[NETFILTER]: ip_conntrack: properly use RCU API for ip_ct_protos array
[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                         int visited = e->comefrom & (1 << hook);
362
363                         if (e->comefrom & (1 << NF_ARP_NUMHOOKS)) {
364                                 printk("arptables: loop hook %u pos %u %08X.\n",
365                                        hook, pos, e->comefrom);
366                                 return 0;
367                         }
368                         e->comefrom
369                                 |= ((1 << hook) | (1 << NF_ARP_NUMHOOKS));
370
371                         /* Unconditional return/END. */
372                         if ((e->target_offset == sizeof(struct arpt_entry)
373                             && (strcmp(t->target.u.user.name,
374                                        ARPT_STANDARD_TARGET) == 0)
375                             && t->verdict < 0
376                             && unconditional(&e->arp)) || visited) {
377                                 unsigned int oldpos, size;
378
379                                 if (t->verdict < -NF_MAX_VERDICT - 1) {
380                                         duprintf("mark_source_chains: bad "
381                                                 "negative verdict (%i)\n",
382                                                                 t->verdict);
383                                         return 0;
384                                 }
385
386                                 /* Return: backtrack through the last
387                                  * big jump.
388                                  */
389                                 do {
390                                         e->comefrom ^= (1<<NF_ARP_NUMHOOKS);
391                                         oldpos = pos;
392                                         pos = e->counters.pcnt;
393                                         e->counters.pcnt = 0;
394
395                                         /* We're at the start. */
396                                         if (pos == oldpos)
397                                                 goto next;
398
399                                         e = (struct arpt_entry *)
400                                                 (entry0 + pos);
401                                 } while (oldpos == pos + e->next_offset);
402
403                                 /* Move along one */
404                                 size = e->next_offset;
405                                 e = (struct arpt_entry *)
406                                         (entry0 + pos + size);
407                                 e->counters.pcnt = pos;
408                                 pos += size;
409                         } else {
410                                 int newpos = t->verdict;
411
412                                 if (strcmp(t->target.u.user.name,
413                                            ARPT_STANDARD_TARGET) == 0
414                                     && newpos >= 0) {
415                                         if (newpos > newinfo->size -
416                                                 sizeof(struct arpt_entry)) {
417                                                 duprintf("mark_source_chains: "
418                                                         "bad verdict (%i)\n",
419                                                                 newpos);
420                                                 return 0;
421                                         }
422
423                                         /* This a jump; chase it. */
424                                         duprintf("Jump rule %u -> %u\n",
425                                                  pos, newpos);
426                                 } else {
427                                         /* ... this is a fallthru */
428                                         newpos = pos + e->next_offset;
429                                 }
430                                 e = (struct arpt_entry *)
431                                         (entry0 + newpos);
432                                 e->counters.pcnt = pos;
433                                 pos = newpos;
434                         }
435                 }
436                 next:
437                 duprintf("Finished chain %u\n", hook);
438         }
439         return 1;
440 }
441
442 static inline int standard_check(const struct arpt_entry_target *t,
443                                  unsigned int max_offset)
444 {
445         /* Check standard info. */
446         if (t->u.target_size
447             != ARPT_ALIGN(sizeof(struct arpt_standard_target))) {
448                 duprintf("arpt_standard_check: target size %u != %Zu\n",
449                          t->u.target_size,
450                          ARPT_ALIGN(sizeof(struct arpt_standard_target)));
451                 return 0;
452         }
453
454         return 1;
455 }
456
457 static struct arpt_target arpt_standard_target;
458
459 static inline int check_entry(struct arpt_entry *e, const char *name, unsigned int size,
460                               unsigned int *i)
461 {
462         struct arpt_entry_target *t;
463         struct arpt_target *target;
464         int ret;
465
466         if (!arp_checkentry(&e->arp)) {
467                 duprintf("arp_tables: arp check failed %p %s.\n", e, name);
468                 return -EINVAL;
469         }
470
471         if (e->target_offset + sizeof(struct arpt_entry_target) > e->next_offset)
472                 return -EINVAL;
473
474         t = arpt_get_target(e);
475         if (e->target_offset + t->u.target_size > e->next_offset)
476                 return -EINVAL;
477
478         target = try_then_request_module(xt_find_target(NF_ARP, t->u.user.name,
479                                                         t->u.user.revision),
480                                          "arpt_%s", t->u.user.name);
481         if (IS_ERR(target) || !target) {
482                 duprintf("check_entry: `%s' not found\n", t->u.user.name);
483                 ret = target ? PTR_ERR(target) : -ENOENT;
484                 goto out;
485         }
486         t->u.kernel.target = target;
487
488         ret = xt_check_target(target, NF_ARP, t->u.target_size - sizeof(*t),
489                               name, e->comefrom, 0, 0);
490         if (ret)
491                 goto err;
492
493         if (t->u.kernel.target == &arpt_standard_target) {
494                 if (!standard_check(t, size)) {
495                         ret = -EINVAL;
496                         goto err;
497                 }
498         } else if (t->u.kernel.target->checkentry
499                    && !t->u.kernel.target->checkentry(name, e, target, t->data,
500                                                       e->comefrom)) {
501                 duprintf("arp_tables: check failed for `%s'.\n",
502                          t->u.kernel.target->name);
503                 ret = -EINVAL;
504                 goto err;
505         }
506
507         (*i)++;
508         return 0;
509 err:
510         module_put(t->u.kernel.target->me);
511 out:
512         return ret;
513 }
514
515 static inline int check_entry_size_and_hooks(struct arpt_entry *e,
516                                              struct xt_table_info *newinfo,
517                                              unsigned char *base,
518                                              unsigned char *limit,
519                                              const unsigned int *hook_entries,
520                                              const unsigned int *underflows,
521                                              unsigned int *i)
522 {
523         unsigned int h;
524
525         if ((unsigned long)e % __alignof__(struct arpt_entry) != 0
526             || (unsigned char *)e + sizeof(struct arpt_entry) >= limit) {
527                 duprintf("Bad offset %p\n", e);
528                 return -EINVAL;
529         }
530
531         if (e->next_offset
532             < sizeof(struct arpt_entry) + sizeof(struct arpt_entry_target)) {
533                 duprintf("checking: element %p size %u\n",
534                          e, e->next_offset);
535                 return -EINVAL;
536         }
537
538         /* Check hooks & underflows */
539         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
540                 if ((unsigned char *)e - base == hook_entries[h])
541                         newinfo->hook_entry[h] = hook_entries[h];
542                 if ((unsigned char *)e - base == underflows[h])
543                         newinfo->underflow[h] = underflows[h];
544         }
545
546         /* FIXME: underflows must be unconditional, standard verdicts
547            < 0 (not ARPT_RETURN). --RR */
548
549         /* Clear counters and comefrom */
550         e->counters = ((struct xt_counters) { 0, 0 });
551         e->comefrom = 0;
552
553         (*i)++;
554         return 0;
555 }
556
557 static inline int cleanup_entry(struct arpt_entry *e, unsigned int *i)
558 {
559         struct arpt_entry_target *t;
560
561         if (i && (*i)-- == 0)
562                 return 1;
563
564         t = arpt_get_target(e);
565         if (t->u.kernel.target->destroy)
566                 t->u.kernel.target->destroy(t->u.kernel.target, t->data);
567         module_put(t->u.kernel.target->me);
568         return 0;
569 }
570
571 /* Checks and translates the user-supplied table segment (held in
572  * newinfo).
573  */
574 static int translate_table(const char *name,
575                            unsigned int valid_hooks,
576                            struct xt_table_info *newinfo,
577                            void *entry0,
578                            unsigned int size,
579                            unsigned int number,
580                            const unsigned int *hook_entries,
581                            const unsigned int *underflows)
582 {
583         unsigned int i;
584         int ret;
585
586         newinfo->size = size;
587         newinfo->number = number;
588
589         /* Init all hooks to impossible value. */
590         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
591                 newinfo->hook_entry[i] = 0xFFFFFFFF;
592                 newinfo->underflow[i] = 0xFFFFFFFF;
593         }
594
595         duprintf("translate_table: size %u\n", newinfo->size);
596         i = 0;
597
598         /* Walk through entries, checking offsets. */
599         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
600                                  check_entry_size_and_hooks,
601                                  newinfo,
602                                  entry0,
603                                  entry0 + size,
604                                  hook_entries, underflows, &i);
605         duprintf("translate_table: ARPT_ENTRY_ITERATE gives %d\n", ret);
606         if (ret != 0)
607                 return ret;
608
609         if (i != number) {
610                 duprintf("translate_table: %u not %u entries\n",
611                          i, number);
612                 return -EINVAL;
613         }
614
615         /* Check hooks all assigned */
616         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
617                 /* Only hooks which are valid */
618                 if (!(valid_hooks & (1 << i)))
619                         continue;
620                 if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
621                         duprintf("Invalid hook entry %u %u\n",
622                                  i, hook_entries[i]);
623                         return -EINVAL;
624                 }
625                 if (newinfo->underflow[i] == 0xFFFFFFFF) {
626                         duprintf("Invalid underflow %u %u\n",
627                                  i, underflows[i]);
628                         return -EINVAL;
629                 }
630         }
631
632         if (!mark_source_chains(newinfo, valid_hooks, entry0)) {
633                 duprintf("Looping hook\n");
634                 return -ELOOP;
635         }
636
637         /* Finally, each sanity check must pass */
638         i = 0;
639         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
640                                  check_entry, name, size, &i);
641
642         if (ret != 0) {
643                 ARPT_ENTRY_ITERATE(entry0, newinfo->size,
644                                 cleanup_entry, &i);
645                 return ret;
646         }
647
648         /* And one copy for every other CPU */
649         for_each_possible_cpu(i) {
650                 if (newinfo->entries[i] && newinfo->entries[i] != entry0)
651                         memcpy(newinfo->entries[i], entry0, newinfo->size);
652         }
653
654         return ret;
655 }
656
657 /* Gets counters. */
658 static inline int add_entry_to_counter(const struct arpt_entry *e,
659                                        struct xt_counters total[],
660                                        unsigned int *i)
661 {
662         ADD_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
663
664         (*i)++;
665         return 0;
666 }
667
668 static inline int set_entry_to_counter(const struct arpt_entry *e,
669                                        struct xt_counters total[],
670                                        unsigned int *i)
671 {
672         SET_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
673
674         (*i)++;
675         return 0;
676 }
677
678 static void get_counters(const struct xt_table_info *t,
679                          struct xt_counters counters[])
680 {
681         unsigned int cpu;
682         unsigned int i;
683         unsigned int curcpu;
684
685         /* Instead of clearing (by a previous call to memset())
686          * the counters and using adds, we set the counters
687          * with data used by 'current' CPU
688          * We dont care about preemption here.
689          */
690         curcpu = raw_smp_processor_id();
691
692         i = 0;
693         ARPT_ENTRY_ITERATE(t->entries[curcpu],
694                            t->size,
695                            set_entry_to_counter,
696                            counters,
697                            &i);
698
699         for_each_possible_cpu(cpu) {
700                 if (cpu == curcpu)
701                         continue;
702                 i = 0;
703                 ARPT_ENTRY_ITERATE(t->entries[cpu],
704                                    t->size,
705                                    add_entry_to_counter,
706                                    counters,
707                                    &i);
708         }
709 }
710
711 static int copy_entries_to_user(unsigned int total_size,
712                                 struct arpt_table *table,
713                                 void __user *userptr)
714 {
715         unsigned int off, num, countersize;
716         struct arpt_entry *e;
717         struct xt_counters *counters;
718         struct xt_table_info *private = table->private;
719         int ret = 0;
720         void *loc_cpu_entry;
721
722         /* We need atomic snapshot of counters: rest doesn't change
723          * (other than comefrom, which userspace doesn't care
724          * about).
725          */
726         countersize = sizeof(struct xt_counters) * private->number;
727         counters = vmalloc_node(countersize, numa_node_id());
728
729         if (counters == NULL)
730                 return -ENOMEM;
731
732         /* First, sum counters... */
733         write_lock_bh(&table->lock);
734         get_counters(private, counters);
735         write_unlock_bh(&table->lock);
736
737         loc_cpu_entry = private->entries[raw_smp_processor_id()];
738         /* ... then copy entire thing ... */
739         if (copy_to_user(userptr, loc_cpu_entry, total_size) != 0) {
740                 ret = -EFAULT;
741                 goto free_counters;
742         }
743
744         /* FIXME: use iterator macros --RR */
745         /* ... then go back and fix counters and names */
746         for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
747                 struct arpt_entry_target *t;
748
749                 e = (struct arpt_entry *)(loc_cpu_entry + off);
750                 if (copy_to_user(userptr + off
751                                  + offsetof(struct arpt_entry, counters),
752                                  &counters[num],
753                                  sizeof(counters[num])) != 0) {
754                         ret = -EFAULT;
755                         goto free_counters;
756                 }
757
758                 t = arpt_get_target(e);
759                 if (copy_to_user(userptr + off + e->target_offset
760                                  + offsetof(struct arpt_entry_target,
761                                             u.user.name),
762                                  t->u.kernel.target->name,
763                                  strlen(t->u.kernel.target->name)+1) != 0) {
764                         ret = -EFAULT;
765                         goto free_counters;
766                 }
767         }
768
769  free_counters:
770         vfree(counters);
771         return ret;
772 }
773
774 static int get_entries(const struct arpt_get_entries *entries,
775                        struct arpt_get_entries __user *uptr)
776 {
777         int ret;
778         struct arpt_table *t;
779
780         t = xt_find_table_lock(NF_ARP, entries->name);
781         if (t && !IS_ERR(t)) {
782                 struct xt_table_info *private = t->private;
783                 duprintf("t->private->number = %u\n",
784                          private->number);
785                 if (entries->size == private->size)
786                         ret = copy_entries_to_user(private->size,
787                                                    t, uptr->entrytable);
788                 else {
789                         duprintf("get_entries: I've got %u not %u!\n",
790                                  private->size, entries->size);
791                         ret = -EINVAL;
792                 }
793                 module_put(t->me);
794                 xt_table_unlock(t);
795         } else
796                 ret = t ? PTR_ERR(t) : -ENOENT;
797
798         return ret;
799 }
800
801 static int do_replace(void __user *user, unsigned int len)
802 {
803         int ret;
804         struct arpt_replace tmp;
805         struct arpt_table *t;
806         struct xt_table_info *newinfo, *oldinfo;
807         struct xt_counters *counters;
808         void *loc_cpu_entry, *loc_cpu_old_entry;
809
810         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
811                 return -EFAULT;
812
813         /* Hack: Causes ipchains to give correct error msg --RR */
814         if (len != sizeof(tmp) + tmp.size)
815                 return -ENOPROTOOPT;
816
817         /* overflow check */
818         if (tmp.size >= (INT_MAX - sizeof(struct xt_table_info)) / NR_CPUS -
819                         SMP_CACHE_BYTES)
820                 return -ENOMEM;
821         if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
822                 return -ENOMEM;
823
824         newinfo = xt_alloc_table_info(tmp.size);
825         if (!newinfo)
826                 return -ENOMEM;
827
828         /* choose the copy that is on our node/cpu */
829         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
830         if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
831                            tmp.size) != 0) {
832                 ret = -EFAULT;
833                 goto free_newinfo;
834         }
835
836         counters = vmalloc(tmp.num_counters * sizeof(struct xt_counters));
837         if (!counters) {
838                 ret = -ENOMEM;
839                 goto free_newinfo;
840         }
841
842         ret = translate_table(tmp.name, tmp.valid_hooks,
843                               newinfo, loc_cpu_entry, tmp.size, tmp.num_entries,
844                               tmp.hook_entry, tmp.underflow);
845         if (ret != 0)
846                 goto free_newinfo_counters;
847
848         duprintf("arp_tables: Translated table\n");
849
850         t = try_then_request_module(xt_find_table_lock(NF_ARP, tmp.name),
851                                     "arptable_%s", tmp.name);
852         if (!t || IS_ERR(t)) {
853                 ret = t ? PTR_ERR(t) : -ENOENT;
854                 goto free_newinfo_counters_untrans;
855         }
856
857         /* You lied! */
858         if (tmp.valid_hooks != t->valid_hooks) {
859                 duprintf("Valid hook crap: %08X vs %08X\n",
860                          tmp.valid_hooks, t->valid_hooks);
861                 ret = -EINVAL;
862                 goto put_module;
863         }
864
865         oldinfo = xt_replace_table(t, tmp.num_counters, newinfo, &ret);
866         if (!oldinfo)
867                 goto put_module;
868
869         /* Update module usage count based on number of rules */
870         duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
871                 oldinfo->number, oldinfo->initial_entries, newinfo->number);
872         if ((oldinfo->number > oldinfo->initial_entries) ||
873             (newinfo->number <= oldinfo->initial_entries))
874                 module_put(t->me);
875         if ((oldinfo->number > oldinfo->initial_entries) &&
876             (newinfo->number <= oldinfo->initial_entries))
877                 module_put(t->me);
878
879         /* Get the old counters. */
880         get_counters(oldinfo, counters);
881         /* Decrease module usage counts and free resource */
882         loc_cpu_old_entry = oldinfo->entries[raw_smp_processor_id()];
883         ARPT_ENTRY_ITERATE(loc_cpu_old_entry, oldinfo->size, cleanup_entry,NULL);
884
885         xt_free_table_info(oldinfo);
886         if (copy_to_user(tmp.counters, counters,
887                          sizeof(struct xt_counters) * tmp.num_counters) != 0)
888                 ret = -EFAULT;
889         vfree(counters);
890         xt_table_unlock(t);
891         return ret;
892
893  put_module:
894         module_put(t->me);
895         xt_table_unlock(t);
896  free_newinfo_counters_untrans:
897         ARPT_ENTRY_ITERATE(loc_cpu_entry, newinfo->size, cleanup_entry, NULL);
898  free_newinfo_counters:
899         vfree(counters);
900  free_newinfo:
901         xt_free_table_info(newinfo);
902         return ret;
903 }
904
905 /* We're lazy, and add to the first CPU; overflow works its fey magic
906  * and everything is OK.
907  */
908 static inline int add_counter_to_entry(struct arpt_entry *e,
909                                        const struct xt_counters addme[],
910                                        unsigned int *i)
911 {
912
913         ADD_COUNTER(e->counters, addme[*i].bcnt, addme[*i].pcnt);
914
915         (*i)++;
916         return 0;
917 }
918
919 static int do_add_counters(void __user *user, unsigned int len)
920 {
921         unsigned int i;
922         struct xt_counters_info tmp, *paddc;
923         struct arpt_table *t;
924         struct xt_table_info *private;
925         int ret = 0;
926         void *loc_cpu_entry;
927
928         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
929                 return -EFAULT;
930
931         if (len != sizeof(tmp) + tmp.num_counters*sizeof(struct xt_counters))
932                 return -EINVAL;
933
934         paddc = vmalloc(len);
935         if (!paddc)
936                 return -ENOMEM;
937
938         if (copy_from_user(paddc, user, len) != 0) {
939                 ret = -EFAULT;
940                 goto free;
941         }
942
943         t = xt_find_table_lock(NF_ARP, tmp.name);
944         if (!t || IS_ERR(t)) {
945                 ret = t ? PTR_ERR(t) : -ENOENT;
946                 goto free;
947         }
948
949         write_lock_bh(&t->lock);
950         private = t->private;
951         if (private->number != tmp.num_counters) {
952                 ret = -EINVAL;
953                 goto unlock_up_free;
954         }
955
956         i = 0;
957         /* Choose the copy that is on our node */
958         loc_cpu_entry = private->entries[smp_processor_id()];
959         ARPT_ENTRY_ITERATE(loc_cpu_entry,
960                            private->size,
961                            add_counter_to_entry,
962                            paddc->counters,
963                            &i);
964  unlock_up_free:
965         write_unlock_bh(&t->lock);
966         xt_table_unlock(t);
967         module_put(t->me);
968  free:
969         vfree(paddc);
970
971         return ret;
972 }
973
974 static int do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
975 {
976         int ret;
977
978         if (!capable(CAP_NET_ADMIN))
979                 return -EPERM;
980
981         switch (cmd) {
982         case ARPT_SO_SET_REPLACE:
983                 ret = do_replace(user, len);
984                 break;
985
986         case ARPT_SO_SET_ADD_COUNTERS:
987                 ret = do_add_counters(user, len);
988                 break;
989
990         default:
991                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
992                 ret = -EINVAL;
993         }
994
995         return ret;
996 }
997
998 static int do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
999 {
1000         int ret;
1001
1002         if (!capable(CAP_NET_ADMIN))
1003                 return -EPERM;
1004
1005         switch (cmd) {
1006         case ARPT_SO_GET_INFO: {
1007                 char name[ARPT_TABLE_MAXNAMELEN];
1008                 struct arpt_table *t;
1009
1010                 if (*len != sizeof(struct arpt_getinfo)) {
1011                         duprintf("length %u != %Zu\n", *len,
1012                                  sizeof(struct arpt_getinfo));
1013                         ret = -EINVAL;
1014                         break;
1015                 }
1016
1017                 if (copy_from_user(name, user, sizeof(name)) != 0) {
1018                         ret = -EFAULT;
1019                         break;
1020                 }
1021                 name[ARPT_TABLE_MAXNAMELEN-1] = '\0';
1022
1023                 t = try_then_request_module(xt_find_table_lock(NF_ARP, name),
1024                                             "arptable_%s", name);
1025                 if (t && !IS_ERR(t)) {
1026                         struct arpt_getinfo info;
1027                         struct xt_table_info *private = t->private;
1028
1029                         info.valid_hooks = t->valid_hooks;
1030                         memcpy(info.hook_entry, private->hook_entry,
1031                                sizeof(info.hook_entry));
1032                         memcpy(info.underflow, private->underflow,
1033                                sizeof(info.underflow));
1034                         info.num_entries = private->number;
1035                         info.size = private->size;
1036                         strcpy(info.name, name);
1037
1038                         if (copy_to_user(user, &info, *len) != 0)
1039                                 ret = -EFAULT;
1040                         else
1041                                 ret = 0;
1042                         xt_table_unlock(t);
1043                         module_put(t->me);
1044                 } else
1045                         ret = t ? PTR_ERR(t) : -ENOENT;
1046         }
1047         break;
1048
1049         case ARPT_SO_GET_ENTRIES: {
1050                 struct arpt_get_entries get;
1051
1052                 if (*len < sizeof(get)) {
1053                         duprintf("get_entries: %u < %Zu\n", *len, sizeof(get));
1054                         ret = -EINVAL;
1055                 } else if (copy_from_user(&get, user, sizeof(get)) != 0) {
1056                         ret = -EFAULT;
1057                 } else if (*len != sizeof(struct arpt_get_entries) + get.size) {
1058                         duprintf("get_entries: %u != %Zu\n", *len,
1059                                  sizeof(struct arpt_get_entries) + get.size);
1060                         ret = -EINVAL;
1061                 } else
1062                         ret = get_entries(&get, user);
1063                 break;
1064         }
1065
1066         case ARPT_SO_GET_REVISION_TARGET: {
1067                 struct xt_get_revision rev;
1068
1069                 if (*len != sizeof(rev)) {
1070                         ret = -EINVAL;
1071                         break;
1072                 }
1073                 if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
1074                         ret = -EFAULT;
1075                         break;
1076                 }
1077
1078                 try_then_request_module(xt_find_revision(NF_ARP, rev.name,
1079                                                          rev.revision, 1, &ret),
1080                                         "arpt_%s", rev.name);
1081                 break;
1082         }
1083
1084         default:
1085                 duprintf("do_arpt_get_ctl: unknown request %i\n", cmd);
1086                 ret = -EINVAL;
1087         }
1088
1089         return ret;
1090 }
1091
1092 int arpt_register_table(struct arpt_table *table,
1093                         const struct arpt_replace *repl)
1094 {
1095         int ret;
1096         struct xt_table_info *newinfo;
1097         static struct xt_table_info bootstrap
1098                 = { 0, 0, 0, { 0 }, { 0 }, { } };
1099         void *loc_cpu_entry;
1100
1101         newinfo = xt_alloc_table_info(repl->size);
1102         if (!newinfo) {
1103                 ret = -ENOMEM;
1104                 return ret;
1105         }
1106
1107         /* choose the copy on our node/cpu */
1108         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1109         memcpy(loc_cpu_entry, repl->entries, repl->size);
1110
1111         ret = translate_table(table->name, table->valid_hooks,
1112                               newinfo, loc_cpu_entry, repl->size,
1113                               repl->num_entries,
1114                               repl->hook_entry,
1115                               repl->underflow);
1116
1117         duprintf("arpt_register_table: translate table gives %d\n", ret);
1118         if (ret != 0) {
1119                 xt_free_table_info(newinfo);
1120                 return ret;
1121         }
1122
1123         ret = xt_register_table(table, &bootstrap, newinfo);
1124         if (ret != 0) {
1125                 xt_free_table_info(newinfo);
1126                 return ret;
1127         }
1128
1129         return 0;
1130 }
1131
1132 void arpt_unregister_table(struct arpt_table *table)
1133 {
1134         struct xt_table_info *private;
1135         void *loc_cpu_entry;
1136
1137         private = xt_unregister_table(table);
1138
1139         /* Decrease module usage counts and free resources */
1140         loc_cpu_entry = private->entries[raw_smp_processor_id()];
1141         ARPT_ENTRY_ITERATE(loc_cpu_entry, private->size,
1142                            cleanup_entry, NULL);
1143         xt_free_table_info(private);
1144 }
1145
1146 /* The built-in targets: standard (NULL) and error. */
1147 static struct arpt_target arpt_standard_target = {
1148         .name           = ARPT_STANDARD_TARGET,
1149         .targetsize     = sizeof(int),
1150         .family         = NF_ARP,
1151 };
1152
1153 static struct arpt_target arpt_error_target = {
1154         .name           = ARPT_ERROR_TARGET,
1155         .target         = arpt_error,
1156         .targetsize     = ARPT_FUNCTION_MAXNAMELEN,
1157         .family         = NF_ARP,
1158 };
1159
1160 static struct nf_sockopt_ops arpt_sockopts = {
1161         .pf             = PF_INET,
1162         .set_optmin     = ARPT_BASE_CTL,
1163         .set_optmax     = ARPT_SO_SET_MAX+1,
1164         .set            = do_arpt_set_ctl,
1165         .get_optmin     = ARPT_BASE_CTL,
1166         .get_optmax     = ARPT_SO_GET_MAX+1,
1167         .get            = do_arpt_get_ctl,
1168 };
1169
1170 static int __init arp_tables_init(void)
1171 {
1172         int ret;
1173
1174         ret = xt_proto_init(NF_ARP);
1175         if (ret < 0)
1176                 goto err1;
1177
1178         /* Noone else will be downing sem now, so we won't sleep */
1179         ret = xt_register_target(&arpt_standard_target);
1180         if (ret < 0)
1181                 goto err2;
1182         ret = xt_register_target(&arpt_error_target);
1183         if (ret < 0)
1184                 goto err3;
1185
1186         /* Register setsockopt */
1187         ret = nf_register_sockopt(&arpt_sockopts);
1188         if (ret < 0)
1189                 goto err4;
1190
1191         printk("arp_tables: (C) 2002 David S. Miller\n");
1192         return 0;
1193
1194 err4:
1195         xt_unregister_target(&arpt_error_target);
1196 err3:
1197         xt_unregister_target(&arpt_standard_target);
1198 err2:
1199         xt_proto_fini(NF_ARP);
1200 err1:
1201         return ret;
1202 }
1203
1204 static void __exit arp_tables_fini(void)
1205 {
1206         nf_unregister_sockopt(&arpt_sockopts);
1207         xt_unregister_target(&arpt_error_target);
1208         xt_unregister_target(&arpt_standard_target);
1209         xt_proto_fini(NF_ARP);
1210 }
1211
1212 EXPORT_SYMBOL(arpt_register_table);
1213 EXPORT_SYMBOL(arpt_unregister_table);
1214 EXPORT_SYMBOL(arpt_do_table);
1215
1216 module_init(arp_tables_init);
1217 module_exit(arp_tables_fini);