[NETFILTER]: Add "revision" support to arp_tables and ip6_tables
[safe/jmp/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/config.h>
13 #include <linux/kernel.h>
14 #include <linux/skbuff.h>
15 #include <linux/netdevice.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 <asm/semaphore.h>
25
26 #include <linux/netfilter_arp/arp_tables.h>
27
28 MODULE_LICENSE("GPL");
29 MODULE_AUTHOR("David S. Miller <davem@redhat.com>");
30 MODULE_DESCRIPTION("arptables core");
31
32 /*#define DEBUG_ARP_TABLES*/
33 /*#define DEBUG_ARP_TABLES_USER*/
34
35 #ifdef DEBUG_ARP_TABLES
36 #define dprintf(format, args...)  printk(format , ## args)
37 #else
38 #define dprintf(format, args...)
39 #endif
40
41 #ifdef DEBUG_ARP_TABLES_USER
42 #define duprintf(format, args...) printk(format , ## args)
43 #else
44 #define duprintf(format, args...)
45 #endif
46
47 #ifdef CONFIG_NETFILTER_DEBUG
48 #define ARP_NF_ASSERT(x)                                        \
49 do {                                                            \
50         if (!(x))                                               \
51                 printk("ARP_NF_ASSERT: %s:%s:%u\n",             \
52                        __FUNCTION__, __FILE__, __LINE__);       \
53 } while(0)
54 #else
55 #define ARP_NF_ASSERT(x)
56 #endif
57 #define SMP_ALIGN(x) (((x) + SMP_CACHE_BYTES-1) & ~(SMP_CACHE_BYTES-1))
58
59 static DECLARE_MUTEX(arpt_mutex);
60
61 #define ASSERT_READ_LOCK(x) ARP_NF_ASSERT(down_trylock(&arpt_mutex) != 0)
62 #define ASSERT_WRITE_LOCK(x) ARP_NF_ASSERT(down_trylock(&arpt_mutex) != 0)
63 #include <linux/netfilter_ipv4/listhelp.h>
64
65 struct arpt_table_info {
66         unsigned int size;
67         unsigned int number;
68         unsigned int initial_entries;
69         unsigned int hook_entry[NF_ARP_NUMHOOKS];
70         unsigned int underflow[NF_ARP_NUMHOOKS];
71         char entries[0] __attribute__((aligned(SMP_CACHE_BYTES)));
72 };
73
74 static LIST_HEAD(arpt_target);
75 static LIST_HEAD(arpt_tables);
76 #define ADD_COUNTER(c,b,p) do { (c).bcnt += (b); (c).pcnt += (p); } while(0)
77
78 #ifdef CONFIG_SMP
79 #define TABLE_OFFSET(t,p) (SMP_ALIGN((t)->size)*(p))
80 #else
81 #define TABLE_OFFSET(t,p) 0
82 #endif
83
84 static inline int arp_devaddr_compare(const struct arpt_devaddr_info *ap,
85                                       char *hdr_addr, int len)
86 {
87         int i, ret;
88
89         if (len > ARPT_DEV_ADDR_LEN_MAX)
90                 len = ARPT_DEV_ADDR_LEN_MAX;
91
92         ret = 0;
93         for (i = 0; i < len; i++)
94                 ret |= (hdr_addr[i] ^ ap->addr[i]) & ap->mask[i];
95
96         return (ret != 0);
97 }
98
99 /* Returns whether packet matches rule or not. */
100 static inline int arp_packet_match(const struct arphdr *arphdr,
101                                    struct net_device *dev,
102                                    const char *indev,
103                                    const char *outdev,
104                                    const struct arpt_arp *arpinfo)
105 {
106         char *arpptr = (char *)(arphdr + 1);
107         char *src_devaddr, *tgt_devaddr;
108         u32 src_ipaddr, tgt_ipaddr;
109         int i, ret;
110
111 #define FWINV(bool,invflg) ((bool) ^ !!(arpinfo->invflags & invflg))
112
113         if (FWINV((arphdr->ar_op & arpinfo->arpop_mask) != arpinfo->arpop,
114                   ARPT_INV_ARPOP)) {
115                 dprintf("ARP operation field mismatch.\n");
116                 dprintf("ar_op: %04x info->arpop: %04x info->arpop_mask: %04x\n",
117                         arphdr->ar_op, arpinfo->arpop, arpinfo->arpop_mask);
118                 return 0;
119         }
120
121         if (FWINV((arphdr->ar_hrd & arpinfo->arhrd_mask) != arpinfo->arhrd,
122                   ARPT_INV_ARPHRD)) {
123                 dprintf("ARP hardware address format mismatch.\n");
124                 dprintf("ar_hrd: %04x info->arhrd: %04x info->arhrd_mask: %04x\n",
125                         arphdr->ar_hrd, arpinfo->arhrd, arpinfo->arhrd_mask);
126                 return 0;
127         }
128
129         if (FWINV((arphdr->ar_pro & arpinfo->arpro_mask) != arpinfo->arpro,
130                   ARPT_INV_ARPPRO)) {
131                 dprintf("ARP protocol address format mismatch.\n");
132                 dprintf("ar_pro: %04x info->arpro: %04x info->arpro_mask: %04x\n",
133                         arphdr->ar_pro, arpinfo->arpro, arpinfo->arpro_mask);
134                 return 0;
135         }
136
137         if (FWINV((arphdr->ar_hln & arpinfo->arhln_mask) != arpinfo->arhln,
138                   ARPT_INV_ARPHLN)) {
139                 dprintf("ARP hardware address length mismatch.\n");
140                 dprintf("ar_hln: %02x info->arhln: %02x info->arhln_mask: %02x\n",
141                         arphdr->ar_hln, arpinfo->arhln, arpinfo->arhln_mask);
142                 return 0;
143         }
144
145         src_devaddr = arpptr;
146         arpptr += dev->addr_len;
147         memcpy(&src_ipaddr, arpptr, sizeof(u32));
148         arpptr += sizeof(u32);
149         tgt_devaddr = arpptr;
150         arpptr += dev->addr_len;
151         memcpy(&tgt_ipaddr, arpptr, sizeof(u32));
152
153         if (FWINV(arp_devaddr_compare(&arpinfo->src_devaddr, src_devaddr, dev->addr_len),
154                   ARPT_INV_SRCDEVADDR) ||
155             FWINV(arp_devaddr_compare(&arpinfo->tgt_devaddr, tgt_devaddr, dev->addr_len),
156                   ARPT_INV_TGTDEVADDR)) {
157                 dprintf("Source or target device address mismatch.\n");
158
159                 return 0;
160         }
161
162         if (FWINV((src_ipaddr & arpinfo->smsk.s_addr) != arpinfo->src.s_addr,
163                   ARPT_INV_SRCIP) ||
164             FWINV(((tgt_ipaddr & arpinfo->tmsk.s_addr) != arpinfo->tgt.s_addr),
165                   ARPT_INV_TGTIP)) {
166                 dprintf("Source or target IP address mismatch.\n");
167
168                 dprintf("SRC: %u.%u.%u.%u. Mask: %u.%u.%u.%u. Target: %u.%u.%u.%u.%s\n",
169                         NIPQUAD(src_ipaddr),
170                         NIPQUAD(arpinfo->smsk.s_addr),
171                         NIPQUAD(arpinfo->src.s_addr),
172                         arpinfo->invflags & ARPT_INV_SRCIP ? " (INV)" : "");
173                 dprintf("TGT: %u.%u.%u.%u Mask: %u.%u.%u.%u Target: %u.%u.%u.%u.%s\n",
174                         NIPQUAD(tgt_ipaddr),
175                         NIPQUAD(arpinfo->tmsk.s_addr),
176                         NIPQUAD(arpinfo->tgt.s_addr),
177                         arpinfo->invflags & ARPT_INV_TGTIP ? " (INV)" : "");
178                 return 0;
179         }
180
181         /* Look for ifname matches.  */
182         for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
183                 ret |= (indev[i] ^ arpinfo->iniface[i])
184                         & arpinfo->iniface_mask[i];
185         }
186
187         if (FWINV(ret != 0, ARPT_INV_VIA_IN)) {
188                 dprintf("VIA in mismatch (%s vs %s).%s\n",
189                         indev, arpinfo->iniface,
190                         arpinfo->invflags&ARPT_INV_VIA_IN ?" (INV)":"");
191                 return 0;
192         }
193
194         for (i = 0, ret = 0; i < IFNAMSIZ/sizeof(unsigned long); i++) {
195                 unsigned long odev;
196                 memcpy(&odev, outdev + i*sizeof(unsigned long),
197                        sizeof(unsigned long));
198                 ret |= (odev
199                         ^ ((const unsigned long *)arpinfo->outiface)[i])
200                         & ((const unsigned long *)arpinfo->outiface_mask)[i];
201         }
202
203         if (FWINV(ret != 0, ARPT_INV_VIA_OUT)) {
204                 dprintf("VIA out mismatch (%s vs %s).%s\n",
205                         outdev, arpinfo->outiface,
206                         arpinfo->invflags&ARPT_INV_VIA_OUT ?" (INV)":"");
207                 return 0;
208         }
209
210         return 1;
211 }
212
213 static inline int arp_checkentry(const struct arpt_arp *arp)
214 {
215         if (arp->flags & ~ARPT_F_MASK) {
216                 duprintf("Unknown flag bits set: %08X\n",
217                          arp->flags & ~ARPT_F_MASK);
218                 return 0;
219         }
220         if (arp->invflags & ~ARPT_INV_MASK) {
221                 duprintf("Unknown invflag bits set: %08X\n",
222                          arp->invflags & ~ARPT_INV_MASK);
223                 return 0;
224         }
225
226         return 1;
227 }
228
229 static unsigned int arpt_error(struct sk_buff **pskb,
230                                unsigned int hooknum,
231                                const struct net_device *in,
232                                const struct net_device *out,
233                                const void *targinfo,
234                                void *userinfo)
235 {
236         if (net_ratelimit())
237                 printk("arp_tables: error: '%s'\n", (char *)targinfo);
238
239         return NF_DROP;
240 }
241
242 static inline struct arpt_entry *get_entry(void *base, unsigned int offset)
243 {
244         return (struct arpt_entry *)(base + offset);
245 }
246
247 unsigned int arpt_do_table(struct sk_buff **pskb,
248                            unsigned int hook,
249                            const struct net_device *in,
250                            const struct net_device *out,
251                            struct arpt_table *table,
252                            void *userdata)
253 {
254         static const char nulldevname[IFNAMSIZ];
255         unsigned int verdict = NF_DROP;
256         struct arphdr *arp;
257         int hotdrop = 0;
258         struct arpt_entry *e, *back;
259         const char *indev, *outdev;
260         void *table_base;
261
262         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
263         if (!pskb_may_pull((*pskb), (sizeof(struct arphdr) +
264                                      (2 * (*pskb)->dev->addr_len) +
265                                      (2 * sizeof(u32)))))
266                 return NF_DROP;
267
268         indev = in ? in->name : nulldevname;
269         outdev = out ? out->name : nulldevname;
270
271         read_lock_bh(&table->lock);
272         table_base = (void *)table->private->entries
273                 + TABLE_OFFSET(table->private,
274                                smp_processor_id());
275         e = get_entry(table_base, table->private->hook_entry[hook]);
276         back = get_entry(table_base, table->private->underflow[hook]);
277
278         arp = (*pskb)->nh.arph;
279         do {
280                 if (arp_packet_match(arp, (*pskb)->dev, indev, outdev, &e->arp)) {
281                         struct arpt_entry_target *t;
282                         int hdr_len;
283
284                         hdr_len = sizeof(*arp) + (2 * sizeof(struct in_addr)) +
285                                 (2 * (*pskb)->dev->addr_len);
286                         ADD_COUNTER(e->counters, hdr_len, 1);
287
288                         t = arpt_get_target(e);
289
290                         /* Standard target? */
291                         if (!t->u.kernel.target->target) {
292                                 int v;
293
294                                 v = ((struct arpt_standard_target *)t)->verdict;
295                                 if (v < 0) {
296                                         /* Pop from stack? */
297                                         if (v != ARPT_RETURN) {
298                                                 verdict = (unsigned)(-v) - 1;
299                                                 break;
300                                         }
301                                         e = back;
302                                         back = get_entry(table_base,
303                                                          back->comefrom);
304                                         continue;
305                                 }
306                                 if (table_base + v
307                                     != (void *)e + e->next_offset) {
308                                         /* Save old back ptr in next entry */
309                                         struct arpt_entry *next
310                                                 = (void *)e + e->next_offset;
311                                         next->comefrom =
312                                                 (void *)back - table_base;
313
314                                         /* set back pointer to next entry */
315                                         back = next;
316                                 }
317
318                                 e = get_entry(table_base, v);
319                         } else {
320                                 /* Targets which reenter must return
321                                  * abs. verdicts
322                                  */
323                                 verdict = t->u.kernel.target->target(pskb,
324                                                                      hook,
325                                                                      in, out,
326                                                                      t->data,
327                                                                      userdata);
328
329                                 /* Target might have changed stuff. */
330                                 arp = (*pskb)->nh.arph;
331
332                                 if (verdict == ARPT_CONTINUE)
333                                         e = (void *)e + e->next_offset;
334                                 else
335                                         /* Verdict */
336                                         break;
337                         }
338                 } else {
339                         e = (void *)e + e->next_offset;
340                 }
341         } while (!hotdrop);
342         read_unlock_bh(&table->lock);
343
344         if (hotdrop)
345                 return NF_DROP;
346         else
347                 return verdict;
348 }
349
350 /*
351  * These are weird, but module loading must not be done with mutex
352  * held (since they will register), and we have to have a single
353  * function to use try_then_request_module().
354  */
355
356 /* Find table by name, grabs mutex & ref.  Returns ERR_PTR() on error. */
357 static inline struct arpt_table *find_table_lock(const char *name)
358 {
359         struct arpt_table *t;
360
361         if (down_interruptible(&arpt_mutex) != 0)
362                 return ERR_PTR(-EINTR);
363
364         list_for_each_entry(t, &arpt_tables, list)
365                 if (strcmp(t->name, name) == 0 && try_module_get(t->me))
366                         return t;
367         up(&arpt_mutex);
368         return NULL;
369 }
370
371
372 /* Find target, grabs ref.  Returns ERR_PTR() on error. */
373 static inline struct arpt_target *find_target(const char *name, u8 revision)
374 {
375         struct arpt_target *t;
376         int err = 0;
377
378         if (down_interruptible(&arpt_mutex) != 0)
379                 return ERR_PTR(-EINTR);
380
381         list_for_each_entry(t, &arpt_target, list) {
382                 if (strcmp(t->name, name) == 0) {
383                         if (t->revision == revision) {
384                                 if (try_module_get(t->me)) {
385                                         up(&arpt_mutex);
386                                         return t;
387                                 }
388                         } else
389                                 err = -EPROTOTYPE; /* Found something. */
390                 }
391         }
392         up(&arpt_mutex);
393         return ERR_PTR(err);
394 }
395
396 struct arpt_target *arpt_find_target(const char *name, u8 revision)
397 {
398         struct arpt_target *target;
399
400         target = try_then_request_module(find_target(name, revision),
401                                          "arpt_%s", name);
402         if (IS_ERR(target) || !target)
403                 return NULL;
404         return target;
405 }
406
407 static int target_revfn(const char *name, u8 revision, int *bestp)
408 {
409         struct arpt_target *t;
410         int have_rev = 0;
411
412         list_for_each_entry(t, &arpt_target, list) {
413                 if (strcmp(t->name, name) == 0) {
414                         if (t->revision > *bestp)
415                                 *bestp = t->revision;
416                         if (t->revision == revision)
417                                 have_rev =1;
418                 }
419         }
420         return have_rev;
421 }
422
423 /* Returns true or false (if no such extension at all) */
424 static inline int find_revision(const char *name, u8 revision,
425                                 int (*revfn)(const char *, u8, int *),
426                                 int *err)
427 {
428         int have_rev, best = -1;
429
430         if (down_interruptible(&arpt_mutex) != 0) {
431                 *err = -EINTR;
432                 return 1;
433         }
434         have_rev = revfn(name, revision, &best);
435         up(&arpt_mutex);
436
437         /* Nothing at all?  Return 0 to try loading module. */
438         if (best == -1) {
439                 *err = -ENOENT;
440                 return 0;
441         }
442
443         *err = best;
444         if (!have_rev)
445                 *err = -EPROTONOSUPPORT;
446         return 1;
447 }
448
449
450 /* All zeroes == unconditional rule. */
451 static inline int unconditional(const struct arpt_arp *arp)
452 {
453         unsigned int i;
454
455         for (i = 0; i < sizeof(*arp)/sizeof(__u32); i++)
456                 if (((__u32 *)arp)[i])
457                         return 0;
458
459         return 1;
460 }
461
462 /* Figures out from what hook each rule can be called: returns 0 if
463  * there are loops.  Puts hook bitmask in comefrom.
464  */
465 static int mark_source_chains(struct arpt_table_info *newinfo, unsigned int valid_hooks)
466 {
467         unsigned int hook;
468
469         /* No recursion; use packet counter to save back ptrs (reset
470          * to 0 as we leave), and comefrom to save source hook bitmask.
471          */
472         for (hook = 0; hook < NF_ARP_NUMHOOKS; hook++) {
473                 unsigned int pos = newinfo->hook_entry[hook];
474                 struct arpt_entry *e
475                         = (struct arpt_entry *)(newinfo->entries + pos);
476
477                 if (!(valid_hooks & (1 << hook)))
478                         continue;
479
480                 /* Set initial back pointer. */
481                 e->counters.pcnt = pos;
482
483                 for (;;) {
484                         struct arpt_standard_target *t
485                                 = (void *)arpt_get_target(e);
486
487                         if (e->comefrom & (1 << NF_ARP_NUMHOOKS)) {
488                                 printk("arptables: loop hook %u pos %u %08X.\n",
489                                        hook, pos, e->comefrom);
490                                 return 0;
491                         }
492                         e->comefrom
493                                 |= ((1 << hook) | (1 << NF_ARP_NUMHOOKS));
494
495                         /* Unconditional return/END. */
496                         if (e->target_offset == sizeof(struct arpt_entry)
497                             && (strcmp(t->target.u.user.name,
498                                        ARPT_STANDARD_TARGET) == 0)
499                             && t->verdict < 0
500                             && unconditional(&e->arp)) {
501                                 unsigned int oldpos, size;
502
503                                 /* Return: backtrack through the last
504                                  * big jump.
505                                  */
506                                 do {
507                                         e->comefrom ^= (1<<NF_ARP_NUMHOOKS);
508                                         oldpos = pos;
509                                         pos = e->counters.pcnt;
510                                         e->counters.pcnt = 0;
511
512                                         /* We're at the start. */
513                                         if (pos == oldpos)
514                                                 goto next;
515
516                                         e = (struct arpt_entry *)
517                                                 (newinfo->entries + pos);
518                                 } while (oldpos == pos + e->next_offset);
519
520                                 /* Move along one */
521                                 size = e->next_offset;
522                                 e = (struct arpt_entry *)
523                                         (newinfo->entries + pos + size);
524                                 e->counters.pcnt = pos;
525                                 pos += size;
526                         } else {
527                                 int newpos = t->verdict;
528
529                                 if (strcmp(t->target.u.user.name,
530                                            ARPT_STANDARD_TARGET) == 0
531                                     && newpos >= 0) {
532                                         /* This a jump; chase it. */
533                                         duprintf("Jump rule %u -> %u\n",
534                                                  pos, newpos);
535                                 } else {
536                                         /* ... this is a fallthru */
537                                         newpos = pos + e->next_offset;
538                                 }
539                                 e = (struct arpt_entry *)
540                                         (newinfo->entries + newpos);
541                                 e->counters.pcnt = pos;
542                                 pos = newpos;
543                         }
544                 }
545                 next:
546                 duprintf("Finished chain %u\n", hook);
547         }
548         return 1;
549 }
550
551 static inline int standard_check(const struct arpt_entry_target *t,
552                                  unsigned int max_offset)
553 {
554         struct arpt_standard_target *targ = (void *)t;
555
556         /* Check standard info. */
557         if (t->u.target_size
558             != ARPT_ALIGN(sizeof(struct arpt_standard_target))) {
559                 duprintf("arpt_standard_check: target size %u != %Zu\n",
560                          t->u.target_size,
561                          ARPT_ALIGN(sizeof(struct arpt_standard_target)));
562                 return 0;
563         }
564
565         if (targ->verdict >= 0
566             && targ->verdict > max_offset - sizeof(struct arpt_entry)) {
567                 duprintf("arpt_standard_check: bad verdict (%i)\n",
568                          targ->verdict);
569                 return 0;
570         }
571
572         if (targ->verdict < -NF_MAX_VERDICT - 1) {
573                 duprintf("arpt_standard_check: bad negative verdict (%i)\n",
574                          targ->verdict);
575                 return 0;
576         }
577         return 1;
578 }
579
580 static struct arpt_target arpt_standard_target;
581
582 static inline int check_entry(struct arpt_entry *e, const char *name, unsigned int size,
583                               unsigned int *i)
584 {
585         struct arpt_entry_target *t;
586         struct arpt_target *target;
587         int ret;
588
589         if (!arp_checkentry(&e->arp)) {
590                 duprintf("arp_tables: arp check failed %p %s.\n", e, name);
591                 return -EINVAL;
592         }
593
594         t = arpt_get_target(e);
595         target = try_then_request_module(find_target(t->u.user.name,
596                                                      t->u.user.revision),
597                                          "arpt_%s", t->u.user.name);
598         if (IS_ERR(target) || !target) {
599                 duprintf("check_entry: `%s' not found\n", t->u.user.name);
600                 ret = target ? PTR_ERR(target) : -ENOENT;
601                 goto out;
602         }
603         t->u.kernel.target = target;
604
605         if (t->u.kernel.target == &arpt_standard_target) {
606                 if (!standard_check(t, size)) {
607                         ret = -EINVAL;
608                         goto out;
609                 }
610         } else if (t->u.kernel.target->checkentry
611                    && !t->u.kernel.target->checkentry(name, e, t->data,
612                                                       t->u.target_size
613                                                       - sizeof(*t),
614                                                       e->comefrom)) {
615                 module_put(t->u.kernel.target->me);
616                 duprintf("arp_tables: check failed for `%s'.\n",
617                          t->u.kernel.target->name);
618                 ret = -EINVAL;
619                 goto out;
620         }
621
622         (*i)++;
623         return 0;
624
625 out:
626         return ret;
627 }
628
629 static inline int check_entry_size_and_hooks(struct arpt_entry *e,
630                                              struct arpt_table_info *newinfo,
631                                              unsigned char *base,
632                                              unsigned char *limit,
633                                              const unsigned int *hook_entries,
634                                              const unsigned int *underflows,
635                                              unsigned int *i)
636 {
637         unsigned int h;
638
639         if ((unsigned long)e % __alignof__(struct arpt_entry) != 0
640             || (unsigned char *)e + sizeof(struct arpt_entry) >= limit) {
641                 duprintf("Bad offset %p\n", e);
642                 return -EINVAL;
643         }
644
645         if (e->next_offset
646             < sizeof(struct arpt_entry) + sizeof(struct arpt_entry_target)) {
647                 duprintf("checking: element %p size %u\n",
648                          e, e->next_offset);
649                 return -EINVAL;
650         }
651
652         /* Check hooks & underflows */
653         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
654                 if ((unsigned char *)e - base == hook_entries[h])
655                         newinfo->hook_entry[h] = hook_entries[h];
656                 if ((unsigned char *)e - base == underflows[h])
657                         newinfo->underflow[h] = underflows[h];
658         }
659
660         /* FIXME: underflows must be unconditional, standard verdicts
661            < 0 (not ARPT_RETURN). --RR */
662
663         /* Clear counters and comefrom */
664         e->counters = ((struct arpt_counters) { 0, 0 });
665         e->comefrom = 0;
666
667         (*i)++;
668         return 0;
669 }
670
671 static inline int cleanup_entry(struct arpt_entry *e, unsigned int *i)
672 {
673         struct arpt_entry_target *t;
674
675         if (i && (*i)-- == 0)
676                 return 1;
677
678         t = arpt_get_target(e);
679         if (t->u.kernel.target->destroy)
680                 t->u.kernel.target->destroy(t->data,
681                                             t->u.target_size - sizeof(*t));
682         module_put(t->u.kernel.target->me);
683         return 0;
684 }
685
686 /* Checks and translates the user-supplied table segment (held in
687  * newinfo).
688  */
689 static int translate_table(const char *name,
690                            unsigned int valid_hooks,
691                            struct arpt_table_info *newinfo,
692                            unsigned int size,
693                            unsigned int number,
694                            const unsigned int *hook_entries,
695                            const unsigned int *underflows)
696 {
697         unsigned int i;
698         int ret;
699
700         newinfo->size = size;
701         newinfo->number = number;
702
703         /* Init all hooks to impossible value. */
704         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
705                 newinfo->hook_entry[i] = 0xFFFFFFFF;
706                 newinfo->underflow[i] = 0xFFFFFFFF;
707         }
708
709         duprintf("translate_table: size %u\n", newinfo->size);
710         i = 0;
711
712         /* Walk through entries, checking offsets. */
713         ret = ARPT_ENTRY_ITERATE(newinfo->entries, newinfo->size,
714                                  check_entry_size_and_hooks,
715                                  newinfo,
716                                  newinfo->entries,
717                                  newinfo->entries + size,
718                                  hook_entries, underflows, &i);
719         duprintf("translate_table: ARPT_ENTRY_ITERATE gives %d\n", ret);
720         if (ret != 0)
721                 return ret;
722
723         if (i != number) {
724                 duprintf("translate_table: %u not %u entries\n",
725                          i, number);
726                 return -EINVAL;
727         }
728
729         /* Check hooks all assigned */
730         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
731                 /* Only hooks which are valid */
732                 if (!(valid_hooks & (1 << i)))
733                         continue;
734                 if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
735                         duprintf("Invalid hook entry %u %u\n",
736                                  i, hook_entries[i]);
737                         return -EINVAL;
738                 }
739                 if (newinfo->underflow[i] == 0xFFFFFFFF) {
740                         duprintf("Invalid underflow %u %u\n",
741                                  i, underflows[i]);
742                         return -EINVAL;
743                 }
744         }
745
746         if (!mark_source_chains(newinfo, valid_hooks)) {
747                 duprintf("Looping hook\n");
748                 return -ELOOP;
749         }
750
751         /* Finally, each sanity check must pass */
752         i = 0;
753         ret = ARPT_ENTRY_ITERATE(newinfo->entries, newinfo->size,
754                                  check_entry, name, size, &i);
755
756         if (ret != 0) {
757                 ARPT_ENTRY_ITERATE(newinfo->entries, newinfo->size,
758                                    cleanup_entry, &i);
759                 return ret;
760         }
761
762         /* And one copy for every other CPU */
763         for_each_cpu(i) {
764                 if (i == 0)
765                         continue;
766                 memcpy(newinfo->entries + SMP_ALIGN(newinfo->size) * i,
767                        newinfo->entries,
768                        SMP_ALIGN(newinfo->size));
769         }
770
771         return ret;
772 }
773
774 static struct arpt_table_info *replace_table(struct arpt_table *table,
775                                              unsigned int num_counters,
776                                              struct arpt_table_info *newinfo,
777                                              int *error)
778 {
779         struct arpt_table_info *oldinfo;
780
781         /* Do the substitution. */
782         write_lock_bh(&table->lock);
783         /* Check inside lock: is the old number correct? */
784         if (num_counters != table->private->number) {
785                 duprintf("num_counters != table->private->number (%u/%u)\n",
786                          num_counters, table->private->number);
787                 write_unlock_bh(&table->lock);
788                 *error = -EAGAIN;
789                 return NULL;
790         }
791         oldinfo = table->private;
792         table->private = newinfo;
793         newinfo->initial_entries = oldinfo->initial_entries;
794         write_unlock_bh(&table->lock);
795
796         return oldinfo;
797 }
798
799 /* Gets counters. */
800 static inline int add_entry_to_counter(const struct arpt_entry *e,
801                                        struct arpt_counters total[],
802                                        unsigned int *i)
803 {
804         ADD_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
805
806         (*i)++;
807         return 0;
808 }
809
810 static void get_counters(const struct arpt_table_info *t,
811                          struct arpt_counters counters[])
812 {
813         unsigned int cpu;
814         unsigned int i;
815
816         for_each_cpu(cpu) {
817                 i = 0;
818                 ARPT_ENTRY_ITERATE(t->entries + TABLE_OFFSET(t, cpu),
819                                    t->size,
820                                    add_entry_to_counter,
821                                    counters,
822                                    &i);
823         }
824 }
825
826 static int copy_entries_to_user(unsigned int total_size,
827                                 struct arpt_table *table,
828                                 void __user *userptr)
829 {
830         unsigned int off, num, countersize;
831         struct arpt_entry *e;
832         struct arpt_counters *counters;
833         int ret = 0;
834
835         /* We need atomic snapshot of counters: rest doesn't change
836          * (other than comefrom, which userspace doesn't care
837          * about).
838          */
839         countersize = sizeof(struct arpt_counters) * table->private->number;
840         counters = vmalloc(countersize);
841
842         if (counters == NULL)
843                 return -ENOMEM;
844
845         /* First, sum counters... */
846         memset(counters, 0, countersize);
847         write_lock_bh(&table->lock);
848         get_counters(table->private, counters);
849         write_unlock_bh(&table->lock);
850
851         /* ... then copy entire thing from CPU 0... */
852         if (copy_to_user(userptr, table->private->entries, total_size) != 0) {
853                 ret = -EFAULT;
854                 goto free_counters;
855         }
856
857         /* FIXME: use iterator macros --RR */
858         /* ... then go back and fix counters and names */
859         for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
860                 struct arpt_entry_target *t;
861
862                 e = (struct arpt_entry *)(table->private->entries + off);
863                 if (copy_to_user(userptr + off
864                                  + offsetof(struct arpt_entry, counters),
865                                  &counters[num],
866                                  sizeof(counters[num])) != 0) {
867                         ret = -EFAULT;
868                         goto free_counters;
869                 }
870
871                 t = arpt_get_target(e);
872                 if (copy_to_user(userptr + off + e->target_offset
873                                  + offsetof(struct arpt_entry_target,
874                                             u.user.name),
875                                  t->u.kernel.target->name,
876                                  strlen(t->u.kernel.target->name)+1) != 0) {
877                         ret = -EFAULT;
878                         goto free_counters;
879                 }
880         }
881
882  free_counters:
883         vfree(counters);
884         return ret;
885 }
886
887 static int get_entries(const struct arpt_get_entries *entries,
888                        struct arpt_get_entries __user *uptr)
889 {
890         int ret;
891         struct arpt_table *t;
892
893         t = find_table_lock(entries->name);
894         if (t || !IS_ERR(t)) {
895                 duprintf("t->private->number = %u\n",
896                          t->private->number);
897                 if (entries->size == t->private->size)
898                         ret = copy_entries_to_user(t->private->size,
899                                                    t, uptr->entrytable);
900                 else {
901                         duprintf("get_entries: I've got %u not %u!\n",
902                                  t->private->size,
903                                  entries->size);
904                         ret = -EINVAL;
905                 }
906                 module_put(t->me);
907                 up(&arpt_mutex);
908         } else
909                 ret = t ? PTR_ERR(t) : -ENOENT;
910
911         return ret;
912 }
913
914 static int do_replace(void __user *user, unsigned int len)
915 {
916         int ret;
917         struct arpt_replace tmp;
918         struct arpt_table *t;
919         struct arpt_table_info *newinfo, *oldinfo;
920         struct arpt_counters *counters;
921
922         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
923                 return -EFAULT;
924
925         /* Hack: Causes ipchains to give correct error msg --RR */
926         if (len != sizeof(tmp) + tmp.size)
927                 return -ENOPROTOOPT;
928
929         /* Pedantry: prevent them from hitting BUG() in vmalloc.c --RR */
930         if ((SMP_ALIGN(tmp.size) >> PAGE_SHIFT) + 2 > num_physpages)
931                 return -ENOMEM;
932
933         newinfo = vmalloc(sizeof(struct arpt_table_info)
934                           + SMP_ALIGN(tmp.size) *
935                                         (highest_possible_processor_id()+1));
936         if (!newinfo)
937                 return -ENOMEM;
938
939         if (copy_from_user(newinfo->entries, user + sizeof(tmp),
940                            tmp.size) != 0) {
941                 ret = -EFAULT;
942                 goto free_newinfo;
943         }
944
945         counters = vmalloc(tmp.num_counters * sizeof(struct arpt_counters));
946         if (!counters) {
947                 ret = -ENOMEM;
948                 goto free_newinfo;
949         }
950         memset(counters, 0, tmp.num_counters * sizeof(struct arpt_counters));
951
952         ret = translate_table(tmp.name, tmp.valid_hooks,
953                               newinfo, tmp.size, tmp.num_entries,
954                               tmp.hook_entry, tmp.underflow);
955         if (ret != 0)
956                 goto free_newinfo_counters;
957
958         duprintf("arp_tables: Translated table\n");
959
960         t = try_then_request_module(find_table_lock(tmp.name),
961                                     "arptable_%s", tmp.name);
962         if (!t || IS_ERR(t)) {
963                 ret = t ? PTR_ERR(t) : -ENOENT;
964                 goto free_newinfo_counters_untrans;
965         }
966
967         /* You lied! */
968         if (tmp.valid_hooks != t->valid_hooks) {
969                 duprintf("Valid hook crap: %08X vs %08X\n",
970                          tmp.valid_hooks, t->valid_hooks);
971                 ret = -EINVAL;
972                 goto put_module;
973         }
974
975         oldinfo = replace_table(t, tmp.num_counters, newinfo, &ret);
976         if (!oldinfo)
977                 goto put_module;
978
979         /* Update module usage count based on number of rules */
980         duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
981                 oldinfo->number, oldinfo->initial_entries, newinfo->number);
982         if ((oldinfo->number > oldinfo->initial_entries) || 
983             (newinfo->number <= oldinfo->initial_entries)) 
984                 module_put(t->me);
985         if ((oldinfo->number > oldinfo->initial_entries) &&
986             (newinfo->number <= oldinfo->initial_entries))
987                 module_put(t->me);
988
989         /* Get the old counters. */
990         get_counters(oldinfo, counters);
991         /* Decrease module usage counts and free resource */
992         ARPT_ENTRY_ITERATE(oldinfo->entries, oldinfo->size, cleanup_entry,NULL);
993         vfree(oldinfo);
994         if (copy_to_user(tmp.counters, counters,
995                          sizeof(struct arpt_counters) * tmp.num_counters) != 0)
996                 ret = -EFAULT;
997         vfree(counters);
998         up(&arpt_mutex);
999         return ret;
1000
1001  put_module:
1002         module_put(t->me);
1003         up(&arpt_mutex);
1004  free_newinfo_counters_untrans:
1005         ARPT_ENTRY_ITERATE(newinfo->entries, newinfo->size, cleanup_entry, NULL);
1006  free_newinfo_counters:
1007         vfree(counters);
1008  free_newinfo:
1009         vfree(newinfo);
1010         return ret;
1011 }
1012
1013 /* We're lazy, and add to the first CPU; overflow works its fey magic
1014  * and everything is OK.
1015  */
1016 static inline int add_counter_to_entry(struct arpt_entry *e,
1017                                        const struct arpt_counters addme[],
1018                                        unsigned int *i)
1019 {
1020
1021         ADD_COUNTER(e->counters, addme[*i].bcnt, addme[*i].pcnt);
1022
1023         (*i)++;
1024         return 0;
1025 }
1026
1027 static int do_add_counters(void __user *user, unsigned int len)
1028 {
1029         unsigned int i;
1030         struct arpt_counters_info tmp, *paddc;
1031         struct arpt_table *t;
1032         int ret = 0;
1033
1034         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
1035                 return -EFAULT;
1036
1037         if (len != sizeof(tmp) + tmp.num_counters*sizeof(struct arpt_counters))
1038                 return -EINVAL;
1039
1040         paddc = vmalloc(len);
1041         if (!paddc)
1042                 return -ENOMEM;
1043
1044         if (copy_from_user(paddc, user, len) != 0) {
1045                 ret = -EFAULT;
1046                 goto free;
1047         }
1048
1049         t = find_table_lock(tmp.name);
1050         if (!t || IS_ERR(t)) {
1051                 ret = t ? PTR_ERR(t) : -ENOENT;
1052                 goto free;
1053         }
1054
1055         write_lock_bh(&t->lock);
1056         if (t->private->number != paddc->num_counters) {
1057                 ret = -EINVAL;
1058                 goto unlock_up_free;
1059         }
1060
1061         i = 0;
1062         ARPT_ENTRY_ITERATE(t->private->entries,
1063                            t->private->size,
1064                            add_counter_to_entry,
1065                            paddc->counters,
1066                            &i);
1067  unlock_up_free:
1068         write_unlock_bh(&t->lock);
1069         up(&arpt_mutex);
1070         module_put(t->me);
1071  free:
1072         vfree(paddc);
1073
1074         return ret;
1075 }
1076
1077 static int do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
1078 {
1079         int ret;
1080
1081         if (!capable(CAP_NET_ADMIN))
1082                 return -EPERM;
1083
1084         switch (cmd) {
1085         case ARPT_SO_SET_REPLACE:
1086                 ret = do_replace(user, len);
1087                 break;
1088
1089         case ARPT_SO_SET_ADD_COUNTERS:
1090                 ret = do_add_counters(user, len);
1091                 break;
1092
1093         default:
1094                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
1095                 ret = -EINVAL;
1096         }
1097
1098         return ret;
1099 }
1100
1101 static int do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
1102 {
1103         int ret;
1104
1105         if (!capable(CAP_NET_ADMIN))
1106                 return -EPERM;
1107
1108         switch (cmd) {
1109         case ARPT_SO_GET_INFO: {
1110                 char name[ARPT_TABLE_MAXNAMELEN];
1111                 struct arpt_table *t;
1112
1113                 if (*len != sizeof(struct arpt_getinfo)) {
1114                         duprintf("length %u != %Zu\n", *len,
1115                                  sizeof(struct arpt_getinfo));
1116                         ret = -EINVAL;
1117                         break;
1118                 }
1119
1120                 if (copy_from_user(name, user, sizeof(name)) != 0) {
1121                         ret = -EFAULT;
1122                         break;
1123                 }
1124                 name[ARPT_TABLE_MAXNAMELEN-1] = '\0';
1125
1126                 t = try_then_request_module(find_table_lock(name),
1127                                             "arptable_%s", name);
1128                 if (t && !IS_ERR(t)) {
1129                         struct arpt_getinfo info;
1130
1131                         info.valid_hooks = t->valid_hooks;
1132                         memcpy(info.hook_entry, t->private->hook_entry,
1133                                sizeof(info.hook_entry));
1134                         memcpy(info.underflow, t->private->underflow,
1135                                sizeof(info.underflow));
1136                         info.num_entries = t->private->number;
1137                         info.size = t->private->size;
1138                         strcpy(info.name, name);
1139
1140                         if (copy_to_user(user, &info, *len) != 0)
1141                                 ret = -EFAULT;
1142                         else
1143                                 ret = 0;
1144                         up(&arpt_mutex);
1145                         module_put(t->me);
1146                 } else
1147                         ret = t ? PTR_ERR(t) : -ENOENT;
1148         }
1149         break;
1150
1151         case ARPT_SO_GET_ENTRIES: {
1152                 struct arpt_get_entries get;
1153
1154                 if (*len < sizeof(get)) {
1155                         duprintf("get_entries: %u < %Zu\n", *len, sizeof(get));
1156                         ret = -EINVAL;
1157                 } else if (copy_from_user(&get, user, sizeof(get)) != 0) {
1158                         ret = -EFAULT;
1159                 } else if (*len != sizeof(struct arpt_get_entries) + get.size) {
1160                         duprintf("get_entries: %u != %Zu\n", *len,
1161                                  sizeof(struct arpt_get_entries) + get.size);
1162                         ret = -EINVAL;
1163                 } else
1164                         ret = get_entries(&get, user);
1165                 break;
1166         }
1167
1168         case ARPT_SO_GET_REVISION_TARGET: {
1169                 struct arpt_get_revision rev;
1170
1171                 if (*len != sizeof(rev)) {
1172                         ret = -EINVAL;
1173                         break;
1174                 }
1175                 if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
1176                         ret = -EFAULT;
1177                         break;
1178                 }
1179
1180                 try_then_request_module(find_revision(rev.name, rev.revision,
1181                                                       target_revfn, &ret),
1182                                         "arpt_%s", rev.name);
1183                 break;
1184         }
1185
1186         default:
1187                 duprintf("do_arpt_get_ctl: unknown request %i\n", cmd);
1188                 ret = -EINVAL;
1189         }
1190
1191         return ret;
1192 }
1193
1194 /* Registration hooks for targets. */
1195 int arpt_register_target(struct arpt_target *target)
1196 {
1197         int ret;
1198
1199         ret = down_interruptible(&arpt_mutex);
1200         if (ret != 0)
1201                 return ret;
1202
1203         list_add(&target->list, &arpt_target);
1204         up(&arpt_mutex);
1205
1206         return ret;
1207 }
1208
1209 void arpt_unregister_target(struct arpt_target *target)
1210 {
1211         down(&arpt_mutex);
1212         LIST_DELETE(&arpt_target, target);
1213         up(&arpt_mutex);
1214 }
1215
1216 int arpt_register_table(struct arpt_table *table,
1217                         const struct arpt_replace *repl)
1218 {
1219         int ret;
1220         struct arpt_table_info *newinfo;
1221         static struct arpt_table_info bootstrap
1222                 = { 0, 0, 0, { 0 }, { 0 }, { } };
1223
1224         newinfo = vmalloc(sizeof(struct arpt_table_info)
1225                           + SMP_ALIGN(repl->size) *
1226                                         (highest_possible_processor_id()+1));
1227         if (!newinfo) {
1228                 ret = -ENOMEM;
1229                 return ret;
1230         }
1231         memcpy(newinfo->entries, repl->entries, repl->size);
1232
1233         ret = translate_table(table->name, table->valid_hooks,
1234                               newinfo, repl->size,
1235                               repl->num_entries,
1236                               repl->hook_entry,
1237                               repl->underflow);
1238         duprintf("arpt_register_table: translate table gives %d\n", ret);
1239         if (ret != 0) {
1240                 vfree(newinfo);
1241                 return ret;
1242         }
1243
1244         ret = down_interruptible(&arpt_mutex);
1245         if (ret != 0) {
1246                 vfree(newinfo);
1247                 return ret;
1248         }
1249
1250         /* Don't autoload: we'd eat our tail... */
1251         if (list_named_find(&arpt_tables, table->name)) {
1252                 ret = -EEXIST;
1253                 goto free_unlock;
1254         }
1255
1256         /* Simplifies replace_table code. */
1257         table->private = &bootstrap;
1258         if (!replace_table(table, 0, newinfo, &ret))
1259                 goto free_unlock;
1260
1261         duprintf("table->private->number = %u\n",
1262                  table->private->number);
1263         
1264         /* save number of initial entries */
1265         table->private->initial_entries = table->private->number;
1266
1267         rwlock_init(&table->lock);
1268         list_prepend(&arpt_tables, table);
1269
1270  unlock:
1271         up(&arpt_mutex);
1272         return ret;
1273
1274  free_unlock:
1275         vfree(newinfo);
1276         goto unlock;
1277 }
1278
1279 void arpt_unregister_table(struct arpt_table *table)
1280 {
1281         down(&arpt_mutex);
1282         LIST_DELETE(&arpt_tables, table);
1283         up(&arpt_mutex);
1284
1285         /* Decrease module usage counts and free resources */
1286         ARPT_ENTRY_ITERATE(table->private->entries, table->private->size,
1287                            cleanup_entry, NULL);
1288         vfree(table->private);
1289 }
1290
1291 /* The built-in targets: standard (NULL) and error. */
1292 static struct arpt_target arpt_standard_target = {
1293         .name           = ARPT_STANDARD_TARGET,
1294 };
1295
1296 static struct arpt_target arpt_error_target = {
1297         .name           = ARPT_ERROR_TARGET,
1298         .target         = arpt_error,
1299 };
1300
1301 static struct nf_sockopt_ops arpt_sockopts = {
1302         .pf             = PF_INET,
1303         .set_optmin     = ARPT_BASE_CTL,
1304         .set_optmax     = ARPT_SO_SET_MAX+1,
1305         .set            = do_arpt_set_ctl,
1306         .get_optmin     = ARPT_BASE_CTL,
1307         .get_optmax     = ARPT_SO_GET_MAX+1,
1308         .get            = do_arpt_get_ctl,
1309 };
1310
1311 #ifdef CONFIG_PROC_FS
1312 static inline int print_name(const struct arpt_table *t,
1313                              off_t start_offset, char *buffer, int length,
1314                              off_t *pos, unsigned int *count)
1315 {
1316         if ((*count)++ >= start_offset) {
1317                 unsigned int namelen;
1318
1319                 namelen = sprintf(buffer + *pos, "%s\n", t->name);
1320                 if (*pos + namelen > length) {
1321                         /* Stop iterating */
1322                         return 1;
1323                 }
1324                 *pos += namelen;
1325         }
1326         return 0;
1327 }
1328
1329 static int arpt_get_tables(char *buffer, char **start, off_t offset, int length)
1330 {
1331         off_t pos = 0;
1332         unsigned int count = 0;
1333
1334         if (down_interruptible(&arpt_mutex) != 0)
1335                 return 0;
1336
1337         LIST_FIND(&arpt_tables, print_name, struct arpt_table *,
1338                   offset, buffer, length, &pos, &count);
1339
1340         up(&arpt_mutex);
1341
1342         /* `start' hack - see fs/proc/generic.c line ~105 */
1343         *start=(char *)((unsigned long)count-offset);
1344         return pos;
1345 }
1346 #endif /*CONFIG_PROC_FS*/
1347
1348 static int __init init(void)
1349 {
1350         int ret;
1351
1352         /* Noone else will be downing sem now, so we won't sleep */
1353         down(&arpt_mutex);
1354         list_append(&arpt_target, &arpt_standard_target);
1355         list_append(&arpt_target, &arpt_error_target);
1356         up(&arpt_mutex);
1357
1358         /* Register setsockopt */
1359         ret = nf_register_sockopt(&arpt_sockopts);
1360         if (ret < 0) {
1361                 duprintf("Unable to register sockopts.\n");
1362                 return ret;
1363         }
1364
1365 #ifdef CONFIG_PROC_FS
1366         {
1367                 struct proc_dir_entry *proc;
1368
1369                 proc = proc_net_create("arp_tables_names", 0, arpt_get_tables);
1370                 if (!proc) {
1371                         nf_unregister_sockopt(&arpt_sockopts);
1372                         return -ENOMEM;
1373                 }
1374                 proc->owner = THIS_MODULE;
1375         }
1376 #endif
1377
1378         printk("arp_tables: (C) 2002 David S. Miller\n");
1379         return 0;
1380 }
1381
1382 static void __exit fini(void)
1383 {
1384         nf_unregister_sockopt(&arpt_sockopts);
1385 #ifdef CONFIG_PROC_FS
1386         proc_net_remove("arp_tables_names");
1387 #endif
1388 }
1389
1390 EXPORT_SYMBOL(arpt_register_table);
1391 EXPORT_SYMBOL(arpt_unregister_table);
1392 EXPORT_SYMBOL(arpt_do_table);
1393 EXPORT_SYMBOL(arpt_register_target);
1394 EXPORT_SYMBOL(arpt_unregister_target);
1395
1396 module_init(init);
1397 module_exit(fini);