_HASHLIMIT_INVERT);
669         }
670
671         spin_unlock_bh(&hinfo->lock);
672         /* default match is underlimit - so over the limit, we need to invert */
673         return info->cfg.mode & XT_HASHLIMIT_INVERT;
674
675  hotdrop:
676         *par->hotdrop = true;
677         return false;
678 }
679
680 static bool hashlimit_mt_check_v0(const struct xt_mtchk_param *par)
681 {
682         struct net *net = par->net;
683         struct xt_hashlimit_info *r = par->matchinfo;
684
685         /* Check for overflow. */
686         if (r->cfg.burst == 0 ||
687             user2credits(r->cfg.avg * r->cfg.burst) < user2credits(r->cfg.avg)) {
688                 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
689                        r->cfg.avg, r->cfg.burst);
690                 return false;
691         }
692         if (r->cfg.mode == 0 ||
693             r->cfg.mode > (XT_HASHLIMIT_HASH_DPT |
694                            XT_HASHLIMIT_HASH_DIP |
695                            XT_HASHLIMIT_HASH_SIP |
696                            XT_HASHLIMIT_HASH_SPT))
697                 return false;
698         if (!r->cfg.gc_interval)
699                 return false;
700         if (!r->cfg.expire)
701                 return false;
702         if (r->name[sizeof(r->name) - 1] != '\0')
703                 return false;
704
705         mutex_lock(&hashlimit_mutex);
706         r->hinfo = htable_find_get(net, r->name, par->match->family);
707         if (!r->hinfo && htable_create_v0(net, r, par->match->family) != 0) {
708                 mutex_unlock(&hashlimit_mutex);
709                 return false;
710         }
711         mutex_unlock(&hashlimit_mutex);
712
713         return true;
714 }
715
716 static bool hashlimit_mt_check(const struct xt_mtchk_param *par)
717 {
718         struct net *net = par->net;
719         struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
720
721         /* Check for overflow. */
722         if (info->cfg.burst == 0 ||
723             user2credits(info->cfg.avg * info->cfg.burst) <
724             user2credits(info->cfg.avg)) {
725                 printk(KERN_ERR "xt_hashlimit: overflow, try lower: %u/%u\n",
726                        info->cfg.avg, info->cfg.burst);
727                 return false;
728         }
729         if (info->cfg.gc_interval == 0 || info->cfg.expire == 0)
730                 return false;
731         if (info->name[sizeof(info->name)-1] != '\0')
732                 return false;
733         if (par->match->family == NFPROTO_IPV4) {
734                 if (info->cfg.srcmask > 32 || info->cfg.dstmask > 32)
735                         return false;
736         } else {
737                 if (info->cfg.srcmask > 128 || info->cfg.dstmask > 128)
738                         return false;
739         }
740
741         mutex_lock(&hashlimit_mutex);
742         info->hinfo = htable_find_get(net, info->name, par->match->family);
743         if (!info->hinfo && htable_create(net, info, par->match->family) != 0) {
744                 mutex_unlock(&hashlimit_mutex);
745                 return false;
746         }
747         mutex_unlock(&hashlimit_mutex);
748         return true;
749 }
750
751 static void
752 hashlimit_mt_destroy_v0(const struct xt_mtdtor_param *par)
753 {
754         const struct xt_hashlimit_info *r = par->matchinfo;
755
756         htable_put(r->hinfo);
757 }
758
759 static void hashlimit_mt_destroy(const struct xt_mtdtor_param *par)
760 {
761         const struct xt_hashlimit_mtinfo1 *info = par->matchinfo;
762
763         htable_put(info->hinfo);
764 }
765
766 #ifdef CONFIG_COMPAT
767 struct compat_xt_hashlimit_info {
768         char name[IFNAMSIZ];
769         struct hashlimit_cfg cfg;
770         compat_uptr_t hinfo;
771         compat_uptr_t master;
772 };
773
774 static void hashlimit_mt_compat_from_user(void *dst, const void *src)
775 {
776         int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
777
778         memcpy(dst, src, off);
779         memset(dst + off, 0, sizeof(struct compat_xt_hashlimit_info) - off);
780 }
781
782 static int hashlimit_mt_compat_to_user(void __user *dst, const void *src)
783 {
784         int off = offsetof(struct compat_xt_hashlimit_info, hinfo);
785
786         return copy_to_user(dst, src, off) ? -EFAULT : 0;
787 }
788 #endif
789
790 static struct xt_match hashlimit_mt_reg[] __read_mostly = {
791         {
792                 .name           = "hashlimit",
793                 .revision       = 0,
794                 .family         = NFPROTO_IPV4,
795                 .match          = hashlimit_mt_v0,
796                 .matchsize      = sizeof(struct xt_hashlimit_info),
797 #ifdef CONFIG_COMPAT
798                 .compatsize     = sizeof(struct compat_xt_hashlimit_info),
799                 .compat_from_user = hashlimit_mt_compat_from_user,
800                 .compat_to_user = hashlimit_mt_compat_to_user,
801 #endif
802                 .checkentry     = hashlimit_mt_check_v0,
803                 .destroy        = hashlimit_mt_destroy_v0,
804                 .me             = THIS_MODULE
805         },
806         {
807                 .name           = "hashlimit",
808                 .revision       = 1,
809                 .family         = NFPROTO_IPV4,
810                 .match          = hashlimit_mt,
811                 .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
812                 .checkentry     = hashlimit_mt_check,
813                 .destroy        = hashlimit_mt_destroy,
814                 .me             = THIS_MODULE,
815         },
816 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
817         {
818                 .name           = "hashlimit",
819                 .family         = NFPROTO_IPV6,
820                 .match          = hashlimit_mt_v0,
821                 .matchsize      = sizeof(struct xt_hashlimit_info),
822 #ifdef CONFIG_COMPAT
823                 .compatsize     = sizeof(struct compat_xt_hashlimit_info),
824                 .compat_from_user = hashlimit_mt_compat_from_user,
825                 .compat_to_user = hashlimit_mt_compat_to_user,
826 #endif
827                 .checkentry     = hashlimit_mt_check_v0,
828                 .destroy        = hashlimit_mt_destroy_v0,
829                 .me             = THIS_MODULE
830         },
831         {
832                 .name           = "hashlimit",
833                 .revision       = 1,
834                 .family         = NFPROTO_IPV6,
835                 .match          = hashlimit_mt,
836                 .matchsize      = sizeof(struct xt_hashlimit_mtinfo1),
837                 .checkentry     = hashlimit_mt_check,
838                 .destroy        = hashlimit_mt_destroy,
839                 .me             = THIS_MODULE,
840         },
841 #endif
842 };
843
844 /* PROC stuff */
845 static void *dl_seq_start(struct seq_file *s, loff_t *pos)
846         __acquires(htable->lock)
847 {
848         struct xt_hashlimit_htable *htable = s->private;
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 xt_hashlimit_htable *htable = s->private;
866         unsigned int *bucket = (unsigned int *)v;
867
868         *pos = ++(*bucket);
869         if (*pos >= htable->cfg.size) {
870                 kfree(v);
871                 return NULL;
872         }
873         return bucket;
874 }
875
876 static void dl_seq_stop(struct seq_file *s, void *v)
877         __releases(htable->lock)
878 {
879         struct xt_hashlimit_htable *htable = s->private;
880         unsigned int *bucket = (unsigned int *)v;
881
882         kfree(bucket);
883         spin_unlock_bh(&htable->lock);
884 }
885
886 static int dl_seq_real_show(struct dsthash_ent *ent, u_int8_t family,
887                                    struct seq_file *s)
888 {
889         /* recalculate to show accurate numbers */
890         rateinfo_recalc(ent, jiffies);
891
892         switch (family) {
893         case NFPROTO_IPV4:
894                 return seq_printf(s, "%ld %pI4:%u->%pI4:%u %u %u %u\n",
895                                  (long)(ent->expires - jiffies)/HZ,
896                                  &ent->dst.ip.src,
897                                  ntohs(ent->dst.src_port),
898                                  &ent->dst.ip.dst,
899                                  ntohs(ent->dst.dst_port),
900                                  ent->rateinfo.credit, ent->rateinfo.credit_cap,
901                                  ent->rateinfo.cost);
902 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
903         case NFPROTO_IPV6:
904                 return seq_printf(s, "%ld %pI6:%u->%pI6:%u %u %u %u\n",
905                                  (long)(ent->expires - jiffies)/HZ,
906                                  &ent->dst.ip6.src,
907                                  ntohs(ent->dst.src_port),
908                                  &ent->dst.ip6.dst,
909                                  ntohs(ent->dst.dst_port),
910                                  ent->rateinfo.credit, ent->rateinfo.credit_cap,
911                                  ent->rateinfo.cost);
912 #endif
913         default:
914                 BUG();
915                 return 0;
916         }
917 }
918
919 static int dl_seq_show(struct seq_file *s, void *v)
920 {
921         struct xt_hashlimit_htable *htable = s->private;
922         unsigned int *bucket = (unsigned int *)v;
923         struct dsthash_ent *ent;
924         struct hlist_node *pos;
925
926         if (!hlist_empty(&htable->hash[*bucket])) {
927                 hlist_for_each_entry(ent, pos, &htable->hash[*bucket], node)
928                         if (dl_seq_real_show(ent, htable->family, s))
929                                 return -1;
930         }
931         return 0;
932 }
933
934 static const struct seq_operations dl_seq_ops = {
935         .start = dl_seq_start,
936         .next  = dl_seq_next,
937         .stop  = dl_seq_stop,
938         .show  = dl_seq_show
939 };
940
941 static int dl_proc_open(struct inode *inode, struct file *file)
942 {
943         int ret = seq_open(file, &dl_seq_ops);
944
945         if (!ret) {
946                 struct seq_file *sf = file->private_data;
947                 sf->private = PDE(inode)->data;
948         }
949         return ret;
950 }
951
952 static const struct file_operations dl_file_ops = {
953         .owner   = THIS_MODULE,
954         .open    = dl_proc_open,
955         .read    = seq_read,
956         .llseek  = seq_lseek,
957         .release = seq_release
958 };
959
960 static int __net_init hashlimit_proc_net_init(struct net *net)
961 {
962         struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
963
964         hashlimit_net->ipt_hashlimit = proc_mkdir("ipt_hashlimit", net->proc_net);
965         if (!hashlimit_net->ipt_hashlimit)
966                 return -ENOMEM;
967 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
968         hashlimit_net->ip6t_hashlimit = proc_mkdir("ip6t_hashlimit", net->proc_net);
969         if (!hashlimit_net->ip6t_hashlimit) {
970                 proc_net_remove(net, "ipt_hashlimit");
971                 return -ENOMEM;
972         }
973 #endif
974         return 0;
975 }
976
977 static void __net_exit hashlimit_proc_net_exit(struct net *net)
978 {
979         proc_net_remove(net, "ipt_hashlimit");
980 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
981         proc_net_remove(net, "ip6t_hashlimit");
982 #endif
983 }
984
985 static int __net_init hashlimit_net_init(struct net *net)
986 {
987         struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
988
989         INIT_HLIST_HEAD(&hashlimit_net->htables);
990         return hashlimit_proc_net_init(net);
991 }
992
993 static void __net_exit hashlimit_net_exit(struct net *net)
994 {
995         struct hashlimit_net *hashlimit_net = hashlimit_pernet(net);
996
997         BUG_ON(!hlist_empty(&hashlimit_net->htables));
998         hashlimit_proc_net_exit(net);
999 }
1000
1001 static struct pernet_operations hashlimit_net_ops = {
1002         .init   = hashlimit_net_init,
1003         .exit   = hashlimit_net_exit,
1004         .id     = &hashlimit_net_id,
1005         .size   = sizeof(struct hashlimit_net),
1006 };
1007
1008 static int __init hashlimit_mt_init(void)
1009 {
1010         int err;
1011
1012         err = register_pernet_subsys(&hashlimit_net_ops);
1013         if (err < 0)
1014                 return err;
1015         err = xt_register_matches(hashlimit_mt_reg,
1016               ARRAY_SIZE(hashlimit_mt_reg));
1017         if (err < 0)
1018                 goto err1;
1019
1020         err = -ENOMEM;
1021         hashlimit_cachep = kmem_cache_create("xt_hashlimit",
1022                                             sizeof(struct dsthash_ent), 0, 0,
1023                                             NULL);
1024         if (!hashlimit_cachep) {
1025                 printk(KERN_ERR "xt_hashlimit: unable to create slab cache\n");
1026                 goto err2;
1027         }
1028         return 0;
1029
1030 err2:
1031         xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1032 err1:
1033         unregister_pernet_subsys(&hashlimit_net_ops);
1034         return err;
1035
1036 }
1037
1038 static void __exit hashlimit_mt_exit(void)
1039 {
1040         kmem_cache_destroy(hashlimit_cachep);
1041         xt_unregister_matches(hashlimit_mt_reg, ARRAY_SIZE(hashlimit_mt_reg));
1042         unregister_pernet_subsys(&hashlimit_net_ops);
1043 }
1044
1045 module_init(hashlimit_mt_init);
1046 module_exit(hashlimit_mt_exit);