netfilter: xtables: use percpu rule counters
[firefly-linux-kernel-4.4.55.git] / 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  * Copyright (C) 2006-2009 Patrick McHardy <kaber@trash.net>
10  *
11  */
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/kernel.h>
14 #include <linux/skbuff.h>
15 #include <linux/netdevice.h>
16 #include <linux/capability.h>
17 #include <linux/if_arp.h>
18 #include <linux/kmod.h>
19 #include <linux/vmalloc.h>
20 #include <linux/proc_fs.h>
21 #include <linux/module.h>
22 #include <linux/init.h>
23 #include <linux/mutex.h>
24 #include <linux/err.h>
25 #include <net/compat.h>
26 #include <net/sock.h>
27 #include <asm/uaccess.h>
28
29 #include <linux/netfilter/x_tables.h>
30 #include <linux/netfilter_arp/arp_tables.h>
31 #include "../../netfilter/xt_repldata.h"
32
33 MODULE_LICENSE("GPL");
34 MODULE_AUTHOR("David S. Miller <davem@redhat.com>");
35 MODULE_DESCRIPTION("arptables core");
36
37 /*#define DEBUG_ARP_TABLES*/
38 /*#define DEBUG_ARP_TABLES_USER*/
39
40 #ifdef DEBUG_ARP_TABLES
41 #define dprintf(format, args...)  printk(format , ## args)
42 #else
43 #define dprintf(format, args...)
44 #endif
45
46 #ifdef DEBUG_ARP_TABLES_USER
47 #define duprintf(format, args...) printk(format , ## args)
48 #else
49 #define duprintf(format, args...)
50 #endif
51
52 #ifdef CONFIG_NETFILTER_DEBUG
53 #define ARP_NF_ASSERT(x)        WARN_ON(!(x))
54 #else
55 #define ARP_NF_ASSERT(x)
56 #endif
57
58 void *arpt_alloc_initial_table(const struct xt_table *info)
59 {
60         return xt_alloc_initial_table(arpt, ARPT);
61 }
62 EXPORT_SYMBOL_GPL(arpt_alloc_initial_table);
63
64 static inline int arp_devaddr_compare(const struct arpt_devaddr_info *ap,
65                                       const char *hdr_addr, int len)
66 {
67         int i, ret;
68
69         if (len > ARPT_DEV_ADDR_LEN_MAX)
70                 len = ARPT_DEV_ADDR_LEN_MAX;
71
72         ret = 0;
73         for (i = 0; i < len; i++)
74                 ret |= (hdr_addr[i] ^ ap->addr[i]) & ap->mask[i];
75
76         return ret != 0;
77 }
78
79 /*
80  * Unfortunately, _b and _mask are not aligned to an int (or long int)
81  * Some arches dont care, unrolling the loop is a win on them.
82  * For other arches, we only have a 16bit alignement.
83  */
84 static unsigned long ifname_compare(const char *_a, const char *_b, const char *_mask)
85 {
86 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
87         unsigned long ret = ifname_compare_aligned(_a, _b, _mask);
88 #else
89         unsigned long ret = 0;
90         const u16 *a = (const u16 *)_a;
91         const u16 *b = (const u16 *)_b;
92         const u16 *mask = (const u16 *)_mask;
93         int i;
94
95         for (i = 0; i < IFNAMSIZ/sizeof(u16); i++)
96                 ret |= (a[i] ^ b[i]) & mask[i];
97 #endif
98         return ret;
99 }
100
101 /* Returns whether packet matches rule or not. */
102 static inline int arp_packet_match(const struct arphdr *arphdr,
103                                    struct net_device *dev,
104                                    const char *indev,
105                                    const char *outdev,
106                                    const struct arpt_arp *arpinfo)
107 {
108         const char *arpptr = (char *)(arphdr + 1);
109         const char *src_devaddr, *tgt_devaddr;
110         __be32 src_ipaddr, tgt_ipaddr;
111         long ret;
112
113 #define FWINV(bool, invflg) ((bool) ^ !!(arpinfo->invflags & (invflg)))
114
115         if (FWINV((arphdr->ar_op & arpinfo->arpop_mask) != arpinfo->arpop,
116                   ARPT_INV_ARPOP)) {
117                 dprintf("ARP operation field mismatch.\n");
118                 dprintf("ar_op: %04x info->arpop: %04x info->arpop_mask: %04x\n",
119                         arphdr->ar_op, arpinfo->arpop, arpinfo->arpop_mask);
120                 return 0;
121         }
122
123         if (FWINV((arphdr->ar_hrd & arpinfo->arhrd_mask) != arpinfo->arhrd,
124                   ARPT_INV_ARPHRD)) {
125                 dprintf("ARP hardware address format mismatch.\n");
126                 dprintf("ar_hrd: %04x info->arhrd: %04x info->arhrd_mask: %04x\n",
127                         arphdr->ar_hrd, arpinfo->arhrd, arpinfo->arhrd_mask);
128                 return 0;
129         }
130
131         if (FWINV((arphdr->ar_pro & arpinfo->arpro_mask) != arpinfo->arpro,
132                   ARPT_INV_ARPPRO)) {
133                 dprintf("ARP protocol address format mismatch.\n");
134                 dprintf("ar_pro: %04x info->arpro: %04x info->arpro_mask: %04x\n",
135                         arphdr->ar_pro, arpinfo->arpro, arpinfo->arpro_mask);
136                 return 0;
137         }
138
139         if (FWINV((arphdr->ar_hln & arpinfo->arhln_mask) != arpinfo->arhln,
140                   ARPT_INV_ARPHLN)) {
141                 dprintf("ARP hardware address length mismatch.\n");
142                 dprintf("ar_hln: %02x info->arhln: %02x info->arhln_mask: %02x\n",
143                         arphdr->ar_hln, arpinfo->arhln, arpinfo->arhln_mask);
144                 return 0;
145         }
146
147         src_devaddr = arpptr;
148         arpptr += dev->addr_len;
149         memcpy(&src_ipaddr, arpptr, sizeof(u32));
150         arpptr += sizeof(u32);
151         tgt_devaddr = arpptr;
152         arpptr += dev->addr_len;
153         memcpy(&tgt_ipaddr, arpptr, sizeof(u32));
154
155         if (FWINV(arp_devaddr_compare(&arpinfo->src_devaddr, src_devaddr, dev->addr_len),
156                   ARPT_INV_SRCDEVADDR) ||
157             FWINV(arp_devaddr_compare(&arpinfo->tgt_devaddr, tgt_devaddr, dev->addr_len),
158                   ARPT_INV_TGTDEVADDR)) {
159                 dprintf("Source or target device address mismatch.\n");
160
161                 return 0;
162         }
163
164         if (FWINV((src_ipaddr & arpinfo->smsk.s_addr) != arpinfo->src.s_addr,
165                   ARPT_INV_SRCIP) ||
166             FWINV(((tgt_ipaddr & arpinfo->tmsk.s_addr) != arpinfo->tgt.s_addr),
167                   ARPT_INV_TGTIP)) {
168                 dprintf("Source or target IP address mismatch.\n");
169
170                 dprintf("SRC: %pI4. Mask: %pI4. Target: %pI4.%s\n",
171                         &src_ipaddr,
172                         &arpinfo->smsk.s_addr,
173                         &arpinfo->src.s_addr,
174                         arpinfo->invflags & ARPT_INV_SRCIP ? " (INV)" : "");
175                 dprintf("TGT: %pI4 Mask: %pI4 Target: %pI4.%s\n",
176                         &tgt_ipaddr,
177                         &arpinfo->tmsk.s_addr,
178                         &arpinfo->tgt.s_addr,
179                         arpinfo->invflags & ARPT_INV_TGTIP ? " (INV)" : "");
180                 return 0;
181         }
182
183         /* Look for ifname matches.  */
184         ret = ifname_compare(indev, arpinfo->iniface, arpinfo->iniface_mask);
185
186         if (FWINV(ret != 0, ARPT_INV_VIA_IN)) {
187                 dprintf("VIA in mismatch (%s vs %s).%s\n",
188                         indev, arpinfo->iniface,
189                         arpinfo->invflags&ARPT_INV_VIA_IN ?" (INV)":"");
190                 return 0;
191         }
192
193         ret = ifname_compare(outdev, arpinfo->outiface, arpinfo->outiface_mask);
194
195         if (FWINV(ret != 0, ARPT_INV_VIA_OUT)) {
196                 dprintf("VIA out mismatch (%s vs %s).%s\n",
197                         outdev, arpinfo->outiface,
198                         arpinfo->invflags&ARPT_INV_VIA_OUT ?" (INV)":"");
199                 return 0;
200         }
201
202         return 1;
203 #undef FWINV
204 }
205
206 static inline int arp_checkentry(const struct arpt_arp *arp)
207 {
208         if (arp->flags & ~ARPT_F_MASK) {
209                 duprintf("Unknown flag bits set: %08X\n",
210                          arp->flags & ~ARPT_F_MASK);
211                 return 0;
212         }
213         if (arp->invflags & ~ARPT_INV_MASK) {
214                 duprintf("Unknown invflag bits set: %08X\n",
215                          arp->invflags & ~ARPT_INV_MASK);
216                 return 0;
217         }
218
219         return 1;
220 }
221
222 static unsigned int
223 arpt_error(struct sk_buff *skb, const struct xt_action_param *par)
224 {
225         net_err_ratelimited("arp_tables: error: '%s'\n",
226                             (const char *)par->targinfo);
227
228         return NF_DROP;
229 }
230
231 static inline const struct xt_entry_target *
232 arpt_get_target_c(const struct arpt_entry *e)
233 {
234         return arpt_get_target((struct arpt_entry *)e);
235 }
236
237 static inline struct arpt_entry *
238 get_entry(const void *base, unsigned int offset)
239 {
240         return (struct arpt_entry *)(base + offset);
241 }
242
243 static inline __pure
244 struct arpt_entry *arpt_next_entry(const struct arpt_entry *entry)
245 {
246         return (void *)entry + entry->next_offset;
247 }
248
249 unsigned int arpt_do_table(struct sk_buff *skb,
250                            unsigned int hook,
251                            const struct nf_hook_state *state,
252                            struct xt_table *table)
253 {
254         static const char nulldevname[IFNAMSIZ] __attribute__((aligned(sizeof(long))));
255         unsigned int verdict = NF_DROP;
256         const struct arphdr *arp;
257         struct arpt_entry *e, *back;
258         const char *indev, *outdev;
259         void *table_base;
260         const struct xt_table_info *private;
261         struct xt_action_param acpar;
262         unsigned int addend;
263
264         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
265                 return NF_DROP;
266
267         indev = state->in ? state->in->name : nulldevname;
268         outdev = state->out ? state->out->name : nulldevname;
269
270         local_bh_disable();
271         addend = xt_write_recseq_begin();
272         private = table->private;
273         /*
274          * Ensure we load private-> members after we've fetched the base
275          * pointer.
276          */
277         smp_read_barrier_depends();
278         table_base = private->entries[smp_processor_id()];
279
280         e = get_entry(table_base, private->hook_entry[hook]);
281         back = get_entry(table_base, private->underflow[hook]);
282
283         acpar.in      = state->in;
284         acpar.out     = state->out;
285         acpar.hooknum = hook;
286         acpar.family  = NFPROTO_ARP;
287         acpar.hotdrop = false;
288
289         arp = arp_hdr(skb);
290         do {
291                 const struct xt_entry_target *t;
292                 struct xt_counters *counter;
293
294                 if (!arp_packet_match(arp, skb->dev, indev, outdev, &e->arp)) {
295                         e = arpt_next_entry(e);
296                         continue;
297                 }
298
299                 counter = xt_get_this_cpu_counter(&e->counters);
300                 ADD_COUNTER(*counter, arp_hdr_len(skb->dev), 1);
301
302                 t = arpt_get_target_c(e);
303
304                 /* Standard target? */
305                 if (!t->u.kernel.target->target) {
306                         int v;
307
308                         v = ((struct xt_standard_target *)t)->verdict;
309                         if (v < 0) {
310                                 /* Pop from stack? */
311                                 if (v != XT_RETURN) {
312                                         verdict = (unsigned int)(-v) - 1;
313                                         break;
314                                 }
315                                 e = back;
316                                 back = get_entry(table_base, back->comefrom);
317                                 continue;
318                         }
319                         if (table_base + v
320                             != arpt_next_entry(e)) {
321                                 /* Save old back ptr in next entry */
322                                 struct arpt_entry *next = arpt_next_entry(e);
323                                 next->comefrom = (void *)back - table_base;
324
325                                 /* set back pointer to next entry */
326                                 back = next;
327                         }
328
329                         e = get_entry(table_base, v);
330                         continue;
331                 }
332
333                 /* Targets which reenter must return
334                  * abs. verdicts
335                  */
336                 acpar.target   = t->u.kernel.target;
337                 acpar.targinfo = t->data;
338                 verdict = t->u.kernel.target->target(skb, &acpar);
339
340                 /* Target might have changed stuff. */
341                 arp = arp_hdr(skb);
342
343                 if (verdict == XT_CONTINUE)
344                         e = arpt_next_entry(e);
345                 else
346                         /* Verdict */
347                         break;
348         } while (!acpar.hotdrop);
349         xt_write_recseq_end(addend);
350         local_bh_enable();
351
352         if (acpar.hotdrop)
353                 return NF_DROP;
354         else
355                 return verdict;
356 }
357
358 /* All zeroes == unconditional rule. */
359 static inline bool unconditional(const struct arpt_arp *arp)
360 {
361         static const struct arpt_arp uncond;
362
363         return memcmp(arp, &uncond, sizeof(uncond)) == 0;
364 }
365
366 /* Figures out from what hook each rule can be called: returns 0 if
367  * there are loops.  Puts hook bitmask in comefrom.
368  */
369 static int mark_source_chains(const struct xt_table_info *newinfo,
370                               unsigned int valid_hooks, void *entry0)
371 {
372         unsigned int hook;
373
374         /* No recursion; use packet counter to save back ptrs (reset
375          * to 0 as we leave), and comefrom to save source hook bitmask.
376          */
377         for (hook = 0; hook < NF_ARP_NUMHOOKS; hook++) {
378                 unsigned int pos = newinfo->hook_entry[hook];
379                 struct arpt_entry *e
380                         = (struct arpt_entry *)(entry0 + pos);
381
382                 if (!(valid_hooks & (1 << hook)))
383                         continue;
384
385                 /* Set initial back pointer. */
386                 e->counters.pcnt = pos;
387
388                 for (;;) {
389                         const struct xt_standard_target *t
390                                 = (void *)arpt_get_target_c(e);
391                         int visited = e->comefrom & (1 << hook);
392
393                         if (e->comefrom & (1 << NF_ARP_NUMHOOKS)) {
394                                 pr_notice("arptables: loop hook %u pos %u %08X.\n",
395                                        hook, pos, e->comefrom);
396                                 return 0;
397                         }
398                         e->comefrom
399                                 |= ((1 << hook) | (1 << NF_ARP_NUMHOOKS));
400
401                         /* Unconditional return/END. */
402                         if ((e->target_offset == sizeof(struct arpt_entry) &&
403                              (strcmp(t->target.u.user.name,
404                                      XT_STANDARD_TARGET) == 0) &&
405                              t->verdict < 0 && unconditional(&e->arp)) ||
406                             visited) {
407                                 unsigned int oldpos, size;
408
409                                 if ((strcmp(t->target.u.user.name,
410                                             XT_STANDARD_TARGET) == 0) &&
411                                     t->verdict < -NF_MAX_VERDICT - 1) {
412                                         duprintf("mark_source_chains: bad "
413                                                 "negative verdict (%i)\n",
414                                                                 t->verdict);
415                                         return 0;
416                                 }
417
418                                 /* Return: backtrack through the last
419                                  * big jump.
420                                  */
421                                 do {
422                                         e->comefrom ^= (1<<NF_ARP_NUMHOOKS);
423                                         oldpos = pos;
424                                         pos = e->counters.pcnt;
425                                         e->counters.pcnt = 0;
426
427                                         /* We're at the start. */
428                                         if (pos == oldpos)
429                                                 goto next;
430
431                                         e = (struct arpt_entry *)
432                                                 (entry0 + pos);
433                                 } while (oldpos == pos + e->next_offset);
434
435                                 /* Move along one */
436                                 size = e->next_offset;
437                                 e = (struct arpt_entry *)
438                                         (entry0 + pos + size);
439                                 e->counters.pcnt = pos;
440                                 pos += size;
441                         } else {
442                                 int newpos = t->verdict;
443
444                                 if (strcmp(t->target.u.user.name,
445                                            XT_STANDARD_TARGET) == 0 &&
446                                     newpos >= 0) {
447                                         if (newpos > newinfo->size -
448                                                 sizeof(struct arpt_entry)) {
449                                                 duprintf("mark_source_chains: "
450                                                         "bad verdict (%i)\n",
451                                                                 newpos);
452                                                 return 0;
453                                         }
454
455                                         /* This a jump; chase it. */
456                                         duprintf("Jump rule %u -> %u\n",
457                                                  pos, newpos);
458                                 } else {
459                                         /* ... this is a fallthru */
460                                         newpos = pos + e->next_offset;
461                                 }
462                                 e = (struct arpt_entry *)
463                                         (entry0 + newpos);
464                                 e->counters.pcnt = pos;
465                                 pos = newpos;
466                         }
467                 }
468                 next:
469                 duprintf("Finished chain %u\n", hook);
470         }
471         return 1;
472 }
473
474 static inline int check_entry(const struct arpt_entry *e, const char *name)
475 {
476         const struct xt_entry_target *t;
477
478         if (!arp_checkentry(&e->arp)) {
479                 duprintf("arp_tables: arp check failed %p %s.\n", e, name);
480                 return -EINVAL;
481         }
482
483         if (e->target_offset + sizeof(struct xt_entry_target) > e->next_offset)
484                 return -EINVAL;
485
486         t = arpt_get_target_c(e);
487         if (e->target_offset + t->u.target_size > e->next_offset)
488                 return -EINVAL;
489
490         return 0;
491 }
492
493 static inline int check_target(struct arpt_entry *e, const char *name)
494 {
495         struct xt_entry_target *t = arpt_get_target(e);
496         int ret;
497         struct xt_tgchk_param par = {
498                 .table     = name,
499                 .entryinfo = e,
500                 .target    = t->u.kernel.target,
501                 .targinfo  = t->data,
502                 .hook_mask = e->comefrom,
503                 .family    = NFPROTO_ARP,
504         };
505
506         ret = xt_check_target(&par, t->u.target_size - sizeof(*t), 0, false);
507         if (ret < 0) {
508                 duprintf("arp_tables: check failed for `%s'.\n",
509                          t->u.kernel.target->name);
510                 return ret;
511         }
512         return 0;
513 }
514
515 static inline int
516 find_check_entry(struct arpt_entry *e, const char *name, unsigned int size)
517 {
518         struct xt_entry_target *t;
519         struct xt_target *target;
520         int ret;
521
522         ret = check_entry(e, name);
523         if (ret)
524                 return ret;
525
526         e->counters.pcnt = xt_percpu_counter_alloc();
527         if (IS_ERR_VALUE(e->counters.pcnt))
528                 return -ENOMEM;
529
530         t = arpt_get_target(e);
531         target = xt_request_find_target(NFPROTO_ARP, t->u.user.name,
532                                         t->u.user.revision);
533         if (IS_ERR(target)) {
534                 duprintf("find_check_entry: `%s' not found\n", t->u.user.name);
535                 ret = PTR_ERR(target);
536                 goto out;
537         }
538         t->u.kernel.target = target;
539
540         ret = check_target(e, name);
541         if (ret)
542                 goto err;
543         return 0;
544 err:
545         module_put(t->u.kernel.target->me);
546 out:
547         xt_percpu_counter_free(e->counters.pcnt);
548
549         return ret;
550 }
551
552 static bool check_underflow(const struct arpt_entry *e)
553 {
554         const struct xt_entry_target *t;
555         unsigned int verdict;
556
557         if (!unconditional(&e->arp))
558                 return false;
559         t = arpt_get_target_c(e);
560         if (strcmp(t->u.user.name, XT_STANDARD_TARGET) != 0)
561                 return false;
562         verdict = ((struct xt_standard_target *)t)->verdict;
563         verdict = -verdict - 1;
564         return verdict == NF_DROP || verdict == NF_ACCEPT;
565 }
566
567 static inline int check_entry_size_and_hooks(struct arpt_entry *e,
568                                              struct xt_table_info *newinfo,
569                                              const unsigned char *base,
570                                              const unsigned char *limit,
571                                              const unsigned int *hook_entries,
572                                              const unsigned int *underflows,
573                                              unsigned int valid_hooks)
574 {
575         unsigned int h;
576
577         if ((unsigned long)e % __alignof__(struct arpt_entry) != 0 ||
578             (unsigned char *)e + sizeof(struct arpt_entry) >= limit) {
579                 duprintf("Bad offset %p\n", e);
580                 return -EINVAL;
581         }
582
583         if (e->next_offset
584             < sizeof(struct arpt_entry) + sizeof(struct xt_entry_target)) {
585                 duprintf("checking: element %p size %u\n",
586                          e, e->next_offset);
587                 return -EINVAL;
588         }
589
590         /* Check hooks & underflows */
591         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
592                 if (!(valid_hooks & (1 << h)))
593                         continue;
594                 if ((unsigned char *)e - base == hook_entries[h])
595                         newinfo->hook_entry[h] = hook_entries[h];
596                 if ((unsigned char *)e - base == underflows[h]) {
597                         if (!check_underflow(e)) {
598                                 pr_err("Underflows must be unconditional and "
599                                        "use the STANDARD target with "
600                                        "ACCEPT/DROP\n");
601                                 return -EINVAL;
602                         }
603                         newinfo->underflow[h] = underflows[h];
604                 }
605         }
606
607         /* Clear counters and comefrom */
608         e->counters = ((struct xt_counters) { 0, 0 });
609         e->comefrom = 0;
610         return 0;
611 }
612
613 static inline void cleanup_entry(struct arpt_entry *e)
614 {
615         struct xt_tgdtor_param par;
616         struct xt_entry_target *t;
617
618         t = arpt_get_target(e);
619         par.target   = t->u.kernel.target;
620         par.targinfo = t->data;
621         par.family   = NFPROTO_ARP;
622         if (par.target->destroy != NULL)
623                 par.target->destroy(&par);
624         module_put(par.target->me);
625         xt_percpu_counter_free(e->counters.pcnt);
626 }
627
628 /* Checks and translates the user-supplied table segment (held in
629  * newinfo).
630  */
631 static int translate_table(struct xt_table_info *newinfo, void *entry0,
632                            const struct arpt_replace *repl)
633 {
634         struct arpt_entry *iter;
635         unsigned int i;
636         int ret = 0;
637
638         newinfo->size = repl->size;
639         newinfo->number = repl->num_entries;
640
641         /* Init all hooks to impossible value. */
642         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
643                 newinfo->hook_entry[i] = 0xFFFFFFFF;
644                 newinfo->underflow[i] = 0xFFFFFFFF;
645         }
646
647         duprintf("translate_table: size %u\n", newinfo->size);
648         i = 0;
649
650         /* Walk through entries, checking offsets. */
651         xt_entry_foreach(iter, entry0, newinfo->size) {
652                 ret = check_entry_size_and_hooks(iter, newinfo, entry0,
653                                                  entry0 + repl->size,
654                                                  repl->hook_entry,
655                                                  repl->underflow,
656                                                  repl->valid_hooks);
657                 if (ret != 0)
658                         break;
659                 ++i;
660                 if (strcmp(arpt_get_target(iter)->u.user.name,
661                     XT_ERROR_TARGET) == 0)
662                         ++newinfo->stacksize;
663         }
664         duprintf("translate_table: ARPT_ENTRY_ITERATE gives %d\n", ret);
665         if (ret != 0)
666                 return ret;
667
668         if (i != repl->num_entries) {
669                 duprintf("translate_table: %u not %u entries\n",
670                          i, repl->num_entries);
671                 return -EINVAL;
672         }
673
674         /* Check hooks all assigned */
675         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
676                 /* Only hooks which are valid */
677                 if (!(repl->valid_hooks & (1 << i)))
678                         continue;
679                 if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
680                         duprintf("Invalid hook entry %u %u\n",
681                                  i, repl->hook_entry[i]);
682                         return -EINVAL;
683                 }
684                 if (newinfo->underflow[i] == 0xFFFFFFFF) {
685                         duprintf("Invalid underflow %u %u\n",
686                                  i, repl->underflow[i]);
687                         return -EINVAL;
688                 }
689         }
690
691         if (!mark_source_chains(newinfo, repl->valid_hooks, entry0)) {
692                 duprintf("Looping hook\n");
693                 return -ELOOP;
694         }
695
696         /* Finally, each sanity check must pass */
697         i = 0;
698         xt_entry_foreach(iter, entry0, newinfo->size) {
699                 ret = find_check_entry(iter, repl->name, repl->size);
700                 if (ret != 0)
701                         break;
702                 ++i;
703         }
704
705         if (ret != 0) {
706                 xt_entry_foreach(iter, entry0, newinfo->size) {
707                         if (i-- == 0)
708                                 break;
709                         cleanup_entry(iter);
710                 }
711                 return ret;
712         }
713
714         /* And one copy for every other CPU */
715         for_each_possible_cpu(i) {
716                 if (newinfo->entries[i] && newinfo->entries[i] != entry0)
717                         memcpy(newinfo->entries[i], entry0, newinfo->size);
718         }
719
720         return ret;
721 }
722
723 static void get_counters(const struct xt_table_info *t,
724                          struct xt_counters counters[])
725 {
726         struct arpt_entry *iter;
727         unsigned int cpu;
728         unsigned int i;
729
730         for_each_possible_cpu(cpu) {
731                 seqcount_t *s = &per_cpu(xt_recseq, cpu);
732
733                 i = 0;
734                 xt_entry_foreach(iter, t->entries[cpu], t->size) {
735                         struct xt_counters *tmp;
736                         u64 bcnt, pcnt;
737                         unsigned int start;
738
739                         tmp = xt_get_per_cpu_counter(&iter->counters, cpu);
740                         do {
741                                 start = read_seqcount_begin(s);
742                                 bcnt = tmp->bcnt;
743                                 pcnt = tmp->pcnt;
744                         } while (read_seqcount_retry(s, start));
745
746                         ADD_COUNTER(counters[i], bcnt, pcnt);
747                         ++i;
748                 }
749         }
750 }
751
752 static struct xt_counters *alloc_counters(const struct xt_table *table)
753 {
754         unsigned int countersize;
755         struct xt_counters *counters;
756         const struct xt_table_info *private = table->private;
757
758         /* We need atomic snapshot of counters: rest doesn't change
759          * (other than comefrom, which userspace doesn't care
760          * about).
761          */
762         countersize = sizeof(struct xt_counters) * private->number;
763         counters = vzalloc(countersize);
764
765         if (counters == NULL)
766                 return ERR_PTR(-ENOMEM);
767
768         get_counters(private, counters);
769
770         return counters;
771 }
772
773 static int copy_entries_to_user(unsigned int total_size,
774                                 const struct xt_table *table,
775                                 void __user *userptr)
776 {
777         unsigned int off, num;
778         const struct arpt_entry *e;
779         struct xt_counters *counters;
780         struct xt_table_info *private = table->private;
781         int ret = 0;
782         void *loc_cpu_entry;
783
784         counters = alloc_counters(table);
785         if (IS_ERR(counters))
786                 return PTR_ERR(counters);
787
788         loc_cpu_entry = private->entries[raw_smp_processor_id()];
789         /* ... then copy entire thing ... */
790         if (copy_to_user(userptr, loc_cpu_entry, total_size) != 0) {
791                 ret = -EFAULT;
792                 goto free_counters;
793         }
794
795         /* FIXME: use iterator macros --RR */
796         /* ... then go back and fix counters and names */
797         for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
798                 const struct xt_entry_target *t;
799
800                 e = (struct arpt_entry *)(loc_cpu_entry + off);
801                 if (copy_to_user(userptr + off
802                                  + offsetof(struct arpt_entry, counters),
803                                  &counters[num],
804                                  sizeof(counters[num])) != 0) {
805                         ret = -EFAULT;
806                         goto free_counters;
807                 }
808
809                 t = arpt_get_target_c(e);
810                 if (copy_to_user(userptr + off + e->target_offset
811                                  + offsetof(struct xt_entry_target,
812                                             u.user.name),
813                                  t->u.kernel.target->name,
814                                  strlen(t->u.kernel.target->name)+1) != 0) {
815                         ret = -EFAULT;
816                         goto free_counters;
817                 }
818         }
819
820  free_counters:
821         vfree(counters);
822         return ret;
823 }
824
825 #ifdef CONFIG_COMPAT
826 static void compat_standard_from_user(void *dst, const void *src)
827 {
828         int v = *(compat_int_t *)src;
829
830         if (v > 0)
831                 v += xt_compat_calc_jump(NFPROTO_ARP, v);
832         memcpy(dst, &v, sizeof(v));
833 }
834
835 static int compat_standard_to_user(void __user *dst, const void *src)
836 {
837         compat_int_t cv = *(int *)src;
838
839         if (cv > 0)
840                 cv -= xt_compat_calc_jump(NFPROTO_ARP, cv);
841         return copy_to_user(dst, &cv, sizeof(cv)) ? -EFAULT : 0;
842 }
843
844 static int compat_calc_entry(const struct arpt_entry *e,
845                              const struct xt_table_info *info,
846                              const void *base, struct xt_table_info *newinfo)
847 {
848         const struct xt_entry_target *t;
849         unsigned int entry_offset;
850         int off, i, ret;
851
852         off = sizeof(struct arpt_entry) - sizeof(struct compat_arpt_entry);
853         entry_offset = (void *)e - base;
854
855         t = arpt_get_target_c(e);
856         off += xt_compat_target_offset(t->u.kernel.target);
857         newinfo->size -= off;
858         ret = xt_compat_add_offset(NFPROTO_ARP, entry_offset, off);
859         if (ret)
860                 return ret;
861
862         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
863                 if (info->hook_entry[i] &&
864                     (e < (struct arpt_entry *)(base + info->hook_entry[i])))
865                         newinfo->hook_entry[i] -= off;
866                 if (info->underflow[i] &&
867                     (e < (struct arpt_entry *)(base + info->underflow[i])))
868                         newinfo->underflow[i] -= off;
869         }
870         return 0;
871 }
872
873 static int compat_table_info(const struct xt_table_info *info,
874                              struct xt_table_info *newinfo)
875 {
876         struct arpt_entry *iter;
877         void *loc_cpu_entry;
878         int ret;
879
880         if (!newinfo || !info)
881                 return -EINVAL;
882
883         /* we dont care about newinfo->entries[] */
884         memcpy(newinfo, info, offsetof(struct xt_table_info, entries));
885         newinfo->initial_entries = 0;
886         loc_cpu_entry = info->entries[raw_smp_processor_id()];
887         xt_compat_init_offsets(NFPROTO_ARP, info->number);
888         xt_entry_foreach(iter, loc_cpu_entry, info->size) {
889                 ret = compat_calc_entry(iter, info, loc_cpu_entry, newinfo);
890                 if (ret != 0)
891                         return ret;
892         }
893         return 0;
894 }
895 #endif
896
897 static int get_info(struct net *net, void __user *user,
898                     const int *len, int compat)
899 {
900         char name[XT_TABLE_MAXNAMELEN];
901         struct xt_table *t;
902         int ret;
903
904         if (*len != sizeof(struct arpt_getinfo)) {
905                 duprintf("length %u != %Zu\n", *len,
906                          sizeof(struct arpt_getinfo));
907                 return -EINVAL;
908         }
909
910         if (copy_from_user(name, user, sizeof(name)) != 0)
911                 return -EFAULT;
912
913         name[XT_TABLE_MAXNAMELEN-1] = '\0';
914 #ifdef CONFIG_COMPAT
915         if (compat)
916                 xt_compat_lock(NFPROTO_ARP);
917 #endif
918         t = try_then_request_module(xt_find_table_lock(net, NFPROTO_ARP, name),
919                                     "arptable_%s", name);
920         if (!IS_ERR_OR_NULL(t)) {
921                 struct arpt_getinfo info;
922                 const struct xt_table_info *private = t->private;
923 #ifdef CONFIG_COMPAT
924                 struct xt_table_info tmp;
925
926                 if (compat) {
927                         ret = compat_table_info(private, &tmp);
928                         xt_compat_flush_offsets(NFPROTO_ARP);
929                         private = &tmp;
930                 }
931 #endif
932                 memset(&info, 0, sizeof(info));
933                 info.valid_hooks = t->valid_hooks;
934                 memcpy(info.hook_entry, private->hook_entry,
935                        sizeof(info.hook_entry));
936                 memcpy(info.underflow, private->underflow,
937                        sizeof(info.underflow));
938                 info.num_entries = private->number;
939                 info.size = private->size;
940                 strcpy(info.name, name);
941
942                 if (copy_to_user(user, &info, *len) != 0)
943                         ret = -EFAULT;
944                 else
945                         ret = 0;
946                 xt_table_unlock(t);
947                 module_put(t->me);
948         } else
949                 ret = t ? PTR_ERR(t) : -ENOENT;
950 #ifdef CONFIG_COMPAT
951         if (compat)
952                 xt_compat_unlock(NFPROTO_ARP);
953 #endif
954         return ret;
955 }
956
957 static int get_entries(struct net *net, struct arpt_get_entries __user *uptr,
958                        const int *len)
959 {
960         int ret;
961         struct arpt_get_entries get;
962         struct xt_table *t;
963
964         if (*len < sizeof(get)) {
965                 duprintf("get_entries: %u < %Zu\n", *len, sizeof(get));
966                 return -EINVAL;
967         }
968         if (copy_from_user(&get, uptr, sizeof(get)) != 0)
969                 return -EFAULT;
970         if (*len != sizeof(struct arpt_get_entries) + get.size) {
971                 duprintf("get_entries: %u != %Zu\n", *len,
972                          sizeof(struct arpt_get_entries) + get.size);
973                 return -EINVAL;
974         }
975
976         t = xt_find_table_lock(net, NFPROTO_ARP, get.name);
977         if (!IS_ERR_OR_NULL(t)) {
978                 const struct xt_table_info *private = t->private;
979
980                 duprintf("t->private->number = %u\n",
981                          private->number);
982                 if (get.size == private->size)
983                         ret = copy_entries_to_user(private->size,
984                                                    t, uptr->entrytable);
985                 else {
986                         duprintf("get_entries: I've got %u not %u!\n",
987                                  private->size, get.size);
988                         ret = -EAGAIN;
989                 }
990                 module_put(t->me);
991                 xt_table_unlock(t);
992         } else
993                 ret = t ? PTR_ERR(t) : -ENOENT;
994
995         return ret;
996 }
997
998 static int __do_replace(struct net *net, const char *name,
999                         unsigned int valid_hooks,
1000                         struct xt_table_info *newinfo,
1001                         unsigned int num_counters,
1002                         void __user *counters_ptr)
1003 {
1004         int ret;
1005         struct xt_table *t;
1006         struct xt_table_info *oldinfo;
1007         struct xt_counters *counters;
1008         void *loc_cpu_old_entry;
1009         struct arpt_entry *iter;
1010
1011         ret = 0;
1012         counters = vzalloc(num_counters * sizeof(struct xt_counters));
1013         if (!counters) {
1014                 ret = -ENOMEM;
1015                 goto out;
1016         }
1017
1018         t = try_then_request_module(xt_find_table_lock(net, NFPROTO_ARP, name),
1019                                     "arptable_%s", name);
1020         if (IS_ERR_OR_NULL(t)) {
1021                 ret = t ? PTR_ERR(t) : -ENOENT;
1022                 goto free_newinfo_counters_untrans;
1023         }
1024
1025         /* You lied! */
1026         if (valid_hooks != t->valid_hooks) {
1027                 duprintf("Valid hook crap: %08X vs %08X\n",
1028                          valid_hooks, t->valid_hooks);
1029                 ret = -EINVAL;
1030                 goto put_module;
1031         }
1032
1033         oldinfo = xt_replace_table(t, num_counters, newinfo, &ret);
1034         if (!oldinfo)
1035                 goto put_module;
1036
1037         /* Update module usage count based on number of rules */
1038         duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
1039                 oldinfo->number, oldinfo->initial_entries, newinfo->number);
1040         if ((oldinfo->number > oldinfo->initial_entries) ||
1041             (newinfo->number <= oldinfo->initial_entries))
1042                 module_put(t->me);
1043         if ((oldinfo->number > oldinfo->initial_entries) &&
1044             (newinfo->number <= oldinfo->initial_entries))
1045                 module_put(t->me);
1046
1047         /* Get the old counters, and synchronize with replace */
1048         get_counters(oldinfo, counters);
1049
1050         /* Decrease module usage counts and free resource */
1051         loc_cpu_old_entry = oldinfo->entries[raw_smp_processor_id()];
1052         xt_entry_foreach(iter, loc_cpu_old_entry, oldinfo->size)
1053                 cleanup_entry(iter);
1054
1055         xt_free_table_info(oldinfo);
1056         if (copy_to_user(counters_ptr, counters,
1057                          sizeof(struct xt_counters) * num_counters) != 0) {
1058                 /* Silent error, can't fail, new table is already in place */
1059                 net_warn_ratelimited("arptables: counters copy to user failed while replacing table\n");
1060         }
1061         vfree(counters);
1062         xt_table_unlock(t);
1063         return ret;
1064
1065  put_module:
1066         module_put(t->me);
1067         xt_table_unlock(t);
1068  free_newinfo_counters_untrans:
1069         vfree(counters);
1070  out:
1071         return ret;
1072 }
1073
1074 static int do_replace(struct net *net, const void __user *user,
1075                       unsigned int len)
1076 {
1077         int ret;
1078         struct arpt_replace tmp;
1079         struct xt_table_info *newinfo;
1080         void *loc_cpu_entry;
1081         struct arpt_entry *iter;
1082
1083         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
1084                 return -EFAULT;
1085
1086         /* overflow check */
1087         if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
1088                 return -ENOMEM;
1089         if (tmp.num_counters == 0)
1090                 return -EINVAL;
1091
1092         tmp.name[sizeof(tmp.name)-1] = 0;
1093
1094         newinfo = xt_alloc_table_info(tmp.size);
1095         if (!newinfo)
1096                 return -ENOMEM;
1097
1098         /* choose the copy that is on our node/cpu */
1099         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1100         if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
1101                            tmp.size) != 0) {
1102                 ret = -EFAULT;
1103                 goto free_newinfo;
1104         }
1105
1106         ret = translate_table(newinfo, loc_cpu_entry, &tmp);
1107         if (ret != 0)
1108                 goto free_newinfo;
1109
1110         duprintf("arp_tables: Translated table\n");
1111
1112         ret = __do_replace(net, tmp.name, tmp.valid_hooks, newinfo,
1113                            tmp.num_counters, tmp.counters);
1114         if (ret)
1115                 goto free_newinfo_untrans;
1116         return 0;
1117
1118  free_newinfo_untrans:
1119         xt_entry_foreach(iter, loc_cpu_entry, newinfo->size)
1120                 cleanup_entry(iter);
1121  free_newinfo:
1122         xt_free_table_info(newinfo);
1123         return ret;
1124 }
1125
1126 static int do_add_counters(struct net *net, const void __user *user,
1127                            unsigned int len, int compat)
1128 {
1129         unsigned int i, curcpu;
1130         struct xt_counters_info tmp;
1131         struct xt_counters *paddc;
1132         unsigned int num_counters;
1133         const char *name;
1134         int size;
1135         void *ptmp;
1136         struct xt_table *t;
1137         const struct xt_table_info *private;
1138         int ret = 0;
1139         void *loc_cpu_entry;
1140         struct arpt_entry *iter;
1141         unsigned int addend;
1142 #ifdef CONFIG_COMPAT
1143         struct compat_xt_counters_info compat_tmp;
1144
1145         if (compat) {
1146                 ptmp = &compat_tmp;
1147                 size = sizeof(struct compat_xt_counters_info);
1148         } else
1149 #endif
1150         {
1151                 ptmp = &tmp;
1152                 size = sizeof(struct xt_counters_info);
1153         }
1154
1155         if (copy_from_user(ptmp, user, size) != 0)
1156                 return -EFAULT;
1157
1158 #ifdef CONFIG_COMPAT
1159         if (compat) {
1160                 num_counters = compat_tmp.num_counters;
1161                 name = compat_tmp.name;
1162         } else
1163 #endif
1164         {
1165                 num_counters = tmp.num_counters;
1166                 name = tmp.name;
1167         }
1168
1169         if (len != size + num_counters * sizeof(struct xt_counters))
1170                 return -EINVAL;
1171
1172         paddc = vmalloc(len - size);
1173         if (!paddc)
1174                 return -ENOMEM;
1175
1176         if (copy_from_user(paddc, user + size, len - size) != 0) {
1177                 ret = -EFAULT;
1178                 goto free;
1179         }
1180
1181         t = xt_find_table_lock(net, NFPROTO_ARP, name);
1182         if (IS_ERR_OR_NULL(t)) {
1183                 ret = t ? PTR_ERR(t) : -ENOENT;
1184                 goto free;
1185         }
1186
1187         local_bh_disable();
1188         private = t->private;
1189         if (private->number != num_counters) {
1190                 ret = -EINVAL;
1191                 goto unlock_up_free;
1192         }
1193
1194         i = 0;
1195         /* Choose the copy that is on our node */
1196         curcpu = smp_processor_id();
1197         loc_cpu_entry = private->entries[curcpu];
1198         addend = xt_write_recseq_begin();
1199         xt_entry_foreach(iter, loc_cpu_entry, private->size) {
1200                 struct xt_counters *tmp;
1201
1202                 tmp = xt_get_this_cpu_counter(&iter->counters);
1203                 ADD_COUNTER(*tmp, paddc[i].bcnt, paddc[i].pcnt);
1204                 ++i;
1205         }
1206         xt_write_recseq_end(addend);
1207  unlock_up_free:
1208         local_bh_enable();
1209         xt_table_unlock(t);
1210         module_put(t->me);
1211  free:
1212         vfree(paddc);
1213
1214         return ret;
1215 }
1216
1217 #ifdef CONFIG_COMPAT
1218 static inline void compat_release_entry(struct compat_arpt_entry *e)
1219 {
1220         struct xt_entry_target *t;
1221
1222         t = compat_arpt_get_target(e);
1223         module_put(t->u.kernel.target->me);
1224 }
1225
1226 static inline int
1227 check_compat_entry_size_and_hooks(struct compat_arpt_entry *e,
1228                                   struct xt_table_info *newinfo,
1229                                   unsigned int *size,
1230                                   const unsigned char *base,
1231                                   const unsigned char *limit,
1232                                   const unsigned int *hook_entries,
1233                                   const unsigned int *underflows,
1234                                   const char *name)
1235 {
1236         struct xt_entry_target *t;
1237         struct xt_target *target;
1238         unsigned int entry_offset;
1239         int ret, off, h;
1240
1241         duprintf("check_compat_entry_size_and_hooks %p\n", e);
1242         if ((unsigned long)e % __alignof__(struct compat_arpt_entry) != 0 ||
1243             (unsigned char *)e + sizeof(struct compat_arpt_entry) >= limit) {
1244                 duprintf("Bad offset %p, limit = %p\n", e, limit);
1245                 return -EINVAL;
1246         }
1247
1248         if (e->next_offset < sizeof(struct compat_arpt_entry) +
1249                              sizeof(struct compat_xt_entry_target)) {
1250                 duprintf("checking: element %p size %u\n",
1251                          e, e->next_offset);
1252                 return -EINVAL;
1253         }
1254
1255         /* For purposes of check_entry casting the compat entry is fine */
1256         ret = check_entry((struct arpt_entry *)e, name);
1257         if (ret)
1258                 return ret;
1259
1260         off = sizeof(struct arpt_entry) - sizeof(struct compat_arpt_entry);
1261         entry_offset = (void *)e - (void *)base;
1262
1263         t = compat_arpt_get_target(e);
1264         target = xt_request_find_target(NFPROTO_ARP, t->u.user.name,
1265                                         t->u.user.revision);
1266         if (IS_ERR(target)) {
1267                 duprintf("check_compat_entry_size_and_hooks: `%s' not found\n",
1268                          t->u.user.name);
1269                 ret = PTR_ERR(target);
1270                 goto out;
1271         }
1272         t->u.kernel.target = target;
1273
1274         off += xt_compat_target_offset(target);
1275         *size += off;
1276         ret = xt_compat_add_offset(NFPROTO_ARP, entry_offset, off);
1277         if (ret)
1278                 goto release_target;
1279
1280         /* Check hooks & underflows */
1281         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
1282                 if ((unsigned char *)e - base == hook_entries[h])
1283                         newinfo->hook_entry[h] = hook_entries[h];
1284                 if ((unsigned char *)e - base == underflows[h])
1285                         newinfo->underflow[h] = underflows[h];
1286         }
1287
1288         /* Clear counters and comefrom */
1289         memset(&e->counters, 0, sizeof(e->counters));
1290         e->comefrom = 0;
1291         return 0;
1292
1293 release_target:
1294         module_put(t->u.kernel.target->me);
1295 out:
1296         return ret;
1297 }
1298
1299 static int
1300 compat_copy_entry_from_user(struct compat_arpt_entry *e, void **dstptr,
1301                             unsigned int *size, const char *name,
1302                             struct xt_table_info *newinfo, unsigned char *base)
1303 {
1304         struct xt_entry_target *t;
1305         struct xt_target *target;
1306         struct arpt_entry *de;
1307         unsigned int origsize;
1308         int ret, h;
1309
1310         ret = 0;
1311         origsize = *size;
1312         de = (struct arpt_entry *)*dstptr;
1313         memcpy(de, e, sizeof(struct arpt_entry));
1314         memcpy(&de->counters, &e->counters, sizeof(e->counters));
1315
1316         *dstptr += sizeof(struct arpt_entry);
1317         *size += sizeof(struct arpt_entry) - sizeof(struct compat_arpt_entry);
1318
1319         de->target_offset = e->target_offset - (origsize - *size);
1320         t = compat_arpt_get_target(e);
1321         target = t->u.kernel.target;
1322         xt_compat_target_from_user(t, dstptr, size);
1323
1324         de->next_offset = e->next_offset - (origsize - *size);
1325         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
1326                 if ((unsigned char *)de - base < newinfo->hook_entry[h])
1327                         newinfo->hook_entry[h] -= origsize - *size;
1328                 if ((unsigned char *)de - base < newinfo->underflow[h])
1329                         newinfo->underflow[h] -= origsize - *size;
1330         }
1331         return ret;
1332 }
1333
1334 static int translate_compat_table(const char *name,
1335                                   unsigned int valid_hooks,
1336                                   struct xt_table_info **pinfo,
1337                                   void **pentry0,
1338                                   unsigned int total_size,
1339                                   unsigned int number,
1340                                   unsigned int *hook_entries,
1341                                   unsigned int *underflows)
1342 {
1343         unsigned int i, j;
1344         struct xt_table_info *newinfo, *info;
1345         void *pos, *entry0, *entry1;
1346         struct compat_arpt_entry *iter0;
1347         struct arpt_entry *iter1;
1348         unsigned int size;
1349         int ret = 0;
1350
1351         info = *pinfo;
1352         entry0 = *pentry0;
1353         size = total_size;
1354         info->number = number;
1355
1356         /* Init all hooks to impossible value. */
1357         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
1358                 info->hook_entry[i] = 0xFFFFFFFF;
1359                 info->underflow[i] = 0xFFFFFFFF;
1360         }
1361
1362         duprintf("translate_compat_table: size %u\n", info->size);
1363         j = 0;
1364         xt_compat_lock(NFPROTO_ARP);
1365         xt_compat_init_offsets(NFPROTO_ARP, number);
1366         /* Walk through entries, checking offsets. */
1367         xt_entry_foreach(iter0, entry0, total_size) {
1368                 ret = check_compat_entry_size_and_hooks(iter0, info, &size,
1369                                                         entry0,
1370                                                         entry0 + total_size,
1371                                                         hook_entries,
1372                                                         underflows,
1373                                                         name);
1374                 if (ret != 0)
1375                         goto out_unlock;
1376                 ++j;
1377         }
1378
1379         ret = -EINVAL;
1380         if (j != number) {
1381                 duprintf("translate_compat_table: %u not %u entries\n",
1382                          j, number);
1383                 goto out_unlock;
1384         }
1385
1386         /* Check hooks all assigned */
1387         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
1388                 /* Only hooks which are valid */
1389                 if (!(valid_hooks & (1 << i)))
1390                         continue;
1391                 if (info->hook_entry[i] == 0xFFFFFFFF) {
1392                         duprintf("Invalid hook entry %u %u\n",
1393                                  i, hook_entries[i]);
1394                         goto out_unlock;
1395                 }
1396                 if (info->underflow[i] == 0xFFFFFFFF) {
1397                         duprintf("Invalid underflow %u %u\n",
1398                                  i, underflows[i]);
1399                         goto out_unlock;
1400                 }
1401         }
1402
1403         ret = -ENOMEM;
1404         newinfo = xt_alloc_table_info(size);
1405         if (!newinfo)
1406                 goto out_unlock;
1407
1408         newinfo->number = number;
1409         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
1410                 newinfo->hook_entry[i] = info->hook_entry[i];
1411                 newinfo->underflow[i] = info->underflow[i];
1412         }
1413         entry1 = newinfo->entries[raw_smp_processor_id()];
1414         pos = entry1;
1415         size = total_size;
1416         xt_entry_foreach(iter0, entry0, total_size) {
1417                 ret = compat_copy_entry_from_user(iter0, &pos, &size,
1418                                                   name, newinfo, entry1);
1419                 if (ret != 0)
1420                         break;
1421         }
1422         xt_compat_flush_offsets(NFPROTO_ARP);
1423         xt_compat_unlock(NFPROTO_ARP);
1424         if (ret)
1425                 goto free_newinfo;
1426
1427         ret = -ELOOP;
1428         if (!mark_source_chains(newinfo, valid_hooks, entry1))
1429                 goto free_newinfo;
1430
1431         i = 0;
1432         xt_entry_foreach(iter1, entry1, newinfo->size) {
1433                 iter1->counters.pcnt = xt_percpu_counter_alloc();
1434                 if (IS_ERR_VALUE(iter1->counters.pcnt)) {
1435                         ret = -ENOMEM;
1436                         break;
1437                 }
1438
1439                 ret = check_target(iter1, name);
1440                 if (ret != 0) {
1441                         xt_percpu_counter_free(iter1->counters.pcnt);
1442                         break;
1443                 }
1444                 ++i;
1445                 if (strcmp(arpt_get_target(iter1)->u.user.name,
1446                     XT_ERROR_TARGET) == 0)
1447                         ++newinfo->stacksize;
1448         }
1449         if (ret) {
1450                 /*
1451                  * The first i matches need cleanup_entry (calls ->destroy)
1452                  * because they had called ->check already. The other j-i
1453                  * entries need only release.
1454                  */
1455                 int skip = i;
1456                 j -= i;
1457                 xt_entry_foreach(iter0, entry0, newinfo->size) {
1458                         if (skip-- > 0)
1459                                 continue;
1460                         if (j-- == 0)
1461                                 break;
1462                         compat_release_entry(iter0);
1463                 }
1464                 xt_entry_foreach(iter1, entry1, newinfo->size) {
1465                         if (i-- == 0)
1466                                 break;
1467                         cleanup_entry(iter1);
1468                 }
1469                 xt_free_table_info(newinfo);
1470                 return ret;
1471         }
1472
1473         /* And one copy for every other CPU */
1474         for_each_possible_cpu(i)
1475                 if (newinfo->entries[i] && newinfo->entries[i] != entry1)
1476                         memcpy(newinfo->entries[i], entry1, newinfo->size);
1477
1478         *pinfo = newinfo;
1479         *pentry0 = entry1;
1480         xt_free_table_info(info);
1481         return 0;
1482
1483 free_newinfo:
1484         xt_free_table_info(newinfo);
1485 out:
1486         xt_entry_foreach(iter0, entry0, total_size) {
1487                 if (j-- == 0)
1488                         break;
1489                 compat_release_entry(iter0);
1490         }
1491         return ret;
1492 out_unlock:
1493         xt_compat_flush_offsets(NFPROTO_ARP);
1494         xt_compat_unlock(NFPROTO_ARP);
1495         goto out;
1496 }
1497
1498 struct compat_arpt_replace {
1499         char                            name[XT_TABLE_MAXNAMELEN];
1500         u32                             valid_hooks;
1501         u32                             num_entries;
1502         u32                             size;
1503         u32                             hook_entry[NF_ARP_NUMHOOKS];
1504         u32                             underflow[NF_ARP_NUMHOOKS];
1505         u32                             num_counters;
1506         compat_uptr_t                   counters;
1507         struct compat_arpt_entry        entries[0];
1508 };
1509
1510 static int compat_do_replace(struct net *net, void __user *user,
1511                              unsigned int len)
1512 {
1513         int ret;
1514         struct compat_arpt_replace tmp;
1515         struct xt_table_info *newinfo;
1516         void *loc_cpu_entry;
1517         struct arpt_entry *iter;
1518
1519         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
1520                 return -EFAULT;
1521
1522         /* overflow check */
1523         if (tmp.size >= INT_MAX / num_possible_cpus())
1524                 return -ENOMEM;
1525         if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
1526                 return -ENOMEM;
1527         if (tmp.num_counters == 0)
1528                 return -EINVAL;
1529
1530         tmp.name[sizeof(tmp.name)-1] = 0;
1531
1532         newinfo = xt_alloc_table_info(tmp.size);
1533         if (!newinfo)
1534                 return -ENOMEM;
1535
1536         /* choose the copy that is on our node/cpu */
1537         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1538         if (copy_from_user(loc_cpu_entry, user + sizeof(tmp), tmp.size) != 0) {
1539                 ret = -EFAULT;
1540                 goto free_newinfo;
1541         }
1542
1543         ret = translate_compat_table(tmp.name, tmp.valid_hooks,
1544                                      &newinfo, &loc_cpu_entry, tmp.size,
1545                                      tmp.num_entries, tmp.hook_entry,
1546                                      tmp.underflow);
1547         if (ret != 0)
1548                 goto free_newinfo;
1549
1550         duprintf("compat_do_replace: Translated table\n");
1551
1552         ret = __do_replace(net, tmp.name, tmp.valid_hooks, newinfo,
1553                            tmp.num_counters, compat_ptr(tmp.counters));
1554         if (ret)
1555                 goto free_newinfo_untrans;
1556         return 0;
1557
1558  free_newinfo_untrans:
1559         xt_entry_foreach(iter, loc_cpu_entry, newinfo->size)
1560                 cleanup_entry(iter);
1561  free_newinfo:
1562         xt_free_table_info(newinfo);
1563         return ret;
1564 }
1565
1566 static int compat_do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user,
1567                                   unsigned int len)
1568 {
1569         int ret;
1570
1571         if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1572                 return -EPERM;
1573
1574         switch (cmd) {
1575         case ARPT_SO_SET_REPLACE:
1576                 ret = compat_do_replace(sock_net(sk), user, len);
1577                 break;
1578
1579         case ARPT_SO_SET_ADD_COUNTERS:
1580                 ret = do_add_counters(sock_net(sk), user, len, 1);
1581                 break;
1582
1583         default:
1584                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
1585                 ret = -EINVAL;
1586         }
1587
1588         return ret;
1589 }
1590
1591 static int compat_copy_entry_to_user(struct arpt_entry *e, void __user **dstptr,
1592                                      compat_uint_t *size,
1593                                      struct xt_counters *counters,
1594                                      unsigned int i)
1595 {
1596         struct xt_entry_target *t;
1597         struct compat_arpt_entry __user *ce;
1598         u_int16_t target_offset, next_offset;
1599         compat_uint_t origsize;
1600         int ret;
1601
1602         origsize = *size;
1603         ce = (struct compat_arpt_entry __user *)*dstptr;
1604         if (copy_to_user(ce, e, sizeof(struct arpt_entry)) != 0 ||
1605             copy_to_user(&ce->counters, &counters[i],
1606             sizeof(counters[i])) != 0)
1607                 return -EFAULT;
1608
1609         *dstptr += sizeof(struct compat_arpt_entry);
1610         *size -= sizeof(struct arpt_entry) - sizeof(struct compat_arpt_entry);
1611
1612         target_offset = e->target_offset - (origsize - *size);
1613
1614         t = arpt_get_target(e);
1615         ret = xt_compat_target_to_user(t, dstptr, size);
1616         if (ret)
1617                 return ret;
1618         next_offset = e->next_offset - (origsize - *size);
1619         if (put_user(target_offset, &ce->target_offset) != 0 ||
1620             put_user(next_offset, &ce->next_offset) != 0)
1621                 return -EFAULT;
1622         return 0;
1623 }
1624
1625 static int compat_copy_entries_to_user(unsigned int total_size,
1626                                        struct xt_table *table,
1627                                        void __user *userptr)
1628 {
1629         struct xt_counters *counters;
1630         const struct xt_table_info *private = table->private;
1631         void __user *pos;
1632         unsigned int size;
1633         int ret = 0;
1634         void *loc_cpu_entry;
1635         unsigned int i = 0;
1636         struct arpt_entry *iter;
1637
1638         counters = alloc_counters(table);
1639         if (IS_ERR(counters))
1640                 return PTR_ERR(counters);
1641
1642         /* choose the copy on our node/cpu */
1643         loc_cpu_entry = private->entries[raw_smp_processor_id()];
1644         pos = userptr;
1645         size = total_size;
1646         xt_entry_foreach(iter, loc_cpu_entry, total_size) {
1647                 ret = compat_copy_entry_to_user(iter, &pos,
1648                                                 &size, counters, i++);
1649                 if (ret != 0)
1650                         break;
1651         }
1652         vfree(counters);
1653         return ret;
1654 }
1655
1656 struct compat_arpt_get_entries {
1657         char name[XT_TABLE_MAXNAMELEN];
1658         compat_uint_t size;
1659         struct compat_arpt_entry entrytable[0];
1660 };
1661
1662 static int compat_get_entries(struct net *net,
1663                               struct compat_arpt_get_entries __user *uptr,
1664                               int *len)
1665 {
1666         int ret;
1667         struct compat_arpt_get_entries get;
1668         struct xt_table *t;
1669
1670         if (*len < sizeof(get)) {
1671                 duprintf("compat_get_entries: %u < %zu\n", *len, sizeof(get));
1672                 return -EINVAL;
1673         }
1674         if (copy_from_user(&get, uptr, sizeof(get)) != 0)
1675                 return -EFAULT;
1676         if (*len != sizeof(struct compat_arpt_get_entries) + get.size) {
1677                 duprintf("compat_get_entries: %u != %zu\n",
1678                          *len, sizeof(get) + get.size);
1679                 return -EINVAL;
1680         }
1681
1682         xt_compat_lock(NFPROTO_ARP);
1683         t = xt_find_table_lock(net, NFPROTO_ARP, get.name);
1684         if (!IS_ERR_OR_NULL(t)) {
1685                 const struct xt_table_info *private = t->private;
1686                 struct xt_table_info info;
1687
1688                 duprintf("t->private->number = %u\n", private->number);
1689                 ret = compat_table_info(private, &info);
1690                 if (!ret && get.size == info.size) {
1691                         ret = compat_copy_entries_to_user(private->size,
1692                                                           t, uptr->entrytable);
1693                 } else if (!ret) {
1694                         duprintf("compat_get_entries: I've got %u not %u!\n",
1695                                  private->size, get.size);
1696                         ret = -EAGAIN;
1697                 }
1698                 xt_compat_flush_offsets(NFPROTO_ARP);
1699                 module_put(t->me);
1700                 xt_table_unlock(t);
1701         } else
1702                 ret = t ? PTR_ERR(t) : -ENOENT;
1703
1704         xt_compat_unlock(NFPROTO_ARP);
1705         return ret;
1706 }
1707
1708 static int do_arpt_get_ctl(struct sock *, int, void __user *, int *);
1709
1710 static int compat_do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user,
1711                                   int *len)
1712 {
1713         int ret;
1714
1715         if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1716                 return -EPERM;
1717
1718         switch (cmd) {
1719         case ARPT_SO_GET_INFO:
1720                 ret = get_info(sock_net(sk), user, len, 1);
1721                 break;
1722         case ARPT_SO_GET_ENTRIES:
1723                 ret = compat_get_entries(sock_net(sk), user, len);
1724                 break;
1725         default:
1726                 ret = do_arpt_get_ctl(sk, cmd, user, len);
1727         }
1728         return ret;
1729 }
1730 #endif
1731
1732 static int do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
1733 {
1734         int ret;
1735
1736         if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1737                 return -EPERM;
1738
1739         switch (cmd) {
1740         case ARPT_SO_SET_REPLACE:
1741                 ret = do_replace(sock_net(sk), user, len);
1742                 break;
1743
1744         case ARPT_SO_SET_ADD_COUNTERS:
1745                 ret = do_add_counters(sock_net(sk), user, len, 0);
1746                 break;
1747
1748         default:
1749                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
1750                 ret = -EINVAL;
1751         }
1752
1753         return ret;
1754 }
1755
1756 static int do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
1757 {
1758         int ret;
1759
1760         if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1761                 return -EPERM;
1762
1763         switch (cmd) {
1764         case ARPT_SO_GET_INFO:
1765                 ret = get_info(sock_net(sk), user, len, 0);
1766                 break;
1767
1768         case ARPT_SO_GET_ENTRIES:
1769                 ret = get_entries(sock_net(sk), user, len);
1770                 break;
1771
1772         case ARPT_SO_GET_REVISION_TARGET: {
1773                 struct xt_get_revision rev;
1774
1775                 if (*len != sizeof(rev)) {
1776                         ret = -EINVAL;
1777                         break;
1778                 }
1779                 if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
1780                         ret = -EFAULT;
1781                         break;
1782                 }
1783                 rev.name[sizeof(rev.name)-1] = 0;
1784
1785                 try_then_request_module(xt_find_revision(NFPROTO_ARP, rev.name,
1786                                                          rev.revision, 1, &ret),
1787                                         "arpt_%s", rev.name);
1788                 break;
1789         }
1790
1791         default:
1792                 duprintf("do_arpt_get_ctl: unknown request %i\n", cmd);
1793                 ret = -EINVAL;
1794         }
1795
1796         return ret;
1797 }
1798
1799 struct xt_table *arpt_register_table(struct net *net,
1800                                      const struct xt_table *table,
1801                                      const struct arpt_replace *repl)
1802 {
1803         int ret;
1804         struct xt_table_info *newinfo;
1805         struct xt_table_info bootstrap = {0};
1806         void *loc_cpu_entry;
1807         struct xt_table *new_table;
1808
1809         newinfo = xt_alloc_table_info(repl->size);
1810         if (!newinfo) {
1811                 ret = -ENOMEM;
1812                 goto out;
1813         }
1814
1815         /* choose the copy on our node/cpu */
1816         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1817         memcpy(loc_cpu_entry, repl->entries, repl->size);
1818
1819         ret = translate_table(newinfo, loc_cpu_entry, repl);
1820         duprintf("arpt_register_table: translate table gives %d\n", ret);
1821         if (ret != 0)
1822                 goto out_free;
1823
1824         new_table = xt_register_table(net, table, &bootstrap, newinfo);
1825         if (IS_ERR(new_table)) {
1826                 ret = PTR_ERR(new_table);
1827                 goto out_free;
1828         }
1829         return new_table;
1830
1831 out_free:
1832         xt_free_table_info(newinfo);
1833 out:
1834         return ERR_PTR(ret);
1835 }
1836
1837 void arpt_unregister_table(struct xt_table *table)
1838 {
1839         struct xt_table_info *private;
1840         void *loc_cpu_entry;
1841         struct module *table_owner = table->me;
1842         struct arpt_entry *iter;
1843
1844         private = xt_unregister_table(table);
1845
1846         /* Decrease module usage counts and free resources */
1847         loc_cpu_entry = private->entries[raw_smp_processor_id()];
1848         xt_entry_foreach(iter, loc_cpu_entry, private->size)
1849                 cleanup_entry(iter);
1850         if (private->number > private->initial_entries)
1851                 module_put(table_owner);
1852         xt_free_table_info(private);
1853 }
1854
1855 /* The built-in targets: standard (NULL) and error. */
1856 static struct xt_target arpt_builtin_tg[] __read_mostly = {
1857         {
1858                 .name             = XT_STANDARD_TARGET,
1859                 .targetsize       = sizeof(int),
1860                 .family           = NFPROTO_ARP,
1861 #ifdef CONFIG_COMPAT
1862                 .compatsize       = sizeof(compat_int_t),
1863                 .compat_from_user = compat_standard_from_user,
1864                 .compat_to_user   = compat_standard_to_user,
1865 #endif
1866         },
1867         {
1868                 .name             = XT_ERROR_TARGET,
1869                 .target           = arpt_error,
1870                 .targetsize       = XT_FUNCTION_MAXNAMELEN,
1871                 .family           = NFPROTO_ARP,
1872         },
1873 };
1874
1875 static struct nf_sockopt_ops arpt_sockopts = {
1876         .pf             = PF_INET,
1877         .set_optmin     = ARPT_BASE_CTL,
1878         .set_optmax     = ARPT_SO_SET_MAX+1,
1879         .set            = do_arpt_set_ctl,
1880 #ifdef CONFIG_COMPAT
1881         .compat_set     = compat_do_arpt_set_ctl,
1882 #endif
1883         .get_optmin     = ARPT_BASE_CTL,
1884         .get_optmax     = ARPT_SO_GET_MAX+1,
1885         .get            = do_arpt_get_ctl,
1886 #ifdef CONFIG_COMPAT
1887         .compat_get     = compat_do_arpt_get_ctl,
1888 #endif
1889         .owner          = THIS_MODULE,
1890 };
1891
1892 static int __net_init arp_tables_net_init(struct net *net)
1893 {
1894         return xt_proto_init(net, NFPROTO_ARP);
1895 }
1896
1897 static void __net_exit arp_tables_net_exit(struct net *net)
1898 {
1899         xt_proto_fini(net, NFPROTO_ARP);
1900 }
1901
1902 static struct pernet_operations arp_tables_net_ops = {
1903         .init = arp_tables_net_init,
1904         .exit = arp_tables_net_exit,
1905 };
1906
1907 static int __init arp_tables_init(void)
1908 {
1909         int ret;
1910
1911         ret = register_pernet_subsys(&arp_tables_net_ops);
1912         if (ret < 0)
1913                 goto err1;
1914
1915         /* No one else will be downing sem now, so we won't sleep */
1916         ret = xt_register_targets(arpt_builtin_tg, ARRAY_SIZE(arpt_builtin_tg));
1917         if (ret < 0)
1918                 goto err2;
1919
1920         /* Register setsockopt */
1921         ret = nf_register_sockopt(&arpt_sockopts);
1922         if (ret < 0)
1923                 goto err4;
1924
1925         printk(KERN_INFO "arp_tables: (C) 2002 David S. Miller\n");
1926         return 0;
1927
1928 err4:
1929         xt_unregister_targets(arpt_builtin_tg, ARRAY_SIZE(arpt_builtin_tg));
1930 err2:
1931         unregister_pernet_subsys(&arp_tables_net_ops);
1932 err1:
1933         return ret;
1934 }
1935
1936 static void __exit arp_tables_fini(void)
1937 {
1938         nf_unregister_sockopt(&arpt_sockopts);
1939         xt_unregister_targets(arpt_builtin_tg, ARRAY_SIZE(arpt_builtin_tg));
1940         unregister_pernet_subsys(&arp_tables_net_ops);
1941 }
1942
1943 EXPORT_SYMBOL(arpt_register_table);
1944 EXPORT_SYMBOL(arpt_unregister_table);
1945 EXPORT_SYMBOL(arpt_do_table);
1946
1947 module_init(arp_tables_init);
1948 module_exit(arp_tables_fini);