net/ipv4/netfilter/arp_tables.c

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