[NETFILTER]: x_tables: add missing try to load conntrack from match/targets
[linux-2.6] / net / ipv4 / netfilter / arp_tables.c
1 /*
2  * Packet matching code for ARP packets.
3  *
4  * Based heavily, if not almost entirely, upon ip_tables.c framework.
5  *
6  * Some ARP specific bits are:
7  *
8  * Copyright (C) 2002 David S. Miller (davem@redhat.com)
9  *
10  */
11
12 #include <linux/kernel.h>
13 #include <linux/skbuff.h>
14 #include <linux/netdevice.h>
15 #include <linux/capability.h>
16 #include <linux/if_arp.h>
17 #include <linux/kmod.h>
18 #include <linux/vmalloc.h>
19 #include <linux/proc_fs.h>
20 #include <linux/module.h>
21 #include <linux/init.h>
22
23 #include <asm/uaccess.h>
24 #include <linux/mutex.h>
25
26 #include <linux/netfilter/x_tables.h>
27 #include <linux/netfilter_arp/arp_tables.h>
28
29 MODULE_LICENSE("GPL");
30 MODULE_AUTHOR("David S. Miller <davem@redhat.com>");
31 MODULE_DESCRIPTION("arptables core");
32
33 /*#define DEBUG_ARP_TABLES*/
34 /*#define DEBUG_ARP_TABLES_USER*/
35
36 #ifdef DEBUG_ARP_TABLES
37 #define dprintf(format, args...)  printk(format , ## args)
38 #else
39 #define dprintf(format, args...)
40 #endif
41
42 #ifdef DEBUG_ARP_TABLES_USER
43 #define duprintf(format, args...) printk(format , ## args)
44 #else
45 #define duprintf(format, args...)
46 #endif
47
48 #ifdef CONFIG_NETFILTER_DEBUG
49 #define ARP_NF_ASSERT(x)                                        \
50 do {                                                            \
51         if (!(x))                                               \
52                 printk("ARP_NF_ASSERT: %s:%s:%u\n",             \
53                        __FUNCTION__, __FILE__, __LINE__);       \
54 } while(0)
55 #else
56 #define ARP_NF_ASSERT(x)
57 #endif
58
59 static inline int arp_devaddr_compare(const struct arpt_devaddr_info *ap,
60                                       char *hdr_addr, int len)
61 {
62         int i, ret;
63
64         if (len > ARPT_DEV_ADDR_LEN_MAX)
65                 len = ARPT_DEV_ADDR_LEN_MAX;
66
67         ret = 0;
68         for (i = 0; i < len; i++)
69                 ret |= (hdr_addr[i] ^ ap->addr[i]) & ap->mask[i];
70
71         return (ret != 0);
72 }
73
74 /* Returns whether packet matches rule or not. */
75 static inline int arp_packet_match(const struct arphdr *arphdr,
76                                    struct net_device *dev,
77                                    const char *indev,
78                                    const char *outdev,
79                                    const struct arpt_arp *arpinfo)
80 {
81         char *arpptr = (char *)(arphdr + 1);
82         char *src_devaddr, *tgt_devaddr;
83         __be32 src_ipaddr, tgt_ipaddr;
84         int i, ret;
85
86 #define FWINV(bool,invflg) ((bool) ^ !!(arpinfo->invflags & invflg))
87
88         if (FWINV((arphdr->ar_op & arpinfo->arpop_mask) != arpinfo->arpop,
89                   ARPT_INV_ARPOP)) {
90                 dprintf("ARP operation field mismatch.\n");
91                 dprintf("ar_op: %04x info->arpop: %04x info->arpop_mask: %04x\n",
92                         arphdr->ar_op, arpinfo->arpop, arpinfo->arpop_mask);
93                 return 0;
94         }
95
96         if (FWINV((arphdr->ar_hrd & arpinfo->arhrd_mask) != arpinfo->arhrd,
97                   ARPT_INV_ARPHRD)) {
98                 dprintf("ARP hardware address format mismatch.\n");
99                 dprintf("ar_hrd: %04x info->arhrd: %04x info->arhrd_mask: %04x\n",
100                         arphdr->ar_hrd, arpinfo->arhrd, arpinfo->arhrd_mask);
101                 return 0;
102         }
103
104         if (FWINV((arphdr->ar_pro & arpinfo->arpro_mask) != arpinfo->arpro,
105                   ARPT_INV_ARPPRO)) {
106                 dprintf("ARP protocol address format mismatch.\n");
107                 dprintf("ar_pro: %04x info->arpro: %04x info->arpro_mask: %04x\n",
108                         arphdr->ar_pro, arpinfo->arpro, arpinfo->arpro_mask);
109                 return 0;
110         }
111
112         if (FWINV((arphdr->ar_hln & arpinfo->arhln_mask) != arpinfo->arhln,
113                   ARPT_INV_ARPHLN)) {
114                 dprintf("ARP hardware address length mismatch.\n");
115                 dprintf("ar_hln: %02x info->arhln: %02x info->arhln_mask: %02x\n",
116                         arphdr->ar_hln, arpinfo->arhln, arpinfo->arhln_mask);
117                 return 0;
118         }
119
120         src_devaddr = arpptr;
121         arpptr += dev->addr_len;
122         memcpy(&src_ipaddr, arpptr, sizeof(u32));
123         arpptr += sizeof(u32);
124         tgt_devaddr = arpptr;
125         arpptr += dev->addr_len;
126         memcpy(&tgt_ipaddr, arpptr, sizeof(u32));
127
128         if (FWINV(arp_devaddr_compare(&arpinfo->src_devaddr, src_devaddr, dev->addr_len),
129                   ARPT_INV_SRCDEVADDR) ||
130             FWINV(arp_devaddr_compare(&arpinfo->tgt_devaddr, tgt_devaddr, dev->addr_len),
131                   ARPT_INV_TGTDEVADDR)) {
132                 dprintf("Source or target device address mismatch.\n");
133
134                 return 0;
135         }
136
137         if (FWINV((src_ipaddr & arpinfo->smsk.s_addr) != arpinfo->src.s_addr,
138                   ARPT_INV_SRCIP) ||
139             FWINV(((tgt_ipaddr & arpinfo->tmsk.s_addr) != arpinfo->tgt.s_addr),
140                   ARPT_INV_TGTIP)) {
141                 dprintf("Source or target IP address mismatch.\n");
142
143                 dprintf("SRC: %u.%u.%u.%u. Mask: %u.%u.%u.%u. Target: %u.%u.%u.%u.%s\n",
144                         NIPQUAD(src_ipaddr),
145                         NIPQUAD(arpinfo->smsk.s_addr),
146                         NIPQUAD(arpinfo->src.s_addr),
147                         arpinfo->invflags & ARPT_INV_SRCIP ? " (INV)" : "");
148                 dprintf("TGT: %u.%u.%u.%u Mask: %u.%u.%u.%u Target: %u.%u.%u.%u.%s\n",
149                         NIPQUAD(tgt_ipaddr),
150                         NIPQUAD(arpinfo->tmsk.s_addr),
151                         NIPQUAD(arpinfo->tgt.s_addr),
152                         arpinfo->invflags & ARPT_INV_TGTIP ? " (INV)" : "");
153                 return 0;
154         }
155
156         /* Look for ifname matches.  */
157         for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
158                 ret |= (indev[i] ^ arpinfo->iniface[i])
159                         & arpinfo->iniface_mask[i];
160         }
161
162         if (FWINV(ret != 0, ARPT_INV_VIA_IN)) {
163                 dprintf("VIA in mismatch (%s vs %s).%s\n",
164                         indev, arpinfo->iniface,
165                         arpinfo->invflags&ARPT_INV_VIA_IN ?" (INV)":"");
166                 return 0;
167         }
168
169         for (i = 0, ret = 0; i < IFNAMSIZ/sizeof(unsigned long); i++) {
170                 unsigned long odev;
171                 memcpy(&odev, outdev + i*sizeof(unsigned long),
172                        sizeof(unsigned long));
173                 ret |= (odev
174                         ^ ((const unsigned long *)arpinfo->outiface)[i])
175                         & ((const unsigned long *)arpinfo->outiface_mask)[i];
176         }
177
178         if (FWINV(ret != 0, ARPT_INV_VIA_OUT)) {
179                 dprintf("VIA out mismatch (%s vs %s).%s\n",
180                         outdev, arpinfo->outiface,
181                         arpinfo->invflags&ARPT_INV_VIA_OUT ?" (INV)":"");
182                 return 0;
183         }
184
185         return 1;
186 }
187
188 static inline int arp_checkentry(const struct arpt_arp *arp)
189 {
190         if (arp->flags & ~ARPT_F_MASK) {
191                 duprintf("Unknown flag bits set: %08X\n",
192                          arp->flags & ~ARPT_F_MASK);
193                 return 0;
194         }
195         if (arp->invflags & ~ARPT_INV_MASK) {
196                 duprintf("Unknown invflag bits set: %08X\n",
197                          arp->invflags & ~ARPT_INV_MASK);
198                 return 0;
199         }
200
201         return 1;
202 }
203
204 static unsigned int arpt_error(struct sk_buff **pskb,
205                                const struct net_device *in,
206                                const struct net_device *out,
207                                unsigned int hooknum,
208                                const struct xt_target *target,
209                                const void *targinfo)
210 {
211         if (net_ratelimit())
212                 printk("arp_tables: error: '%s'\n", (char *)targinfo);
213
214         return NF_DROP;
215 }
216
217 static inline struct arpt_entry *get_entry(void *base, unsigned int offset)
218 {
219         return (struct arpt_entry *)(base + offset);
220 }
221
222 unsigned int arpt_do_table(struct sk_buff **pskb,
223                            unsigned int hook,
224                            const struct net_device *in,
225                            const struct net_device *out,
226                            struct arpt_table *table)
227 {
228         static const char nulldevname[IFNAMSIZ];
229         unsigned int verdict = NF_DROP;
230         struct arphdr *arp;
231         int hotdrop = 0;
232         struct arpt_entry *e, *back;
233         const char *indev, *outdev;
234         void *table_base;
235         struct xt_table_info *private;
236
237         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
238         if (!pskb_may_pull((*pskb), (sizeof(struct arphdr) +
239                                      (2 * (*pskb)->dev->addr_len) +
240                                      (2 * sizeof(u32)))))
241                 return NF_DROP;
242
243         indev = in ? in->name : nulldevname;
244         outdev = out ? out->name : nulldevname;
245
246         read_lock_bh(&table->lock);
247         private = table->private;
248         table_base = (void *)private->entries[smp_processor_id()];
249         e = get_entry(table_base, private->hook_entry[hook]);
250         back = get_entry(table_base, private->underflow[hook]);
251
252         arp = (*pskb)->nh.arph;
253         do {
254                 if (arp_packet_match(arp, (*pskb)->dev, indev, outdev, &e->arp)) {
255                         struct arpt_entry_target *t;
256                         int hdr_len;
257
258                         hdr_len = sizeof(*arp) + (2 * sizeof(struct in_addr)) +
259                                 (2 * (*pskb)->dev->addr_len);
260                         ADD_COUNTER(e->counters, hdr_len, 1);
261
262                         t = arpt_get_target(e);
263
264                         /* Standard target? */
265                         if (!t->u.kernel.target->target) {
266                                 int v;
267
268                                 v = ((struct arpt_standard_target *)t)->verdict;
269                                 if (v < 0) {
270                                         /* Pop from stack? */
271                                         if (v != ARPT_RETURN) {
272                                                 verdict = (unsigned)(-v) - 1;
273                                                 break;
274                                         }
275                                         e = back;
276                                         back = get_entry(table_base,
277                                                          back->comefrom);
278                                         continue;
279                                 }
280                                 if (table_base + v
281                                     != (void *)e + e->next_offset) {
282                                         /* Save old back ptr in next entry */
283                                         struct arpt_entry *next
284                                                 = (void *)e + e->next_offset;
285                                         next->comefrom =
286                                                 (void *)back - table_base;
287
288                                         /* set back pointer to next entry */
289                                         back = next;
290                                 }
291
292                                 e = get_entry(table_base, v);
293                         } else {
294                                 /* Targets which reenter must return
295                                  * abs. verdicts
296                                  */
297                                 verdict = t->u.kernel.target->target(pskb,
298                                                                      in, out,
299                                                                      hook,
300                                                                      t->u.kernel.target,
301                                                                      t->data);
302
303                                 /* Target might have changed stuff. */
304                                 arp = (*pskb)->nh.arph;
305
306                                 if (verdict == ARPT_CONTINUE)
307                                         e = (void *)e + e->next_offset;
308                                 else
309                                         /* Verdict */
310                                         break;
311                         }
312                 } else {
313                         e = (void *)e + e->next_offset;
314                 }
315         } while (!hotdrop);
316         read_unlock_bh(&table->lock);
317
318         if (hotdrop)
319                 return NF_DROP;
320         else
321                 return verdict;
322 }
323
324 /* All zeroes == unconditional rule. */
325 static inline int unconditional(const struct arpt_arp *arp)
326 {
327         unsigned int i;
328
329         for (i = 0; i < sizeof(*arp)/sizeof(__u32); i++)
330                 if (((__u32 *)arp)[i])
331                         return 0;
332
333         return 1;
334 }
335
336 /* Figures out from what hook each rule can be called: returns 0 if
337  * there are loops.  Puts hook bitmask in comefrom.
338  */
339 static int mark_source_chains(struct xt_table_info *newinfo,
340                               unsigned int valid_hooks, void *entry0)
341 {
342         unsigned int hook;
343
344         /* No recursion; use packet counter to save back ptrs (reset
345          * to 0 as we leave), and comefrom to save source hook bitmask.
346          */
347         for (hook = 0; hook < NF_ARP_NUMHOOKS; hook++) {
348                 unsigned int pos = newinfo->hook_entry[hook];
349                 struct arpt_entry *e
350                         = (struct arpt_entry *)(entry0 + pos);
351
352                 if (!(valid_hooks & (1 << hook)))
353                         continue;
354
355                 /* Set initial back pointer. */
356                 e->counters.pcnt = pos;
357
358                 for (;;) {
359                         struct arpt_standard_target *t
360                                 = (void *)arpt_get_target(e);
361
362                         if (e->comefrom & (1 << NF_ARP_NUMHOOKS)) {
363                                 printk("arptables: loop hook %u pos %u %08X.\n",
364                                        hook, pos, e->comefrom);
365                                 return 0;
366                         }
367                         e->comefrom
368                                 |= ((1 << hook) | (1 << NF_ARP_NUMHOOKS));
369
370                         /* Unconditional return/END. */
371                         if (e->target_offset == sizeof(struct arpt_entry)
372                             && (strcmp(t->target.u.user.name,
373                                        ARPT_STANDARD_TARGET) == 0)
374                             && t->verdict < 0
375                             && unconditional(&e->arp)) {
376                                 unsigned int oldpos, size;
377
378                                 if (t->verdict < -NF_MAX_VERDICT - 1) {
379                                         duprintf("mark_source_chains: bad "
380                                                 "negative verdict (%i)\n",
381                                                                 t->verdict);
382                                         return 0;
383                                 }
384
385                                 /* Return: backtrack through the last
386                                  * big jump.
387                                  */
388                                 do {
389                                         e->comefrom ^= (1<<NF_ARP_NUMHOOKS);
390                                         oldpos = pos;
391                                         pos = e->counters.pcnt;
392                                         e->counters.pcnt = 0;
393
394                                         /* We're at the start. */
395                                         if (pos == oldpos)
396                                                 goto next;
397
398                                         e = (struct arpt_entry *)
399                                                 (entry0 + pos);
400                                 } while (oldpos == pos + e->next_offset);
401
402                                 /* Move along one */
403                                 size = e->next_offset;
404                                 e = (struct arpt_entry *)
405                                         (entry0 + pos + size);
406                                 e->counters.pcnt = pos;
407                                 pos += size;
408                         } else {
409                                 int newpos = t->verdict;
410
411                                 if (strcmp(t->target.u.user.name,
412                                            ARPT_STANDARD_TARGET) == 0
413                                     && newpos >= 0) {
414                                         if (newpos > newinfo->size -
415                                                 sizeof(struct arpt_entry)) {
416                                                 duprintf("mark_source_chains: "
417                                                         "bad verdict (%i)\n",
418                                                                 newpos);
419                                                 return 0;
420                                         }
421
422                                         /* This a jump; chase it. */
423                                         duprintf("Jump rule %u -> %u\n",
424                                                  pos, newpos);
425                                 } else {
426                                         /* ... this is a fallthru */
427                                         newpos = pos + e->next_offset;
428                                 }
429                                 e = (struct arpt_entry *)
430                                         (entry0 + newpos);
431                                 e->counters.pcnt = pos;
432                                 pos = newpos;
433                         }
434                 }
435                 next:
436                 duprintf("Finished chain %u\n", hook);
437         }
438         return 1;
439 }
440
441 static inline int standard_check(const struct arpt_entry_target *t,
442                                  unsigned int max_offset)
443 {
444         /* Check standard info. */
445         if (t->u.target_size
446             != ARPT_ALIGN(sizeof(struct arpt_standard_target))) {
447                 duprintf("arpt_standard_check: target size %u != %Zu\n",
448                          t->u.target_size,
449                          ARPT_ALIGN(sizeof(struct arpt_standard_target)));
450                 return 0;
451         }
452
453         return 1;
454 }
455
456 static struct arpt_target arpt_standard_target;
457
458 static inline int check_entry(struct arpt_entry *e, const char *name, unsigned int size,
459                               unsigned int *i)
460 {
461         struct arpt_entry_target *t;
462         struct arpt_target *target;
463         int ret;
464
465         if (!arp_checkentry(&e->arp)) {
466                 duprintf("arp_tables: arp check failed %p %s.\n", e, name);
467                 return -EINVAL;
468         }
469
470         if (e->target_offset + sizeof(struct arpt_entry_target) > e->next_offset)
471                 return -EINVAL;
472
473         t = arpt_get_target(e);
474         if (e->target_offset + t->u.target_size > e->next_offset)
475                 return -EINVAL;
476
477         target = try_then_request_module(xt_find_target(NF_ARP, t->u.user.name,
478                                                         t->u.user.revision),
479                                          "arpt_%s", t->u.user.name);
480         if (IS_ERR(target) || !target) {
481                 duprintf("check_entry: `%s' not found\n", t->u.user.name);
482                 ret = target ? PTR_ERR(target) : -ENOENT;
483                 goto out;
484         }
485         t->u.kernel.target = target;
486
487         ret = xt_check_target(target, NF_ARP, t->u.target_size - sizeof(*t),
488                               name, e->comefrom, 0, 0);
489         if (ret)
490                 goto err;
491
492         if (t->u.kernel.target == &arpt_standard_target) {
493                 if (!standard_check(t, size)) {
494                         ret = -EINVAL;
495                         goto err;
496                 }
497         } else if (t->u.kernel.target->checkentry
498                    && !t->u.kernel.target->checkentry(name, e, target, t->data,
499                                                       e->comefrom)) {
500                 duprintf("arp_tables: check failed for `%s'.\n",
501                          t->u.kernel.target->name);
502                 ret = -EINVAL;
503                 goto err;
504         }
505
506         (*i)++;
507         return 0;
508 err:
509         module_put(t->u.kernel.target->me);
510 out:
511         return ret;
512 }
513
514 static inline int check_entry_size_and_hooks(struct arpt_entry *e,
515                                              struct xt_table_info *newinfo,
516                                              unsigned char *base,
517                                              unsigned char *limit,
518                                              const unsigned int *hook_entries,
519                                              const unsigned int *underflows,
520                                              unsigned int *i)
521 {
522         unsigned int h;
523
524         if ((unsigned long)e % __alignof__(struct arpt_entry) != 0
525             || (unsigned char *)e + sizeof(struct arpt_entry) >= limit) {
526                 duprintf("Bad offset %p\n", e);
527                 return -EINVAL;
528         }
529
530         if (e->next_offset
531             < sizeof(struct arpt_entry) + sizeof(struct arpt_entry_target)) {
532                 duprintf("checking: element %p size %u\n",
533                          e, e->next_offset);
534                 return -EINVAL;
535         }
536
537         /* Check hooks & underflows */
538         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
539                 if ((unsigned char *)e - base == hook_entries[h])
540                         newinfo->hook_entry[h] = hook_entries[h];
541                 if ((unsigned char *)e - base == underflows[h])
542                         newinfo->underflow[h] = underflows[h];
543         }
544
545         /* FIXME: underflows must be unconditional, standard verdicts
546            < 0 (not ARPT_RETURN). --RR */
547
548         /* Clear counters and comefrom */
549         e->counters = ((struct xt_counters) { 0, 0 });
550         e->comefrom = 0;
551
552         (*i)++;
553         return 0;
554 }
555
556 static inline int cleanup_entry(struct arpt_entry *e, unsigned int *i)
557 {
558         struct arpt_entry_target *t;
559
560         if (i && (*i)-- == 0)
561                 return 1;
562
563         t = arpt_get_target(e);
564         if (t->u.kernel.target->destroy)
565                 t->u.kernel.target->destroy(t->u.kernel.target, t->data);
566         module_put(t->u.kernel.target->me);
567         return 0;
568 }
569
570 /* Checks and translates the user-supplied table segment (held in
571  * newinfo).
572  */
573 static int translate_table(const char *name,
574                            unsigned int valid_hooks,
575                            struct xt_table_info *newinfo,
576                            void *entry0,
577                            unsigned int size,
578                            unsigned int number,
579                            const unsigned int *hook_entries,
580                            const unsigned int *underflows)
581 {
582         unsigned int i;
583         int ret;
584
585         newinfo->size = size;
586         newinfo->number = number;
587
588         /* Init all hooks to impossible value. */
589         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
590                 newinfo->hook_entry[i] = 0xFFFFFFFF;
591                 newinfo->underflow[i] = 0xFFFFFFFF;
592         }
593
594         duprintf("translate_table: size %u\n", newinfo->size);
595         i = 0;
596
597         /* Walk through entries, checking offsets. */
598         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
599                                  check_entry_size_and_hooks,
600                                  newinfo,
601                                  entry0,
602                                  entry0 + size,
603                                  hook_entries, underflows, &i);
604         duprintf("translate_table: ARPT_ENTRY_ITERATE gives %d\n", ret);
605         if (ret != 0)
606                 return ret;
607
608         if (i != number) {
609                 duprintf("translate_table: %u not %u entries\n",
610                          i, number);
611                 return -EINVAL;
612         }
613
614         /* Check hooks all assigned */
615         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
616                 /* Only hooks which are valid */
617                 if (!(valid_hooks & (1 << i)))
618                         continue;
619                 if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
620                         duprintf("Invalid hook entry %u %u\n",
621                                  i, hook_entries[i]);
622                         return -EINVAL;
623                 }
624                 if (newinfo->underflow[i] == 0xFFFFFFFF) {
625                         duprintf("Invalid underflow %u %u\n",
626                                  i, underflows[i]);
627                         return -EINVAL;
628                 }
629         }
630
631         if (!mark_source_chains(newinfo, valid_hooks, entry0)) {
632                 duprintf("Looping hook\n");
633                 return -ELOOP;
634         }
635
636         /* Finally, each sanity check must pass */
637         i = 0;
638         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
639                                  check_entry, name, size, &i);
640
641         if (ret != 0) {
642                 ARPT_ENTRY_ITERATE(entry0, newinfo->size,
643                                 cleanup_entry, &i);
644                 return ret;
645         }
646
647         /* And one copy for every other CPU */
648         for_each_possible_cpu(i) {
649                 if (newinfo->entries[i] && newinfo->entries[i] != entry0)
650                         memcpy(newinfo->entries[i], entry0, newinfo->size);
651         }
652
653         return ret;
654 }
655
656 /* Gets counters. */
657 static inline int add_entry_to_counter(const struct arpt_entry *e,
658                                        struct xt_counters total[],
659                                        unsigned int *i)
660 {
661         ADD_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
662
663         (*i)++;
664         return 0;
665 }
666
667 static inline int set_entry_to_counter(const struct arpt_entry *e,
668                                        struct xt_counters total[],
669                                        unsigned int *i)
670 {
671         SET_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
672
673         (*i)++;
674         return 0;
675 }
676
677 static void get_counters(const struct xt_table_info *t,
678                          struct xt_counters counters[])
679 {
680         unsigned int cpu;
681         unsigned int i;
682         unsigned int curcpu;
683
684         /* Instead of clearing (by a previous call to memset())
685          * the counters and using adds, we set the counters
686          * with data used by 'current' CPU
687          * We dont care about preemption here.
688          */
689         curcpu = raw_smp_processor_id();
690
691         i = 0;
692         ARPT_ENTRY_ITERATE(t->entries[curcpu],
693                            t->size,
694                            set_entry_to_counter,
695                            counters,
696                            &i);
697
698         for_each_possible_cpu(cpu) {
699                 if (cpu == curcpu)
700                         continue;
701                 i = 0;
702                 ARPT_ENTRY_ITERATE(t->entries[cpu],
703                                    t->size,
704                                    add_entry_to_counter,
705                                    counters,
706                                    &i);
707         }
708 }
709
710 static int copy_entries_to_user(unsigned int total_size,
711                                 struct arpt_table *table,
712                                 void __user *userptr)
713 {
714         unsigned int off, num, countersize;
715         struct arpt_entry *e;
716         struct xt_counters *counters;
717         struct xt_table_info *private = table->private;
718         int ret = 0;
719         void *loc_cpu_entry;
720
721         /* We need atomic snapshot of counters: rest doesn't change
722          * (other than comefrom, which userspace doesn't care
723          * about).
724          */
725         countersize = sizeof(struct xt_counters) * private->number;
726         counters = vmalloc_node(countersize, numa_node_id());
727
728         if (counters == NULL)
729                 return -ENOMEM;
730
731         /* First, sum counters... */
732         write_lock_bh(&table->lock);
733         get_counters(private, counters);
734         write_unlock_bh(&table->lock);
735
736         loc_cpu_entry = private->entries[raw_smp_processor_id()];
737         /* ... then copy entire thing ... */
738         if (copy_to_user(userptr, loc_cpu_entry, total_size) != 0) {
739                 ret = -EFAULT;
740                 goto free_counters;
741         }
742
743         /* FIXME: use iterator macros --RR */
744         /* ... then go back and fix counters and names */
745         for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
746                 struct arpt_entry_target *t;
747
748                 e = (struct arpt_entry *)(loc_cpu_entry + off);
749                 if (copy_to_user(userptr + off
750                                  + offsetof(struct arpt_entry, counters),
751                                  &counters[num],
752                                  sizeof(counters[num])) != 0) {
753                         ret = -EFAULT;
754                         goto free_counters;
755                 }
756
757                 t = arpt_get_target(e);
758                 if (copy_to_user(userptr + off + e->target_offset
759                                  + offsetof(struct arpt_entry_target,
760                                             u.user.name),
761                                  t->u.kernel.target->name,
762                                  strlen(t->u.kernel.target->name)+1) != 0) {
763                         ret = -EFAULT;
764                         goto free_counters;
765                 }
766         }
767
768  free_counters:
769         vfree(counters);
770         return ret;
771 }
772
773 static int get_entries(const struct arpt_get_entries *entries,
774                        struct arpt_get_entries __user *uptr)
775 {
776         int ret;
777         struct arpt_table *t;
778
779         t = xt_find_table_lock(NF_ARP, entries->name);
780         if (t && !IS_ERR(t)) {
781                 struct xt_table_info *private = t->private;
782                 duprintf("t->private->number = %u\n",
783                          private->number);
784                 if (entries->size == private->size)
785                         ret = copy_entries_to_user(private->size,
786                                                    t, uptr->entrytable);
787                 else {
788                         duprintf("get_entries: I've got %u not %u!\n",
789                                  private->size, entries->size);
790                         ret = -EINVAL;
791                 }
792                 module_put(t->me);
793                 xt_table_unlock(t);
794         } else
795                 ret = t ? PTR_ERR(t) : -ENOENT;
796
797         return ret;
798 }
799
800 static int do_replace(void __user *user, unsigned int len)
801 {
802         int ret;
803         struct arpt_replace tmp;
804         struct arpt_table *t;
805         struct xt_table_info *newinfo, *oldinfo;
806         struct xt_counters *counters;
807         void *loc_cpu_entry, *loc_cpu_old_entry;
808
809         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
810                 return -EFAULT;
811
812         /* Hack: Causes ipchains to give correct error msg --RR */
813         if (len != sizeof(tmp) + tmp.size)
814                 return -ENOPROTOOPT;
815
816         /* overflow check */
817         if (tmp.size >= (INT_MAX - sizeof(struct xt_table_info)) / NR_CPUS -
818                         SMP_CACHE_BYTES)
819                 return -ENOMEM;
820         if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
821                 return -ENOMEM;
822
823         newinfo = xt_alloc_table_info(tmp.size);
824         if (!newinfo)
825                 return -ENOMEM;
826
827         /* choose the copy that is on our node/cpu */
828         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
829         if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
830                            tmp.size) != 0) {
831                 ret = -EFAULT;
832                 goto free_newinfo;
833         }
834
835         counters = vmalloc(tmp.num_counters * sizeof(struct xt_counters));
836         if (!counters) {
837                 ret = -ENOMEM;
838                 goto free_newinfo;
839         }
840
841         ret = translate_table(tmp.name, tmp.valid_hooks,
842                               newinfo, loc_cpu_entry, tmp.size, tmp.num_entries,
843                               tmp.hook_entry, tmp.underflow);
844         if (ret != 0)
845                 goto free_newinfo_counters;
846
847         duprintf("arp_tables: Translated table\n");
848
849         t = try_then_request_module(xt_find_table_lock(NF_ARP, tmp.name),
850                                     "arptable_%s", tmp.name);
851         if (!t || IS_ERR(t)) {
852                 ret = t ? PTR_ERR(t) : -ENOENT;
853                 goto free_newinfo_counters_untrans;
854         }
855
856         /* You lied! */
857         if (tmp.valid_hooks != t->valid_hooks) {
858                 duprintf("Valid hook crap: %08X vs %08X\n",
859                          tmp.valid_hooks, t->valid_hooks);
860                 ret = -EINVAL;
861                 goto put_module;
862         }
863
864         oldinfo = xt_replace_table(t, tmp.num_counters, newinfo, &ret);
865         if (!oldinfo)
866                 goto put_module;
867
868         /* Update module usage count based on number of rules */
869         duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
870                 oldinfo->number, oldinfo->initial_entries, newinfo->number);
871         if ((oldinfo->number > oldinfo->initial_entries) || 
872             (newinfo->number <= oldinfo->initial_entries)) 
873                 module_put(t->me);
874         if ((oldinfo->number > oldinfo->initial_entries) &&
875             (newinfo->number <= oldinfo->initial_entries))
876                 module_put(t->me);
877
878         /* Get the old counters. */
879         get_counters(oldinfo, counters);
880         /* Decrease module usage counts and free resource */
881         loc_cpu_old_entry = oldinfo->entries[raw_smp_processor_id()];
882         ARPT_ENTRY_ITERATE(loc_cpu_old_entry, oldinfo->size, cleanup_entry,NULL);
883
884         xt_free_table_info(oldinfo);
885         if (copy_to_user(tmp.counters, counters,
886                          sizeof(struct xt_counters) * tmp.num_counters) != 0)
887                 ret = -EFAULT;
888         vfree(counters);
889         xt_table_unlock(t);
890         return ret;
891
892  put_module:
893         module_put(t->me);
894         xt_table_unlock(t);
895  free_newinfo_counters_untrans:
896         ARPT_ENTRY_ITERATE(loc_cpu_entry, newinfo->size, cleanup_entry, NULL);
897  free_newinfo_counters:
898         vfree(counters);
899  free_newinfo:
900         xt_free_table_info(newinfo);
901         return ret;
902 }
903
904 /* We're lazy, and add to the first CPU; overflow works its fey magic
905  * and everything is OK.
906  */
907 static inline int add_counter_to_entry(struct arpt_entry *e,
908                                        const struct xt_counters addme[],
909                                        unsigned int *i)
910 {
911
912         ADD_COUNTER(e->counters, addme[*i].bcnt, addme[*i].pcnt);
913
914         (*i)++;
915         return 0;
916 }
917
918 static int do_add_counters(void __user *user, unsigned int len)
919 {
920         unsigned int i;
921         struct xt_counters_info tmp, *paddc;
922         struct arpt_table *t;
923         struct xt_table_info *private;
924         int ret = 0;
925         void *loc_cpu_entry;
926
927         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
928                 return -EFAULT;
929
930         if (len != sizeof(tmp) + tmp.num_counters*sizeof(struct xt_counters))
931                 return -EINVAL;
932
933         paddc = vmalloc(len);
934         if (!paddc)
935                 return -ENOMEM;
936
937         if (copy_from_user(paddc, user, len) != 0) {
938                 ret = -EFAULT;
939                 goto free;
940         }
941
942         t = xt_find_table_lock(NF_ARP, tmp.name);
943         if (!t || IS_ERR(t)) {
944                 ret = t ? PTR_ERR(t) : -ENOENT;
945                 goto free;
946         }
947
948         write_lock_bh(&t->lock);
949         private = t->private;
950         if (private->number != tmp.num_counters) {
951                 ret = -EINVAL;
952                 goto unlock_up_free;
953         }
954
955         i = 0;
956         /* Choose the copy that is on our node */
957         loc_cpu_entry = private->entries[smp_processor_id()];
958         ARPT_ENTRY_ITERATE(loc_cpu_entry,
959                            private->size,
960                            add_counter_to_entry,
961                            paddc->counters,
962                            &i);
963  unlock_up_free:
964         write_unlock_bh(&t->lock);
965         xt_table_unlock(t);
966         module_put(t->me);
967  free:
968         vfree(paddc);
969
970         return ret;
971 }
972
973 static int do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
974 {
975         int ret;
976
977         if (!capable(CAP_NET_ADMIN))
978                 return -EPERM;
979
980         switch (cmd) {
981         case ARPT_SO_SET_REPLACE:
982                 ret = do_replace(user, len);
983                 break;
984
985         case ARPT_SO_SET_ADD_COUNTERS:
986                 ret = do_add_counters(user, len);
987                 break;
988
989         default:
990                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
991                 ret = -EINVAL;
992         }
993
994         return ret;
995 }
996
997 static int do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
998 {
999         int ret;
1000
1001         if (!capable(CAP_NET_ADMIN))
1002                 return -EPERM;
1003
1004         switch (cmd) {
1005         case ARPT_SO_GET_INFO: {
1006                 char name[ARPT_TABLE_MAXNAMELEN];
1007                 struct arpt_table *t;
1008
1009                 if (*len != sizeof(struct arpt_getinfo)) {
1010                         duprintf("length %u != %Zu\n", *len,
1011                                  sizeof(struct arpt_getinfo));
1012                         ret = -EINVAL;
1013                         break;
1014                 }
1015
1016                 if (copy_from_user(name, user, sizeof(name)) != 0) {
1017                         ret = -EFAULT;
1018                         break;
1019                 }
1020                 name[ARPT_TABLE_MAXNAMELEN-1] = '\0';
1021
1022                 t = try_then_request_module(xt_find_table_lock(NF_ARP, name),
1023                                             "arptable_%s", name);
1024                 if (t && !IS_ERR(t)) {
1025                         struct arpt_getinfo info;
1026                         struct xt_table_info *private = t->private;
1027
1028                         info.valid_hooks = t->valid_hooks;
1029                         memcpy(info.hook_entry, private->hook_entry,
1030                                sizeof(info.hook_entry));
1031                         memcpy(info.underflow, private->underflow,
1032                                sizeof(info.underflow));
1033                         info.num_entries = private->number;
1034                         info.size = private->size;
1035                         strcpy(info.name, name);
1036
1037                         if (copy_to_user(user, &info, *len) != 0)
1038                                 ret = -EFAULT;
1039                         else
1040                                 ret = 0;
1041                         xt_table_unlock(t);
1042                         module_put(t->me);
1043                 } else
1044                         ret = t ? PTR_ERR(t) : -ENOENT;
1045         }
1046         break;
1047
1048         case ARPT_SO_GET_ENTRIES: {
1049                 struct arpt_get_entries get;
1050
1051                 if (*len < sizeof(get)) {
1052                         duprintf("get_entries: %u < %Zu\n", *len, sizeof(get));
1053                         ret = -EINVAL;
1054                 } else if (copy_from_user(&get, user, sizeof(get)) != 0) {
1055                         ret = -EFAULT;
1056                 } else if (*len != sizeof(struct arpt_get_entries) + get.size) {
1057                         duprintf("get_entries: %u != %Zu\n", *len,
1058                                  sizeof(struct arpt_get_entries) + get.size);
1059                         ret = -EINVAL;
1060                 } else
1061                         ret = get_entries(&get, user);
1062                 break;
1063         }
1064
1065         case ARPT_SO_GET_REVISION_TARGET: {
1066                 struct xt_get_revision rev;
1067
1068                 if (*len != sizeof(rev)) {
1069                         ret = -EINVAL;
1070                         break;
1071                 }
1072                 if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
1073                         ret = -EFAULT;
1074                         break;
1075                 }
1076
1077                 try_then_request_module(xt_find_revision(NF_ARP, rev.name,
1078                                                          rev.revision, 1, &ret),
1079                                         "arpt_%s", rev.name);
1080                 break;
1081         }
1082
1083         default:
1084                 duprintf("do_arpt_get_ctl: unknown request %i\n", cmd);
1085                 ret = -EINVAL;
1086         }
1087
1088         return ret;
1089 }
1090
1091 int arpt_register_table(struct arpt_table *table,
1092                         const struct arpt_replace *repl)
1093 {
1094         int ret;
1095         struct xt_table_info *newinfo;
1096         static struct xt_table_info bootstrap
1097                 = { 0, 0, 0, { 0 }, { 0 }, { } };
1098         void *loc_cpu_entry;
1099
1100         newinfo = xt_alloc_table_info(repl->size);
1101         if (!newinfo) {
1102                 ret = -ENOMEM;
1103                 return ret;
1104         }
1105
1106         /* choose the copy on our node/cpu */
1107         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1108         memcpy(loc_cpu_entry, repl->entries, repl->size);
1109
1110         ret = translate_table(table->name, table->valid_hooks,
1111                               newinfo, loc_cpu_entry, repl->size,
1112                               repl->num_entries,
1113                               repl->hook_entry,
1114                               repl->underflow);
1115
1116         duprintf("arpt_register_table: translate table gives %d\n", ret);
1117         if (ret != 0) {
1118                 xt_free_table_info(newinfo);
1119                 return ret;
1120         }
1121
1122         ret = xt_register_table(table, &bootstrap, newinfo);
1123         if (ret != 0) {
1124                 xt_free_table_info(newinfo);
1125                 return ret;
1126         }
1127
1128         return 0;
1129 }
1130
1131 void arpt_unregister_table(struct arpt_table *table)
1132 {
1133         struct xt_table_info *private;
1134         void *loc_cpu_entry;
1135
1136         private = xt_unregister_table(table);
1137
1138         /* Decrease module usage counts and free resources */
1139         loc_cpu_entry = private->entries[raw_smp_processor_id()];
1140         ARPT_ENTRY_ITERATE(loc_cpu_entry, private->size,
1141                            cleanup_entry, NULL);
1142         xt_free_table_info(private);
1143 }
1144
1145 /* The built-in targets: standard (NULL) and error. */
1146 static struct arpt_target arpt_standard_target = {
1147         .name           = ARPT_STANDARD_TARGET,
1148         .targetsize     = sizeof(int),
1149         .family         = NF_ARP,
1150 };
1151
1152 static struct arpt_target arpt_error_target = {
1153         .name           = ARPT_ERROR_TARGET,
1154         .target         = arpt_error,
1155         .targetsize     = ARPT_FUNCTION_MAXNAMELEN,
1156         .family         = NF_ARP,
1157 };
1158
1159 static struct nf_sockopt_ops arpt_sockopts = {
1160         .pf             = PF_INET,
1161         .set_optmin     = ARPT_BASE_CTL,
1162         .set_optmax     = ARPT_SO_SET_MAX+1,
1163         .set            = do_arpt_set_ctl,
1164         .get_optmin     = ARPT_BASE_CTL,
1165         .get_optmax     = ARPT_SO_GET_MAX+1,
1166         .get            = do_arpt_get_ctl,
1167 };
1168
1169 static int __init arp_tables_init(void)
1170 {
1171         int ret;
1172
1173         ret = xt_proto_init(NF_ARP);
1174         if (ret < 0)
1175                 goto err1;
1176
1177         /* Noone else will be downing sem now, so we won't sleep */
1178         ret = xt_register_target(&arpt_standard_target);
1179         if (ret < 0)
1180                 goto err2;
1181         ret = xt_register_target(&arpt_error_target);
1182         if (ret < 0)
1183                 goto err3;
1184
1185         /* Register setsockopt */
1186         ret = nf_register_sockopt(&arpt_sockopts);
1187         if (ret < 0)
1188                 goto err4;
1189
1190         printk("arp_tables: (C) 2002 David S. Miller\n");
1191         return 0;
1192
1193 err4:
1194         xt_unregister_target(&arpt_error_target);
1195 err3:
1196         xt_unregister_target(&arpt_standard_target);
1197 err2:
1198         xt_proto_fini(NF_ARP);
1199 err1:
1200         return ret;
1201 }
1202
1203 static void __exit arp_tables_fini(void)
1204 {
1205         nf_unregister_sockopt(&arpt_sockopts);
1206         xt_unregister_target(&arpt_error_target);
1207         xt_unregister_target(&arpt_standard_target);
1208         xt_proto_fini(NF_ARP);
1209 }
1210
1211 EXPORT_SYMBOL(arpt_register_table);
1212 EXPORT_SYMBOL(arpt_unregister_table);
1213 EXPORT_SYMBOL(arpt_do_table);
1214
1215 module_init(arp_tables_init);
1216 module_exit(arp_tables_fini);