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/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
  24 #include <asm/uaccess.h>
  25 #include <asm/semaphore.h>
  26
  27 #include <linux/netfilter/x_tables.h>
  28 #include <linux/netfilter_arp/arp_tables.h>
  29
  30 MODULE_LICENSE("GPL");
  31 MODULE_AUTHOR("David S. Miller <davem@redhat.com>");
  32 MODULE_DESCRIPTION("arptables core");
  33
  34 /*#define DEBUG_ARP_TABLES*/
  35 /*#define DEBUG_ARP_TABLES_USER*/
  36
  37 #ifdef DEBUG_ARP_TABLES
  38 #define dprintf(format, args...)  printk(format , ## args)
  39 #else
  40 #define dprintf(format, args...)
  41 #endif
  42
  43 #ifdef DEBUG_ARP_TABLES_USER
  44 #define duprintf(format, args...) printk(format , ## args)
  45 #else
  46 #define duprintf(format, args...)
  47 #endif
  48
  49 #ifdef CONFIG_NETFILTER_DEBUG
  50 #define ARP_NF_ASSERT(x)                                        \
  51 do {                                                            \
  52         if (!(x))                                               \
  53                 printk("ARP_NF_ASSERT: %s:%s:%u\n",             \
  54                        __FUNCTION__, __FILE__, __LINE__);       \
  55 } while(0)
  56 #else
  57 #define ARP_NF_ASSERT(x)
  58 #endif
  59
  60 #include <linux/netfilter_ipv4/listhelp.h>
  61
  62 static inline int arp_devaddr_compare(const struct arpt_devaddr_info *ap,
  63                                       char *hdr_addr, int len)
  64 {
  65         int i, ret;
  66
  67         if (len > ARPT_DEV_ADDR_LEN_MAX)
  68                 len = ARPT_DEV_ADDR_LEN_MAX;
  69
  70         ret = 0;
  71         for (i = 0; i < len; i++)
  72                 ret |= (hdr_addr[i] ^ ap->addr[i]) & ap->mask[i];
  73
  74         return (ret != 0);
  75 }
  76
  77 /* Returns whether packet matches rule or not. */
  78 static inline int arp_packet_match(const struct arphdr *arphdr,
  79                                    struct net_device *dev,
  80                                    const char *indev,
  81                                    const char *outdev,
  82                                    const struct arpt_arp *arpinfo)
  83 {
  84         char *arpptr = (char *)(arphdr + 1);
  85         char *src_devaddr, *tgt_devaddr;
  86         u32 src_ipaddr, tgt_ipaddr;
  87         int i, ret;
  88
  89 #define FWINV(bool,invflg) ((bool) ^ !!(arpinfo->invflags & invflg))
  90
  91         if (FWINV((arphdr->ar_op & arpinfo->arpop_mask) != arpinfo->arpop,
  92                   ARPT_INV_ARPOP)) {
  93                 dprintf("ARP operation field mismatch.\n");
  94                 dprintf("ar_op: %04x info->arpop: %04x info->arpop_mask: %04x\n",
  95                         arphdr->ar_op, arpinfo->arpop, arpinfo->arpop_mask);
  96                 return 0;
  97         }
  98
  99         if (FWINV((arphdr->ar_hrd & arpinfo->arhrd_mask) != arpinfo->arhrd,
 100                   ARPT_INV_ARPHRD)) {
 101                 dprintf("ARP hardware address format mismatch.\n");
 102                 dprintf("ar_hrd: %04x info->arhrd: %04x info->arhrd_mask: %04x\n",
 103                         arphdr->ar_hrd, arpinfo->arhrd, arpinfo->arhrd_mask);
 104                 return 0;
 105         }
 106
 107         if (FWINV((arphdr->ar_pro & arpinfo->arpro_mask) != arpinfo->arpro,
 108                   ARPT_INV_ARPPRO)) {
 109                 dprintf("ARP protocol address format mismatch.\n");
 110                 dprintf("ar_pro: %04x info->arpro: %04x info->arpro_mask: %04x\n",
 111                         arphdr->ar_pro, arpinfo->arpro, arpinfo->arpro_mask);
 112                 return 0;
 113         }
 114
 115         if (FWINV((arphdr->ar_hln & arpinfo->arhln_mask) != arpinfo->arhln,
 116                   ARPT_INV_ARPHLN)) {
 117                 dprintf("ARP hardware address length mismatch.\n");
 118                 dprintf("ar_hln: %02x info->arhln: %02x info->arhln_mask: %02x\n",
 119                         arphdr->ar_hln, arpinfo->arhln, arpinfo->arhln_mask);
 120                 return 0;
 121         }
 122
 123         src_devaddr = arpptr;
 124         arpptr += dev->addr_len;
 125         memcpy(&src_ipaddr, arpptr, sizeof(u32));
 126         arpptr += sizeof(u32);
 127         tgt_devaddr = arpptr;
 128         arpptr += dev->addr_len;
 129         memcpy(&tgt_ipaddr, arpptr, sizeof(u32));
 130
 131         if (FWINV(arp_devaddr_compare(&arpinfo->src_devaddr, src_devaddr, dev->addr_len),
 132                   ARPT_INV_SRCDEVADDR) ||
 133             FWINV(arp_devaddr_compare(&arpinfo->tgt_devaddr, tgt_devaddr, dev->addr_len),
 134                   ARPT_INV_TGTDEVADDR)) {
 135                 dprintf("Source or target device address mismatch.\n");
 136
 137                 return 0;
 138         }
 139
 140         if (FWINV((src_ipaddr & arpinfo->smsk.s_addr) != arpinfo->src.s_addr,
 141                   ARPT_INV_SRCIP) ||
 142             FWINV(((tgt_ipaddr & arpinfo->tmsk.s_addr) != arpinfo->tgt.s_addr),
 143                   ARPT_INV_TGTIP)) {
 144                 dprintf("Source or target IP address mismatch.\n");
 145
 146                 dprintf("SRC: %u.%u.%u.%u. Mask: %u.%u.%u.%u. Target: %u.%u.%u.%u.%s\n",
 147                         NIPQUAD(src_ipaddr),
 148                         NIPQUAD(arpinfo->smsk.s_addr),
 149                         NIPQUAD(arpinfo->src.s_addr),
 150                         arpinfo->invflags & ARPT_INV_SRCIP ? " (INV)" : "");
 151                 dprintf("TGT: %u.%u.%u.%u Mask: %u.%u.%u.%u Target: %u.%u.%u.%u.%s\n",
 152                         NIPQUAD(tgt_ipaddr),
 153                         NIPQUAD(arpinfo->tmsk.s_addr),
 154                         NIPQUAD(arpinfo->tgt.s_addr),
 155                         arpinfo->invflags & ARPT_INV_TGTIP ? " (INV)" : "");
 156                 return 0;
 157         }
 158
 159         /* Look for ifname matches.  */
 160         for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
 161                 ret |= (indev[i] ^ arpinfo->iniface[i])
 162                         & arpinfo->iniface_mask[i];
 163         }
 164
 165         if (FWINV(ret != 0, ARPT_INV_VIA_IN)) {
 166                 dprintf("VIA in mismatch (%s vs %s).%s\n",
 167                         indev, arpinfo->iniface,
 168                         arpinfo->invflags&ARPT_INV_VIA_IN ?" (INV)":"");
 169                 return 0;
 170         }
 171
 172         for (i = 0, ret = 0; i < IFNAMSIZ/sizeof(unsigned long); i++) {
 173                 unsigned long odev;
 174                 memcpy(&odev, outdev + i*sizeof(unsigned long),
 175                        sizeof(unsigned long));
 176                 ret |= (odev
 177                         ^ ((const unsigned long *)arpinfo->outiface)[i])
 178                         & ((const unsigned long *)arpinfo->outiface_mask)[i];
 179         }
 180
 181         if (FWINV(ret != 0, ARPT_INV_VIA_OUT)) {
 182                 dprintf("VIA out mismatch (%s vs %s).%s\n",
 183                         outdev, arpinfo->outiface,
 184                         arpinfo->invflags&ARPT_INV_VIA_OUT ?" (INV)":"");
 185                 return 0;
 186         }
 187
 188         return 1;
 189 }
 190
 191 static inline int arp_checkentry(const struct arpt_arp *arp)
 192 {
 193         if (arp->flags & ~ARPT_F_MASK) {
 194                 duprintf("Unknown flag bits set: %08X\n",
 195                          arp->flags & ~ARPT_F_MASK);
 196                 return 0;
 197         }
 198         if (arp->invflags & ~ARPT_INV_MASK) {
 199                 duprintf("Unknown invflag bits set: %08X\n",
 200                          arp->invflags & ~ARPT_INV_MASK);
 201                 return 0;
 202         }
 203
 204         return 1;
 205 }
 206
 207 static unsigned int arpt_error(struct sk_buff **pskb,
 208                                const struct net_device *in,
 209                                const struct net_device *out,
 210                                unsigned int hooknum,
 211                                const void *targinfo,
 212                                void *userinfo)
 213 {
 214         if (net_ratelimit())
 215                 printk("arp_tables: error: '%s'\n", (char *)targinfo);
 216
 217         return NF_DROP;
 218 }
 219
 220 static inline struct arpt_entry *get_entry(void *base, unsigned int offset)
 221 {
 222         return (struct arpt_entry *)(base + offset);
 223 }
 224
 225 unsigned int arpt_do_table(struct sk_buff **pskb,
 226                            unsigned int hook,
 227                            const struct net_device *in,
 228                            const struct net_device *out,
 229                            struct arpt_table *table,
 230                            void *userdata)
 231 {
 232         static const char nulldevname[IFNAMSIZ];
 233         unsigned int verdict = NF_DROP;
 234         struct arphdr *arp;
 235         int hotdrop = 0;
 236         struct arpt_entry *e, *back;
 237         const char *indev, *outdev;
 238         void *table_base;
 239         struct xt_table_info *private;
 240
 241         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
 242         if (!pskb_may_pull((*pskb), (sizeof(struct arphdr) +
 243                                      (2 * (*pskb)->dev->addr_len) +
 244                                      (2 * sizeof(u32)))))
 245                 return NF_DROP;
 246
 247         indev = in ? in->name : nulldevname;
 248         outdev = out ? out->name : nulldevname;
 249
 250         read_lock_bh(&table->lock);
 251         private = table->private;
 252         table_base = (void *)private->entries[smp_processor_id()];
 253         e = get_entry(table_base, private->hook_entry[hook]);
 254         back = get_entry(table_base, private->underflow[hook]);
 255
 256         arp = (*pskb)->nh.arph;
 257         do {
 258                 if (arp_packet_match(arp, (*pskb)->dev, indev, outdev, &e->arp)) {
 259                         struct arpt_entry_target *t;
 260                         int hdr_len;
 261
 262                         hdr_len = sizeof(*arp) + (2 * sizeof(struct in_addr)) +
 263                                 (2 * (*pskb)->dev->addr_len);
 264                         ADD_COUNTER(e->counters, hdr_len, 1);
 265
 266                         t = arpt_get_target(e);
 267
 268                         /* Standard target? */
 269                         if (!t->u.kernel.target->target) {
 270                                 int v;
 271
 272                                 v = ((struct arpt_standard_target *)t)->verdict;
 273                                 if (v < 0) {
 274                                         /* Pop from stack? */
 275                                         if (v != ARPT_RETURN) {
 276                                                 verdict = (unsigned)(-v) - 1;
 277                                                 break;
 278                                         }
 279                                         e = back;
 280                                         back = get_entry(table_base,
 281                                                          back->comefrom);
 282                                         continue;
 283                                 }
 284                                 if (table_base + v
 285                                     != (void *)e + e->next_offset) {
 286                                         /* Save old back ptr in next entry */
 287                                         struct arpt_entry *next
 288                                                 = (void *)e + e->next_offset;
 289                                         next->comefrom =
 290                                                 (void *)back - table_base;
 291
 292                                         /* set back pointer to next entry */
 293                                         back = next;
 294                                 }
 295
 296                                 e = get_entry(table_base, v);
 297                         } else {
 298                                 /* Targets which reenter must return
 299                                  * abs. verdicts
 300                                  */
 301                                 verdict = t->u.kernel.target->target(pskb,
 302                                                                      in, out,
 303                                                                      hook,
 304                                                                      t->data,
 305                                                                      userdata);
 306
 307                                 /* Target might have changed stuff. */
 308                                 arp = (*pskb)->nh.arph;
 309
 310                                 if (verdict == ARPT_CONTINUE)
 311                                         e = (void *)e + e->next_offset;
 312                                 else
 313                                         /* Verdict */
 314                                         break;
 315                         }
 316                 } else {
 317                         e = (void *)e + e->next_offset;
 318                 }
 319         } while (!hotdrop);
 320         read_unlock_bh(&table->lock);
 321
 322         if (hotdrop)
 323                 return NF_DROP;
 324         else
 325                 return verdict;
 326 }
 327
 328 /* All zeroes == unconditional rule. */
 329 static inline int unconditional(const struct arpt_arp *arp)
 330 {
 331         unsigned int i;
 332
 333         for (i = 0; i < sizeof(*arp)/sizeof(__u32); i++)
 334                 if (((__u32 *)arp)[i])
 335                         return 0;
 336
 337         return 1;
 338 }
 339
 340 /* Figures out from what hook each rule can be called: returns 0 if
 341  * there are loops.  Puts hook bitmask in comefrom.
 342  */
 343 static int mark_source_chains(struct xt_table_info *newinfo,
 344                               unsigned int valid_hooks, void *entry0)
 345 {
 346         unsigned int hook;
 347
 348         /* No recursion; use packet counter to save back ptrs (reset
 349          * to 0 as we leave), and comefrom to save source hook bitmask.
 350          */
 351         for (hook = 0; hook < NF_ARP_NUMHOOKS; hook++) {
 352                 unsigned int pos = newinfo->hook_entry[hook];
 353                 struct arpt_entry *e
 354                         = (struct arpt_entry *)(entry0 + pos);
 355
 356                 if (!(valid_hooks & (1 << hook)))
 357                         continue;
 358
 359                 /* Set initial back pointer. */
 360                 e->counters.pcnt = pos;
 361
 362                 for (;;) {
 363                         struct arpt_standard_target *t
 364                                 = (void *)arpt_get_target(e);
 365
 366                         if (e->comefrom & (1 << NF_ARP_NUMHOOKS)) {
 367                                 printk("arptables: loop hook %u pos %u %08X.\n",
 368                                        hook, pos, e->comefrom);
 369                                 return 0;
 370                         }
 371                         e->comefrom
 372                                 |= ((1 << hook) | (1 << NF_ARP_NUMHOOKS));
 373
 374                         /* Unconditional return/END. */
 375                         if (e->target_offset == sizeof(struct arpt_entry)
 376                             && (strcmp(t->target.u.user.name,
 377                                        ARPT_STANDARD_TARGET) == 0)
 378                             && t->verdict < 0
 379                             && unconditional(&e->arp)) {
 380                                 unsigned int oldpos, size;
 381
 382                                 /* Return: backtrack through the last
 383                                  * big jump.
 384                                  */
 385                                 do {
 386                                         e->comefrom ^= (1<<NF_ARP_NUMHOOKS);
 387                                         oldpos = pos;
 388                                         pos = e->counters.pcnt;
 389                                         e->counters.pcnt = 0;
 390
 391                                         /* We're at the start. */
 392                                         if (pos == oldpos)
 393                                                 goto next;
 394
 395                                         e = (struct arpt_entry *)
 396                                                 (entry0 + pos);
 397                                 } while (oldpos == pos + e->next_offset);
 398
 399                                 /* Move along one */
 400                                 size = e->next_offset;
 401                                 e = (struct arpt_entry *)
 402                                         (entry0 + pos + size);
 403                                 e->counters.pcnt = pos;
 404                                 pos += size;
 405                         } else {
 406                                 int newpos = t->verdict;
 407
 408                                 if (strcmp(t->target.u.user.name,
 409                                            ARPT_STANDARD_TARGET) == 0
 410                                     && newpos >= 0) {
 411                                         /* This a jump; chase it. */
 412                                         duprintf("Jump rule %u -> %u\n",
 413                                                  pos, newpos);
 414                                 } else {
 415                                         /* ... this is a fallthru */
 416                                         newpos = pos + e->next_offset;
 417                                 }
 418                                 e = (struct arpt_entry *)
 419                                         (entry0 + newpos);
 420                                 e->counters.pcnt = pos;
 421                                 pos = newpos;
 422                         }
 423                 }
 424                 next:
 425                 duprintf("Finished chain %u\n", hook);
 426         }
 427         return 1;
 428 }
 429
 430 static inline int standard_check(const struct arpt_entry_target *t,
 431                                  unsigned int max_offset)
 432 {
 433         struct arpt_standard_target *targ = (void *)t;
 434
 435         /* Check standard info. */
 436         if (t->u.target_size
 437             != ARPT_ALIGN(sizeof(struct arpt_standard_target))) {
 438                 duprintf("arpt_standard_check: target size %u != %Zu\n",
 439                          t->u.target_size,
 440                          ARPT_ALIGN(sizeof(struct arpt_standard_target)));
 441                 return 0;
 442         }
 443
 444         if (targ->verdict >= 0
 445             && targ->verdict > max_offset - sizeof(struct arpt_entry)) {
 446                 duprintf("arpt_standard_check: bad verdict (%i)\n",
 447                          targ->verdict);
 448                 return 0;
 449         }
 450
 451         if (targ->verdict < -NF_MAX_VERDICT - 1) {
 452                 duprintf("arpt_standard_check: bad negative verdict (%i)\n",
 453                          targ->verdict);
 454                 return 0;
 455         }
 456         return 1;
 457 }
 458
 459 static struct arpt_target arpt_standard_target;
 460
 461 static inline int check_entry(struct arpt_entry *e, const char *name, unsigned int size,
 462                               unsigned int *i)
 463 {
 464         struct arpt_entry_target *t;
 465         struct arpt_target *target;
 466         int ret;
 467
 468         if (!arp_checkentry(&e->arp)) {
 469                 duprintf("arp_tables: arp check failed %p %s.\n", e, name);
 470                 return -EINVAL;
 471         }
 472
 473         t = arpt_get_target(e);
 474         target = try_then_request_module(xt_find_target(NF_ARP, t->u.user.name,
 475                                                         t->u.user.revision),
 476                                          "arpt_%s", t->u.user.name);
 477         if (IS_ERR(target) || !target) {
 478                 duprintf("check_entry: `%s' not found\n", t->u.user.name);
 479                 ret = target ? PTR_ERR(target) : -ENOENT;
 480                 goto out;
 481         }
 482         t->u.kernel.target = target;
 483
 484         if (t->u.kernel.target == &arpt_standard_target) {
 485                 if (!standard_check(t, size)) {
 486                         ret = -EINVAL;
 487                         goto out;
 488                 }
 489         } else if (t->u.kernel.target->checkentry
 490                    && !t->u.kernel.target->checkentry(name, e, t->data,
 491                                                       t->u.target_size
 492                                                       - sizeof(*t),
 493                                                       e->comefrom)) {
 494                 module_put(t->u.kernel.target->me);
 495                 duprintf("arp_tables: check failed for `%s'.\n",
 496                          t->u.kernel.target->name);
 497                 ret = -EINVAL;
 498                 goto out;
 499         }
 500
 501         (*i)++;
 502         return 0;
 503
 504 out:
 505         return ret;
 506 }
 507
 508 static inline int check_entry_size_and_hooks(struct arpt_entry *e,
 509                                              struct xt_table_info *newinfo,
 510                                              unsigned char *base,
 511                                              unsigned char *limit,
 512                                              const unsigned int *hook_entries,
 513                                              const unsigned int *underflows,
 514                                              unsigned int *i)
 515 {
 516         unsigned int h;
 517
 518         if ((unsigned long)e % __alignof__(struct arpt_entry) != 0
 519             || (unsigned char *)e + sizeof(struct arpt_entry) >= limit) {
 520                 duprintf("Bad offset %p\n", e);
 521                 return -EINVAL;
 522         }
 523
 524         if (e->next_offset
 525             < sizeof(struct arpt_entry) + sizeof(struct arpt_entry_target)) {
 526                 duprintf("checking: element %p size %u\n",
 527                          e, e->next_offset);
 528                 return -EINVAL;
 529         }
 530
 531         /* Check hooks & underflows */
 532         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
 533                 if ((unsigned char *)e - base == hook_entries[h])
 534                         newinfo->hook_entry[h] = hook_entries[h];
 535                 if ((unsigned char *)e - base == underflows[h])
 536                         newinfo->underflow[h] = underflows[h];
 537         }
 538
 539         /* FIXME: underflows must be unconditional, standard verdicts
 540            < 0 (not ARPT_RETURN). --RR */
 541
 542         /* Clear counters and comefrom */
 543         e->counters = ((struct xt_counters) { 0, 0 });
 544         e->comefrom = 0;
 545
 546         (*i)++;
 547         return 0;
 548 }
 549
 550 static inline int cleanup_entry(struct arpt_entry *e, unsigned int *i)
 551 {
 552         struct arpt_entry_target *t;
 553
 554         if (i && (*i)-- == 0)
 555                 return 1;
 556
 557         t = arpt_get_target(e);
 558         if (t->u.kernel.target->destroy)
 559                 t->u.kernel.target->destroy(t->data,
 560                                             t->u.target_size - sizeof(*t));
 561         module_put(t->u.kernel.target->me);
 562         return 0;
 563 }
 564
 565 /* Checks and translates the user-supplied table segment (held in
 566  * newinfo).
 567  */
 568 static int translate_table(const char *name,
 569                            unsigned int valid_hooks,
 570                            struct xt_table_info *newinfo,
 571                            void *entry0,
 572                            unsigned int size,
 573                            unsigned int number,
 574                            const unsigned int *hook_entries,
 575                            const unsigned int *underflows)
 576 {
 577         unsigned int i;
 578         int ret;
 579
 580         newinfo->size = size;
 581         newinfo->number = number;
 582
 583         /* Init all hooks to impossible value. */
 584         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
 585                 newinfo->hook_entry[i] = 0xFFFFFFFF;
 586                 newinfo->underflow[i] = 0xFFFFFFFF;
 587         }
 588
 589         duprintf("translate_table: size %u\n", newinfo->size);
 590         i = 0;
 591
 592         /* Walk through entries, checking offsets. */
 593         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 594                                  check_entry_size_and_hooks,
 595                                  newinfo,
 596                                  entry0,
 597                                  entry0 + size,
 598                                  hook_entries, underflows, &i);
 599         duprintf("translate_table: ARPT_ENTRY_ITERATE gives %d\n", ret);
 600         if (ret != 0)
 601                 return ret;
 602
 603         if (i != number) {
 604                 duprintf("translate_table: %u not %u entries\n",
 605                          i, number);
 606                 return -EINVAL;
 607         }
 608
 609         /* Check hooks all assigned */
 610         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
 611                 /* Only hooks which are valid */
 612                 if (!(valid_hooks & (1 << i)))
 613                         continue;
 614                 if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
 615                         duprintf("Invalid hook entry %u %u\n",
 616                                  i, hook_entries[i]);
 617                         return -EINVAL;
 618                 }
 619                 if (newinfo->underflow[i] == 0xFFFFFFFF) {
 620                         duprintf("Invalid underflow %u %u\n",
 621                                  i, underflows[i]);
 622                         return -EINVAL;
 623                 }
 624         }
 625
 626         if (!mark_source_chains(newinfo, valid_hooks, entry0)) {
 627                 duprintf("Looping hook\n");
 628                 return -ELOOP;
 629         }
 630
 631         /* Finally, each sanity check must pass */
 632         i = 0;
 633         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 634                                  check_entry, name, size, &i);
 635
 636         if (ret != 0) {
 637                 ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 638                                    cleanup_entry, &i);
 639                 return ret;
 640         }
 641
 642         /* And one copy for every other CPU */
 643         for_each_cpu(i) {
 644                 if (newinfo->entries[i] && newinfo->entries[i] != entry0)
 645                         memcpy(newinfo->entries[i], entry0, newinfo->size);
 646         }
 647
 648         return ret;
 649 }
 650
 651 /* Gets counters. */
 652 static inline int add_entry_to_counter(const struct arpt_entry *e,
 653                                        struct xt_counters total[],
 654                                        unsigned int *i)
 655 {
 656         ADD_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
 657
 658         (*i)++;
 659         return 0;
 660 }
 661
 662 static inline int set_entry_to_counter(const struct arpt_entry *e,
 663                                        struct xt_counters total[],
 664                                        unsigned int *i)
 665 {
 666         SET_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
 667
 668         (*i)++;
 669         return 0;
 670 }
 671
 672 static void get_counters(const struct xt_table_info *t,
 673                          struct xt_counters counters[])
 674 {
 675         unsigned int cpu;
 676         unsigned int i;
 677         unsigned int curcpu;
 678
 679         /* Instead of clearing (by a previous call to memset())
 680          * the counters and using adds, we set the counters
 681          * with data used by 'current' CPU
 682          * We dont care about preemption here.
 683          */
 684         curcpu = raw_smp_processor_id();
 685
 686         i = 0;
 687         ARPT_ENTRY_ITERATE(t->entries[curcpu],
 688                            t->size,
 689                            set_entry_to_counter,
 690                            counters,
 691                            &i);
 692
 693         for_each_cpu(cpu) {
 694                 if (cpu == curcpu)
 695                         continue;
 696                 i = 0;
 697                 ARPT_ENTRY_ITERATE(t->entries[cpu],
 698                                    t->size,
 699                                    add_entry_to_counter,
 700                                    counters,
 701                                    &i);
 702         }
 703 }
 704
 705 static int copy_entries_to_user(unsigned int total_size,
 706                                 struct arpt_table *table,
 707                                 void __user *userptr)
 708 {
 709         unsigned int off, num, countersize;
 710         struct arpt_entry *e;
 711         struct xt_counters *counters;
 712         struct xt_table_info *private = table->private;
 713         int ret = 0;
 714         void *loc_cpu_entry;
 715
 716         /* We need atomic snapshot of counters: rest doesn't change
 717          * (other than comefrom, which userspace doesn't care
 718          * about).
 719          */
 720         countersize = sizeof(struct xt_counters) * private->number;
 721         counters = vmalloc_node(countersize, numa_node_id());
 722
 723         if (counters == NULL)
 724                 return -ENOMEM;
 725
 726         /* First, sum counters... */
 727         write_lock_bh(&table->lock);
 728         get_counters(private, counters);
 729         write_unlock_bh(&table->lock);
 730
 731         loc_cpu_entry = private->entries[raw_smp_processor_id()];
 732         /* ... then copy entire thing ... */
 733         if (copy_to_user(userptr, loc_cpu_entry, total_size) != 0) {
 734                 ret = -EFAULT;
 735                 goto free_counters;
 736         }
 737
 738         /* FIXME: use iterator macros --RR */
 739         /* ... then go back and fix counters and names */
 740         for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
 741                 struct arpt_entry_target *t;
 742
 743                 e = (struct arpt_entry *)(loc_cpu_entry + off);
 744                 if (copy_to_user(userptr + off
 745                                  + offsetof(struct arpt_entry, counters),
 746                                  &counters[num],
 747                                  sizeof(counters[num])) != 0) {
 748                         ret = -EFAULT;
 749                         goto free_counters;
 750                 }
 751
 752                 t = arpt_get_target(e);
 753                 if (copy_to_user(userptr + off + e->target_offset
 754                                  + offsetof(struct arpt_entry_target,
 755                                             u.user.name),
 756                                  t->u.kernel.target->name,
 757                                  strlen(t->u.kernel.target->name)+1) != 0) {
 758                         ret = -EFAULT;
 759                         goto free_counters;
 760                 }
 761         }
 762
 763  free_counters:
 764         vfree(counters);
 765         return ret;
 766 }
 767
 768 static int get_entries(const struct arpt_get_entries *entries,
 769                        struct arpt_get_entries __user *uptr)
 770 {
 771         int ret;
 772         struct arpt_table *t;
 773
 774         t = xt_find_table_lock(NF_ARP, entries->name);
 775         if (t && !IS_ERR(t)) {
 776                 struct xt_table_info *private = t->private;
 777                 duprintf("t->private->number = %u\n",
 778                          private->number);
 779                 if (entries->size == private->size)
 780                         ret = copy_entries_to_user(private->size,
 781                                                    t, uptr->entrytable);
 782                 else {
 783                         duprintf("get_entries: I've got %u not %u!\n",
 784                                  private->size, entries->size);
 785                         ret = -EINVAL;
 786                 }
 787                 module_put(t->me);
 788                 xt_table_unlock(t);
 789         } else
 790                 ret = t ? PTR_ERR(t) : -ENOENT;
 791
 792         return ret;
 793 }
 794
 795 static int do_replace(void __user *user, unsigned int len)
 796 {
 797         int ret;
 798         struct arpt_replace tmp;
 799         struct arpt_table *t;
 800         struct xt_table_info *newinfo, *oldinfo;
 801         struct xt_counters *counters;
 802         void *loc_cpu_entry, *loc_cpu_old_entry;
 803
 804         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
 805                 return -EFAULT;
 806
 807         /* Hack: Causes ipchains to give correct error msg --RR */
 808         if (len != sizeof(tmp) + tmp.size)
 809                 return -ENOPROTOOPT;
 810
 811         /* overflow check */
 812         if (tmp.size >= (INT_MAX - sizeof(struct xt_table_info)) / NR_CPUS -
 813                         SMP_CACHE_BYTES)
 814                 return -ENOMEM;
 815         if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
 816                 return -ENOMEM;
 817
 818         newinfo = xt_alloc_table_info(tmp.size);
 819         if (!newinfo)
 820                 return -ENOMEM;
 821
 822         /* choose the copy that is on our node/cpu */
 823         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
 824         if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
 825                            tmp.size) != 0) {
 826                 ret = -EFAULT;
 827                 goto free_newinfo;
 828         }
 829
 830         counters = vmalloc(tmp.num_counters * sizeof(struct xt_counters));
 831         if (!counters) {
 832                 ret = -ENOMEM;
 833                 goto free_newinfo;
 834         }
 835
 836         ret = translate_table(tmp.name, tmp.valid_hooks,
 837                               newinfo, loc_cpu_entry, tmp.size, tmp.num_entries,
 838                               tmp.hook_entry, tmp.underflow);
 839         if (ret != 0)
 840                 goto free_newinfo_counters;
 841
 842         duprintf("arp_tables: Translated table\n");
 843
 844         t = try_then_request_module(xt_find_table_lock(NF_ARP, tmp.name),
 845                                     "arptable_%s", tmp.name);
 846         if (!t || IS_ERR(t)) {
 847                 ret = t ? PTR_ERR(t) : -ENOENT;
 848                 goto free_newinfo_counters_untrans;
 849         }
 850
 851         /* You lied! */
 852         if (tmp.valid_hooks != t->valid_hooks) {
 853                 duprintf("Valid hook crap: %08X vs %08X\n",
 854                          tmp.valid_hooks, t->valid_hooks);
 855                 ret = -EINVAL;
 856                 goto put_module;
 857         }
 858
 859         oldinfo = xt_replace_table(t, tmp.num_counters, newinfo, &ret);
 860         if (!oldinfo)
 861                 goto put_module;
 862
 863         /* Update module usage count based on number of rules */
 864         duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
 865                 oldinfo->number, oldinfo->initial_entries, newinfo->number);
 866         if ((oldinfo->number > oldinfo->initial_entries) ||
 867             (newinfo->number <= oldinfo->initial_entries))
 868                 module_put(t->me);
 869         if ((oldinfo->number > oldinfo->initial_entries) &&
 870             (newinfo->number <= oldinfo->initial_entries))
 871                 module_put(t->me);
 872
 873         /* Get the old counters. */
 874         get_counters(oldinfo, counters);
 875         /* Decrease module usage counts and free resource */
 876         loc_cpu_old_entry = oldinfo->entries[raw_smp_processor_id()];
 877         ARPT_ENTRY_ITERATE(loc_cpu_old_entry, oldinfo->size, cleanup_entry,NULL);
 878
 879         xt_free_table_info(oldinfo);
 880         if (copy_to_user(tmp.counters, counters,
 881                          sizeof(struct xt_counters) * tmp.num_counters) != 0)
 882                 ret = -EFAULT;
 883         vfree(counters);
 884         xt_table_unlock(t);
 885         return ret;
 886
 887  put_module:
 888         module_put(t->me);
 889         xt_table_unlock(t);
 890  free_newinfo_counters_untrans:
 891         ARPT_ENTRY_ITERATE(loc_cpu_entry, newinfo->size, cleanup_entry, NULL);
 892  free_newinfo_counters:
 893         vfree(counters);
 894  free_newinfo:
 895         xt_free_table_info(newinfo);
 896         return ret;
 897 }
 898
 899 /* We're lazy, and add to the first CPU; overflow works its fey magic
 900  * and everything is OK.
 901  */
 902 static inline int add_counter_to_entry(struct arpt_entry *e,
 903                                        const struct xt_counters addme[],
 904                                        unsigned int *i)
 905 {
 906
 907         ADD_COUNTER(e->counters, addme[*i].bcnt, addme[*i].pcnt);
 908
 909         (*i)++;
 910         return 0;
 911 }
 912
 913 static int do_add_counters(void __user *user, unsigned int len)
 914 {
 915         unsigned int i;
 916         struct xt_counters_info tmp, *paddc;
 917         struct arpt_table *t;
 918         struct xt_table_info *private;
 919         int ret = 0;
 920         void *loc_cpu_entry;
 921
 922         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
 923                 return -EFAULT;
 924
 925         if (len != sizeof(tmp) + tmp.num_counters*sizeof(struct xt_counters))
 926                 return -EINVAL;
 927
 928         paddc = vmalloc(len);
 929         if (!paddc)
 930                 return -ENOMEM;
 931
 932         if (copy_from_user(paddc, user, len) != 0) {
 933                 ret = -EFAULT;
 934                 goto free;
 935         }
 936
 937         t = xt_find_table_lock(NF_ARP, tmp.name);
 938         if (!t || IS_ERR(t)) {
 939                 ret = t ? PTR_ERR(t) : -ENOENT;
 940                 goto free;
 941         }
 942
 943         write_lock_bh(&t->lock);
 944         private = t->private;
 945         if (private->number != tmp.num_counters) {
 946                 ret = -EINVAL;
 947                 goto unlock_up_free;
 948         }
 949
 950         i = 0;
 951         /* Choose the copy that is on our node */
 952         loc_cpu_entry = private->entries[smp_processor_id()];
 953         ARPT_ENTRY_ITERATE(loc_cpu_entry,
 954                            private->size,
 955                            add_counter_to_entry,
 956                            paddc->counters,
 957                            &i);
 958  unlock_up_free:
 959         write_unlock_bh(&t->lock);
 960         xt_table_unlock(t);
 961         module_put(t->me);
 962  free:
 963         vfree(paddc);
 964
 965         return ret;
 966 }
 967
 968 static int do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
 969 {
 970         int ret;
 971
 972         if (!capable(CAP_NET_ADMIN))
 973                 return -EPERM;
 974
 975         switch (cmd) {
 976         case ARPT_SO_SET_REPLACE:
 977                 ret = do_replace(user, len);
 978                 break;
 979
 980         case ARPT_SO_SET_ADD_COUNTERS:
 981                 ret = do_add_counters(user, len);
 982                 break;
 983
 984         default:
 985                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
 986                 ret = -EINVAL;
 987         }
 988
 989         return ret;
 990 }
 991
 992 static int do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
 993 {
 994         int ret;
 995
 996         if (!capable(CAP_NET_ADMIN))
 997                 return -EPERM;
 998
 999         switch (cmd) {
1000         case ARPT_SO_GET_INFO: {
1001                 char name[ARPT_TABLE_MAXNAMELEN];
1002                 struct arpt_table *t;
1003
1004                 if (*len != sizeof(struct arpt_getinfo)) {
1005                         duprintf("length %u != %Zu\n", *len,
1006                                  sizeof(struct arpt_getinfo));
1007                         ret = -EINVAL;
1008                         break;
1009                 }
1010
1011                 if (copy_from_user(name, user, sizeof(name)) != 0) {
1012                         ret = -EFAULT;
1013                         break;
1014                 }
1015                 name[ARPT_TABLE_MAXNAMELEN-1] = '\0';
1016
1017                 t = try_then_request_module(xt_find_table_lock(NF_ARP, name),
1018                                             "arptable_%s", name);
1019                 if (t && !IS_ERR(t)) {
1020                         struct arpt_getinfo info;
1021                         struct xt_table_info *private = t->private;
1022
1023                         info.valid_hooks = t->valid_hooks;
1024                         memcpy(info.hook_entry, private->hook_entry,
1025                                sizeof(info.hook_entry));
1026                         memcpy(info.underflow, private->underflow,
1027                                sizeof(info.underflow));
1028                         info.num_entries = private->number;
1029                         info.size = private->size;
1030                         strcpy(info.name, name);
1031
1032                         if (copy_to_user(user, &info, *len) != 0)
1033                                 ret = -EFAULT;
1034                         else
1035                                 ret = 0;
1036                         xt_table_unlock(t);
1037                         module_put(t->me);
1038                 } else
1039                         ret = t ? PTR_ERR(t) : -ENOENT;
1040         }
1041         break;
1042
1043         case ARPT_SO_GET_ENTRIES: {
1044                 struct arpt_get_entries get;
1045
1046                 if (*len < sizeof(get)) {
1047                         duprintf("get_entries: %u < %Zu\n", *len, sizeof(get));
1048                         ret = -EINVAL;
1049                 } else if (copy_from_user(&get, user, sizeof(get)) != 0) {
1050                         ret = -EFAULT;
1051                 } else if (*len != sizeof(struct arpt_get_entries) + get.size) {
1052                         duprintf("get_entries: %u != %Zu\n", *len,
1053                                  sizeof(struct arpt_get_entries) + get.size);
1054                         ret = -EINVAL;
1055                 } else
1056                         ret = get_entries(&get, user);
1057                 break;
1058         }
1059
1060         case ARPT_SO_GET_REVISION_TARGET: {
1061                 struct xt_get_revision rev;
1062
1063                 if (*len != sizeof(rev)) {
1064                         ret = -EINVAL;
1065                         break;
1066                 }
1067                 if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
1068                         ret = -EFAULT;
1069                         break;
1070                 }
1071
1072                 try_then_request_module(xt_find_revision(NF_ARP, rev.name,
1073                                                          rev.revision, 1, &ret),
1074                                         "arpt_%s", rev.name);
1075                 break;
1076         }
1077
1078         default:
1079                 duprintf("do_arpt_get_ctl: unknown request %i\n", cmd);
1080                 ret = -EINVAL;
1081         }
1082
1083         return ret;
1084 }
1085
1086 int arpt_register_table(struct arpt_table *table,
1087                         const struct arpt_replace *repl)
1088 {
1089         int ret;
1090         struct xt_table_info *newinfo;
1091         static struct xt_table_info bootstrap
1092                 = { 0, 0, 0, { 0 }, { 0 }, { } };
1093         void *loc_cpu_entry;
1094
1095         newinfo = xt_alloc_table_info(repl->size);
1096         if (!newinfo) {
1097                 ret = -ENOMEM;
1098                 return ret;
1099         }
1100
1101         /* choose the copy on our node/cpu */
1102         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1103         memcpy(loc_cpu_entry, repl->entries, repl->size);
1104
1105         ret = translate_table(table->name, table->valid_hooks,
1106                               newinfo, loc_cpu_entry, repl->size,
1107                               repl->num_entries,
1108                               repl->hook_entry,
1109                               repl->underflow);
1110
1111         duprintf("arpt_register_table: translate table gives %d\n", ret);
1112         if (ret != 0) {
1113                 xt_free_table_info(newinfo);
1114                 return ret;
1115         }
1116
1117         if (xt_register_table(table, &bootstrap, newinfo) != 0) {
1118                 xt_free_table_info(newinfo);
1119                 return ret;
1120         }
1121
1122         return 0;
1123 }
1124
1125 void arpt_unregister_table(struct arpt_table *table)
1126 {
1127         struct xt_table_info *private;
1128         void *loc_cpu_entry;
1129
1130         private = xt_unregister_table(table);
1131
1132         /* Decrease module usage counts and free resources */
1133         loc_cpu_entry = private->entries[raw_smp_processor_id()];
1134         ARPT_ENTRY_ITERATE(loc_cpu_entry, private->size,
1135                            cleanup_entry, NULL);
1136         xt_free_table_info(private);
1137 }
1138
1139 /* The built-in targets: standard (NULL) and error. */
1140 static struct arpt_target arpt_standard_target = {
1141         .name           = ARPT_STANDARD_TARGET,
1142 };
1143
1144 static struct arpt_target arpt_error_target = {
1145         .name           = ARPT_ERROR_TARGET,
1146         .target         = arpt_error,
1147 };
1148
1149 static struct nf_sockopt_ops arpt_sockopts = {
1150         .pf             = PF_INET,
1151         .set_optmin     = ARPT_BASE_CTL,
1152         .set_optmax     = ARPT_SO_SET_MAX+1,
1153         .set            = do_arpt_set_ctl,
1154         .get_optmin     = ARPT_BASE_CTL,
1155         .get_optmax     = ARPT_SO_GET_MAX+1,
1156         .get            = do_arpt_get_ctl,
1157 };
1158
1159 static int __init init(void)
1160 {
1161         int ret;
1162
1163         xt_proto_init(NF_ARP);
1164
1165         /* Noone else will be downing sem now, so we won't sleep */
1166         xt_register_target(NF_ARP, &arpt_standard_target);
1167         xt_register_target(NF_ARP, &arpt_error_target);
1168
1169         /* Register setsockopt */
1170         ret = nf_register_sockopt(&arpt_sockopts);
1171         if (ret < 0) {
1172                 duprintf("Unable to register sockopts.\n");
1173                 return ret;
1174         }
1175
1176         printk("arp_tables: (C) 2002 David S. Miller\n");
1177         return 0;
1178 }
1179
1180 static void __exit fini(void)
1181 {
1182         nf_unregister_sockopt(&arpt_sockopts);
1183         xt_proto_fini(NF_ARP);
1184 }
1185
1186 EXPORT_SYMBOL(arpt_register_table);
1187 EXPORT_SYMBOL(arpt_unregister_table);
1188 EXPORT_SYMBOL(arpt_do_table);
1189
1190 module_init(init);
1191 module_exit(fini);