SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/ipv4/netfilter/arp_tables.c

   1 /*
   2  * Packet matching code for ARP packets.
   3  *
   4  * Based heavily, if not almost entirely, upon ip_tables.c framework.
   5  *
   6  * Some ARP specific bits are:
   7  *
   8  * Copyright (C) 2002 David S. Miller (davem@redhat.com)
   9  *
  10  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/netdevice.h>
  15 #include <linux/capability.h>
  16 #include <linux/if_arp.h>
  17 #include <linux/kmod.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/module.h>
  21 #include <linux/init.h>
  22
  23 #include <asm/uaccess.h>
  24 #include <linux/mutex.h>
  25
  26 #include <linux/netfilter/x_tables.h>
  27 #include <linux/netfilter_arp/arp_tables.h>
  28
  29 MODULE_LICENSE("GPL");
  30 MODULE_AUTHOR("David S. Miller <davem@redhat.com>");
  31 MODULE_DESCRIPTION("arptables core");
  32
  33 /*#define DEBUG_ARP_TABLES*/
  34 /*#define DEBUG_ARP_TABLES_USER*/
  35
  36 #ifdef DEBUG_ARP_TABLES
  37 #define dprintf(format, args...)  printk(format , ## args)
  38 #else
  39 #define dprintf(format, args...)
  40 #endif
  41
  42 #ifdef DEBUG_ARP_TABLES_USER
  43 #define duprintf(format, args...) printk(format , ## args)
  44 #else
  45 #define duprintf(format, args...)
  46 #endif
  47
  48 #ifdef CONFIG_NETFILTER_DEBUG
  49 #define ARP_NF_ASSERT(x)                                        \
  50 do {                                                            \
  51         if (!(x))                                               \
  52                 printk("ARP_NF_ASSERT: %s:%s:%u\n",             \
  53                        __FUNCTION__, __FILE__, __LINE__);       \
  54 } while(0)
  55 #else
  56 #define ARP_NF_ASSERT(x)
  57 #endif
  58
  59 static inline int arp_devaddr_compare(const struct arpt_devaddr_info *ap,
  60                                       char *hdr_addr, int len)
  61 {
  62         int i, ret;
  63
  64         if (len > ARPT_DEV_ADDR_LEN_MAX)
  65                 len = ARPT_DEV_ADDR_LEN_MAX;
  66
  67         ret = 0;
  68         for (i = 0; i < len; i++)
  69                 ret |= (hdr_addr[i] ^ ap->addr[i]) & ap->mask[i];
  70
  71         return (ret != 0);
  72 }
  73
  74 /* Returns whether packet matches rule or not. */
  75 static inline int arp_packet_match(const struct arphdr *arphdr,
  76                                    struct net_device *dev,
  77                                    const char *indev,
  78                                    const char *outdev,
  79                                    const struct arpt_arp *arpinfo)
  80 {
  81         char *arpptr = (char *)(arphdr + 1);
  82         char *src_devaddr, *tgt_devaddr;
  83         __be32 src_ipaddr, tgt_ipaddr;
  84         int i, ret;
  85
  86 #define FWINV(bool,invflg) ((bool) ^ !!(arpinfo->invflags & invflg))
  87
  88         if (FWINV((arphdr->ar_op & arpinfo->arpop_mask) != arpinfo->arpop,
  89                   ARPT_INV_ARPOP)) {
  90                 dprintf("ARP operation field mismatch.\n");
  91                 dprintf("ar_op: %04x info->arpop: %04x info->arpop_mask: %04x\n",
  92                         arphdr->ar_op, arpinfo->arpop, arpinfo->arpop_mask);
  93                 return 0;
  94         }
  95
  96         if (FWINV((arphdr->ar_hrd & arpinfo->arhrd_mask) != arpinfo->arhrd,
  97                   ARPT_INV_ARPHRD)) {
  98                 dprintf("ARP hardware address format mismatch.\n");
  99                 dprintf("ar_hrd: %04x info->arhrd: %04x info->arhrd_mask: %04x\n",
 100                         arphdr->ar_hrd, arpinfo->arhrd, arpinfo->arhrd_mask);
 101                 return 0;
 102         }
 103
 104         if (FWINV((arphdr->ar_pro & arpinfo->arpro_mask) != arpinfo->arpro,
 105                   ARPT_INV_ARPPRO)) {
 106                 dprintf("ARP protocol address format mismatch.\n");
 107                 dprintf("ar_pro: %04x info->arpro: %04x info->arpro_mask: %04x\n",
 108                         arphdr->ar_pro, arpinfo->arpro, arpinfo->arpro_mask);
 109                 return 0;
 110         }
 111
 112         if (FWINV((arphdr->ar_hln & arpinfo->arhln_mask) != arpinfo->arhln,
 113                   ARPT_INV_ARPHLN)) {
 114                 dprintf("ARP hardware address length mismatch.\n");
 115                 dprintf("ar_hln: %02x info->arhln: %02x info->arhln_mask: %02x\n",
 116                         arphdr->ar_hln, arpinfo->arhln, arpinfo->arhln_mask);
 117                 return 0;
 118         }
 119
 120         src_devaddr = arpptr;
 121         arpptr += dev->addr_len;
 122         memcpy(&src_ipaddr, arpptr, sizeof(u32));
 123         arpptr += sizeof(u32);
 124         tgt_devaddr = arpptr;
 125         arpptr += dev->addr_len;
 126         memcpy(&tgt_ipaddr, arpptr, sizeof(u32));
 127
 128         if (FWINV(arp_devaddr_compare(&arpinfo->src_devaddr, src_devaddr, dev->addr_len),
 129                   ARPT_INV_SRCDEVADDR) ||
 130             FWINV(arp_devaddr_compare(&arpinfo->tgt_devaddr, tgt_devaddr, dev->addr_len),
 131                   ARPT_INV_TGTDEVADDR)) {
 132                 dprintf("Source or target device address mismatch.\n");
 133
 134                 return 0;
 135         }
 136
 137         if (FWINV((src_ipaddr & arpinfo->smsk.s_addr) != arpinfo->src.s_addr,
 138                   ARPT_INV_SRCIP) ||
 139             FWINV(((tgt_ipaddr & arpinfo->tmsk.s_addr) != arpinfo->tgt.s_addr),
 140                   ARPT_INV_TGTIP)) {
 141                 dprintf("Source or target IP address mismatch.\n");
 142
 143                 dprintf("SRC: %u.%u.%u.%u. Mask: %u.%u.%u.%u. Target: %u.%u.%u.%u.%s\n",
 144                         NIPQUAD(src_ipaddr),
 145                         NIPQUAD(arpinfo->smsk.s_addr),
 146                         NIPQUAD(arpinfo->src.s_addr),
 147                         arpinfo->invflags & ARPT_INV_SRCIP ? " (INV)" : "");
 148                 dprintf("TGT: %u.%u.%u.%u Mask: %u.%u.%u.%u Target: %u.%u.%u.%u.%s\n",
 149                         NIPQUAD(tgt_ipaddr),
 150                         NIPQUAD(arpinfo->tmsk.s_addr),
 151                         NIPQUAD(arpinfo->tgt.s_addr),
 152                         arpinfo->invflags & ARPT_INV_TGTIP ? " (INV)" : "");
 153                 return 0;
 154         }
 155
 156         /* Look for ifname matches.  */
 157         for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
 158                 ret |= (indev[i] ^ arpinfo->iniface[i])
 159                         & arpinfo->iniface_mask[i];
 160         }
 161
 162         if (FWINV(ret != 0, ARPT_INV_VIA_IN)) {
 163                 dprintf("VIA in mismatch (%s vs %s).%s\n",
 164                         indev, arpinfo->iniface,
 165                         arpinfo->invflags&ARPT_INV_VIA_IN ?" (INV)":"");
 166                 return 0;
 167         }
 168
 169         for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
 170                 ret |= (outdev[i] ^ arpinfo->outiface[i])
 171                         & arpinfo->outiface_mask[i];
 172         }
 173
 174         if (FWINV(ret != 0, ARPT_INV_VIA_OUT)) {
 175                 dprintf("VIA out mismatch (%s vs %s).%s\n",
 176                         outdev, arpinfo->outiface,
 177                         arpinfo->invflags&ARPT_INV_VIA_OUT ?" (INV)":"");
 178                 return 0;
 179         }
 180
 181         return 1;
 182 }
 183
 184 static inline int arp_checkentry(const struct arpt_arp *arp)
 185 {
 186         if (arp->flags & ~ARPT_F_MASK) {
 187                 duprintf("Unknown flag bits set: %08X\n",
 188                          arp->flags & ~ARPT_F_MASK);
 189                 return 0;
 190         }
 191         if (arp->invflags & ~ARPT_INV_MASK) {
 192                 duprintf("Unknown invflag bits set: %08X\n",
 193                          arp->invflags & ~ARPT_INV_MASK);
 194                 return 0;
 195         }
 196
 197         return 1;
 198 }
 199
 200 static unsigned int arpt_error(struct sk_buff *skb,
 201                                const struct net_device *in,
 202                                const struct net_device *out,
 203                                unsigned int hooknum,
 204                                const struct xt_target *target,
 205                                const void *targinfo)
 206 {
 207         if (net_ratelimit())
 208                 printk("arp_tables: error: '%s'\n", (char *)targinfo);
 209
 210         return NF_DROP;
 211 }
 212
 213 static inline struct arpt_entry *get_entry(void *base, unsigned int offset)
 214 {
 215         return (struct arpt_entry *)(base + offset);
 216 }
 217
 218 unsigned int arpt_do_table(struct sk_buff *skb,
 219                            unsigned int hook,
 220                            const struct net_device *in,
 221                            const struct net_device *out,
 222                            struct arpt_table *table)
 223 {
 224         static const char nulldevname[IFNAMSIZ];
 225         unsigned int verdict = NF_DROP;
 226         struct arphdr *arp;
 227         bool hotdrop = false;
 228         struct arpt_entry *e, *back;
 229         const char *indev, *outdev;
 230         void *table_base;
 231         struct xt_table_info *private;
 232
 233         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
 234         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
 235                                  (2 * skb->dev->addr_len) +
 236                                  (2 * sizeof(u32)))))
 237                 return NF_DROP;
 238
 239         indev = in ? in->name : nulldevname;
 240         outdev = out ? out->name : nulldevname;
 241
 242         read_lock_bh(&table->lock);
 243         private = table->private;
 244         table_base = (void *)private->entries[smp_processor_id()];
 245         e = get_entry(table_base, private->hook_entry[hook]);
 246         back = get_entry(table_base, private->underflow[hook]);
 247
 248         arp = arp_hdr(skb);
 249         do {
 250                 if (arp_packet_match(arp, skb->dev, indev, outdev, &e->arp)) {
 251                         struct arpt_entry_target *t;
 252                         int hdr_len;
 253
 254                         hdr_len = sizeof(*arp) + (2 * sizeof(struct in_addr)) +
 255                                 (2 * skb->dev->addr_len);
 256                         ADD_COUNTER(e->counters, hdr_len, 1);
 257
 258                         t = arpt_get_target(e);
 259
 260                         /* Standard target? */
 261                         if (!t->u.kernel.target->target) {
 262                                 int v;
 263
 264                                 v = ((struct arpt_standard_target *)t)->verdict;
 265                                 if (v < 0) {
 266                                         /* Pop from stack? */
 267                                         if (v != ARPT_RETURN) {
 268                                                 verdict = (unsigned)(-v) - 1;
 269                                                 break;
 270                                         }
 271                                         e = back;
 272                                         back = get_entry(table_base,
 273                                                          back->comefrom);
 274                                         continue;
 275                                 }
 276                                 if (table_base + v
 277                                     != (void *)e + e->next_offset) {
 278                                         /* Save old back ptr in next entry */
 279                                         struct arpt_entry *next
 280                                                 = (void *)e + e->next_offset;
 281                                         next->comefrom =
 282                                                 (void *)back - table_base;
 283
 284                                         /* set back pointer to next entry */
 285                                         back = next;
 286                                 }
 287
 288                                 e = get_entry(table_base, v);
 289                         } else {
 290                                 /* Targets which reenter must return
 291                                  * abs. verdicts
 292                                  */
 293                                 verdict = t->u.kernel.target->target(skb,
 294                                                                      in, out,
 295                                                                      hook,
 296                                                                      t->u.kernel.target,
 297                                                                      t->data);
 298
 299                                 /* Target might have changed stuff. */
 300                                 arp = arp_hdr(skb);
 301
 302                                 if (verdict == ARPT_CONTINUE)
 303                                         e = (void *)e + e->next_offset;
 304                                 else
 305                                         /* Verdict */
 306                                         break;
 307                         }
 308                 } else {
 309                         e = (void *)e + e->next_offset;
 310                 }
 311         } while (!hotdrop);
 312         read_unlock_bh(&table->lock);
 313
 314         if (hotdrop)
 315                 return NF_DROP;
 316         else
 317                 return verdict;
 318 }
 319
 320 /* All zeroes == unconditional rule. */
 321 static inline int unconditional(const struct arpt_arp *arp)
 322 {
 323         unsigned int i;
 324
 325         for (i = 0; i < sizeof(*arp)/sizeof(__u32); i++)
 326                 if (((__u32 *)arp)[i])
 327                         return 0;
 328
 329         return 1;
 330 }
 331
 332 /* Figures out from what hook each rule can be called: returns 0 if
 333  * there are loops.  Puts hook bitmask in comefrom.
 334  */
 335 static int mark_source_chains(struct xt_table_info *newinfo,
 336                               unsigned int valid_hooks, void *entry0)
 337 {
 338         unsigned int hook;
 339
 340         /* No recursion; use packet counter to save back ptrs (reset
 341          * to 0 as we leave), and comefrom to save source hook bitmask.
 342          */
 343         for (hook = 0; hook < NF_ARP_NUMHOOKS; hook++) {
 344                 unsigned int pos = newinfo->hook_entry[hook];
 345                 struct arpt_entry *e
 346                         = (struct arpt_entry *)(entry0 + pos);
 347
 348                 if (!(valid_hooks & (1 << hook)))
 349                         continue;
 350
 351                 /* Set initial back pointer. */
 352                 e->counters.pcnt = pos;
 353
 354                 for (;;) {
 355                         struct arpt_standard_target *t
 356                                 = (void *)arpt_get_target(e);
 357                         int visited = e->comefrom & (1 << hook);
 358
 359                         if (e->comefrom & (1 << NF_ARP_NUMHOOKS)) {
 360                                 printk("arptables: loop hook %u pos %u %08X.\n",
 361                                        hook, pos, e->comefrom);
 362                                 return 0;
 363                         }
 364                         e->comefrom
 365                                 |= ((1 << hook) | (1 << NF_ARP_NUMHOOKS));
 366
 367                         /* Unconditional return/END. */
 368                         if ((e->target_offset == sizeof(struct arpt_entry)
 369                             && (strcmp(t->target.u.user.name,
 370                                        ARPT_STANDARD_TARGET) == 0)
 371                             && t->verdict < 0
 372                             && unconditional(&e->arp)) || visited) {
 373                                 unsigned int oldpos, size;
 374
 375                                 if (t->verdict < -NF_MAX_VERDICT - 1) {
 376                                         duprintf("mark_source_chains: bad "
 377                                                 "negative verdict (%i)\n",
 378                                                                 t->verdict);
 379                                         return 0;
 380                                 }
 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                                         if (newpos > newinfo->size -
 412                                                 sizeof(struct arpt_entry)) {
 413                                                 duprintf("mark_source_chains: "
 414                                                         "bad verdict (%i)\n",
 415                                                                 newpos);
 416                                                 return 0;
 417                                         }
 418
 419                                         /* This a jump; chase it. */
 420                                         duprintf("Jump rule %u -> %u\n",
 421                                                  pos, newpos);
 422                                 } else {
 423                                         /* ... this is a fallthru */
 424                                         newpos = pos + e->next_offset;
 425                                 }
 426                                 e = (struct arpt_entry *)
 427                                         (entry0 + newpos);
 428                                 e->counters.pcnt = pos;
 429                                 pos = newpos;
 430                         }
 431                 }
 432                 next:
 433                 duprintf("Finished chain %u\n", hook);
 434         }
 435         return 1;
 436 }
 437
 438 static inline int check_entry(struct arpt_entry *e, const char *name, unsigned int size,
 439                               unsigned int *i)
 440 {
 441         struct arpt_entry_target *t;
 442         struct arpt_target *target;
 443         int ret;
 444
 445         if (!arp_checkentry(&e->arp)) {
 446                 duprintf("arp_tables: arp check failed %p %s.\n", e, name);
 447                 return -EINVAL;
 448         }
 449
 450         if (e->target_offset + sizeof(struct arpt_entry_target) > e->next_offset)
 451                 return -EINVAL;
 452
 453         t = arpt_get_target(e);
 454         if (e->target_offset + t->u.target_size > e->next_offset)
 455                 return -EINVAL;
 456
 457         target = try_then_request_module(xt_find_target(NF_ARP, t->u.user.name,
 458                                                         t->u.user.revision),
 459                                          "arpt_%s", t->u.user.name);
 460         if (IS_ERR(target) || !target) {
 461                 duprintf("check_entry: `%s' not found\n", t->u.user.name);
 462                 ret = target ? PTR_ERR(target) : -ENOENT;
 463                 goto out;
 464         }
 465         t->u.kernel.target = target;
 466
 467         ret = xt_check_target(target, NF_ARP, t->u.target_size - sizeof(*t),
 468                               name, e->comefrom, 0, 0);
 469         if (ret)
 470                 goto err;
 471
 472         if (t->u.kernel.target->checkentry
 473             && !t->u.kernel.target->checkentry(name, e, target, t->data,
 474                                                e->comefrom)) {
 475                 duprintf("arp_tables: check failed for `%s'.\n",
 476                          t->u.kernel.target->name);
 477                 ret = -EINVAL;
 478                 goto err;
 479         }
 480
 481         (*i)++;
 482         return 0;
 483 err:
 484         module_put(t->u.kernel.target->me);
 485 out:
 486         return ret;
 487 }
 488
 489 static inline int check_entry_size_and_hooks(struct arpt_entry *e,
 490                                              struct xt_table_info *newinfo,
 491                                              unsigned char *base,
 492                                              unsigned char *limit,
 493                                              const unsigned int *hook_entries,
 494                                              const unsigned int *underflows,
 495                                              unsigned int *i)
 496 {
 497         unsigned int h;
 498
 499         if ((unsigned long)e % __alignof__(struct arpt_entry) != 0
 500             || (unsigned char *)e + sizeof(struct arpt_entry) >= limit) {
 501                 duprintf("Bad offset %p\n", e);
 502                 return -EINVAL;
 503         }
 504
 505         if (e->next_offset
 506             < sizeof(struct arpt_entry) + sizeof(struct arpt_entry_target)) {
 507                 duprintf("checking: element %p size %u\n",
 508                          e, e->next_offset);
 509                 return -EINVAL;
 510         }
 511
 512         /* Check hooks & underflows */
 513         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
 514                 if ((unsigned char *)e - base == hook_entries[h])
 515                         newinfo->hook_entry[h] = hook_entries[h];
 516                 if ((unsigned char *)e - base == underflows[h])
 517                         newinfo->underflow[h] = underflows[h];
 518         }
 519
 520         /* FIXME: underflows must be unconditional, standard verdicts
 521            < 0 (not ARPT_RETURN). --RR */
 522
 523         /* Clear counters and comefrom */
 524         e->counters = ((struct xt_counters) { 0, 0 });
 525         e->comefrom = 0;
 526
 527         (*i)++;
 528         return 0;
 529 }
 530
 531 static inline int cleanup_entry(struct arpt_entry *e, unsigned int *i)
 532 {
 533         struct arpt_entry_target *t;
 534
 535         if (i && (*i)-- == 0)
 536                 return 1;
 537
 538         t = arpt_get_target(e);
 539         if (t->u.kernel.target->destroy)
 540                 t->u.kernel.target->destroy(t->u.kernel.target, t->data);
 541         module_put(t->u.kernel.target->me);
 542         return 0;
 543 }
 544
 545 /* Checks and translates the user-supplied table segment (held in
 546  * newinfo).
 547  */
 548 static int translate_table(const char *name,
 549                            unsigned int valid_hooks,
 550                            struct xt_table_info *newinfo,
 551                            void *entry0,
 552                            unsigned int size,
 553                            unsigned int number,
 554                            const unsigned int *hook_entries,
 555                            const unsigned int *underflows)
 556 {
 557         unsigned int i;
 558         int ret;
 559
 560         newinfo->size = size;
 561         newinfo->number = number;
 562
 563         /* Init all hooks to impossible value. */
 564         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
 565                 newinfo->hook_entry[i] = 0xFFFFFFFF;
 566                 newinfo->underflow[i] = 0xFFFFFFFF;
 567         }
 568
 569         duprintf("translate_table: size %u\n", newinfo->size);
 570         i = 0;
 571
 572         /* Walk through entries, checking offsets. */
 573         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 574                                  check_entry_size_and_hooks,
 575                                  newinfo,
 576                                  entry0,
 577                                  entry0 + size,
 578                                  hook_entries, underflows, &i);
 579         duprintf("translate_table: ARPT_ENTRY_ITERATE gives %d\n", ret);
 580         if (ret != 0)
 581                 return ret;
 582
 583         if (i != number) {
 584                 duprintf("translate_table: %u not %u entries\n",
 585                          i, number);
 586                 return -EINVAL;
 587         }
 588
 589         /* Check hooks all assigned */
 590         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
 591                 /* Only hooks which are valid */
 592                 if (!(valid_hooks & (1 << i)))
 593                         continue;
 594                 if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
 595                         duprintf("Invalid hook entry %u %u\n",
 596                                  i, hook_entries[i]);
 597                         return -EINVAL;
 598                 }
 599                 if (newinfo->underflow[i] == 0xFFFFFFFF) {
 600                         duprintf("Invalid underflow %u %u\n",
 601                                  i, underflows[i]);
 602                         return -EINVAL;
 603                 }
 604         }
 605
 606         if (!mark_source_chains(newinfo, valid_hooks, entry0)) {
 607                 duprintf("Looping hook\n");
 608                 return -ELOOP;
 609         }
 610
 611         /* Finally, each sanity check must pass */
 612         i = 0;
 613         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 614                                  check_entry, name, size, &i);
 615
 616         if (ret != 0) {
 617                 ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 618                                 cleanup_entry, &i);
 619                 return ret;
 620         }
 621
 622         /* And one copy for every other CPU */
 623         for_each_possible_cpu(i) {
 624                 if (newinfo->entries[i] && newinfo->entries[i] != entry0)
 625                         memcpy(newinfo->entries[i], entry0, newinfo->size);
 626         }
 627
 628         return ret;
 629 }
 630
 631 /* Gets counters. */
 632 static inline int add_entry_to_counter(const struct arpt_entry *e,
 633                                        struct xt_counters total[],
 634                                        unsigned int *i)
 635 {
 636         ADD_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
 637
 638         (*i)++;
 639         return 0;
 640 }
 641
 642 static inline int set_entry_to_counter(const struct arpt_entry *e,
 643                                        struct xt_counters total[],
 644                                        unsigned int *i)
 645 {
 646         SET_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
 647
 648         (*i)++;
 649         return 0;
 650 }
 651
 652 static void get_counters(const struct xt_table_info *t,
 653                          struct xt_counters counters[])
 654 {
 655         unsigned int cpu;
 656         unsigned int i;
 657         unsigned int curcpu;
 658
 659         /* Instead of clearing (by a previous call to memset())
 660          * the counters and using adds, we set the counters
 661          * with data used by 'current' CPU
 662          * We dont care about preemption here.
 663          */
 664         curcpu = raw_smp_processor_id();
 665
 666         i = 0;
 667         ARPT_ENTRY_ITERATE(t->entries[curcpu],
 668                            t->size,
 669                            set_entry_to_counter,
 670                            counters,
 671                            &i);
 672
 673         for_each_possible_cpu(cpu) {
 674                 if (cpu == curcpu)
 675                         continue;
 676                 i = 0;
 677                 ARPT_ENTRY_ITERATE(t->entries[cpu],
 678                                    t->size,
 679                                    add_entry_to_counter,
 680                                    counters,
 681                                    &i);
 682         }
 683 }
 684
 685 static int copy_entries_to_user(unsigned int total_size,
 686                                 struct arpt_table *table,
 687                                 void __user *userptr)
 688 {
 689         unsigned int off, num, countersize;
 690         struct arpt_entry *e;
 691         struct xt_counters *counters;
 692         struct xt_table_info *private = table->private;
 693         int ret = 0;
 694         void *loc_cpu_entry;
 695
 696         /* We need atomic snapshot of counters: rest doesn't change
 697          * (other than comefrom, which userspace doesn't care
 698          * about).
 699          */
 700         countersize = sizeof(struct xt_counters) * private->number;
 701         counters = vmalloc_node(countersize, numa_node_id());
 702
 703         if (counters == NULL)
 704                 return -ENOMEM;
 705
 706         /* First, sum counters... */
 707         write_lock_bh(&table->lock);
 708         get_counters(private, counters);
 709         write_unlock_bh(&table->lock);
 710
 711         loc_cpu_entry = private->entries[raw_smp_processor_id()];
 712         /* ... then copy entire thing ... */
 713         if (copy_to_user(userptr, loc_cpu_entry, total_size) != 0) {
 714                 ret = -EFAULT;
 715                 goto free_counters;
 716         }
 717
 718         /* FIXME: use iterator macros --RR */
 719         /* ... then go back and fix counters and names */
 720         for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
 721                 struct arpt_entry_target *t;
 722
 723                 e = (struct arpt_entry *)(loc_cpu_entry + off);
 724                 if (copy_to_user(userptr + off
 725                                  + offsetof(struct arpt_entry, counters),
 726                                  &counters[num],
 727                                  sizeof(counters[num])) != 0) {
 728                         ret = -EFAULT;
 729                         goto free_counters;
 730                 }
 731
 732                 t = arpt_get_target(e);
 733                 if (copy_to_user(userptr + off + e->target_offset
 734                                  + offsetof(struct arpt_entry_target,
 735                                             u.user.name),
 736                                  t->u.kernel.target->name,
 737                                  strlen(t->u.kernel.target->name)+1) != 0) {
 738                         ret = -EFAULT;
 739                         goto free_counters;
 740                 }
 741         }
 742
 743  free_counters:
 744         vfree(counters);
 745         return ret;
 746 }
 747
 748 static int get_entries(const struct arpt_get_entries *entries,
 749                        struct arpt_get_entries __user *uptr)
 750 {
 751         int ret;
 752         struct arpt_table *t;
 753
 754         t = xt_find_table_lock(NF_ARP, entries->name);
 755         if (t && !IS_ERR(t)) {
 756                 struct xt_table_info *private = t->private;
 757                 duprintf("t->private->number = %u\n",
 758                          private->number);
 759                 if (entries->size == private->size)
 760                         ret = copy_entries_to_user(private->size,
 761                                                    t, uptr->entrytable);
 762                 else {
 763                         duprintf("get_entries: I've got %u not %u!\n",
 764                                  private->size, entries->size);
 765                         ret = -EINVAL;
 766                 }
 767                 module_put(t->me);
 768                 xt_table_unlock(t);
 769         } else
 770                 ret = t ? PTR_ERR(t) : -ENOENT;
 771
 772         return ret;
 773 }
 774
 775 static int do_replace(void __user *user, unsigned int len)
 776 {
 777         int ret;
 778         struct arpt_replace tmp;
 779         struct arpt_table *t;
 780         struct xt_table_info *newinfo, *oldinfo;
 781         struct xt_counters *counters;
 782         void *loc_cpu_entry, *loc_cpu_old_entry;
 783
 784         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
 785                 return -EFAULT;
 786
 787         /* Hack: Causes ipchains to give correct error msg --RR */
 788         if (len != sizeof(tmp) + tmp.size)
 789                 return -ENOPROTOOPT;
 790
 791         /* overflow check */
 792         if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
 793                 return -ENOMEM;
 794
 795         newinfo = xt_alloc_table_info(tmp.size);
 796         if (!newinfo)
 797                 return -ENOMEM;
 798
 799         /* choose the copy that is on our node/cpu */
 800         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
 801         if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
 802                            tmp.size) != 0) {
 803                 ret = -EFAULT;
 804                 goto free_newinfo;
 805         }
 806
 807         counters = vmalloc_node(tmp.num_counters * sizeof(struct xt_counters),
 808                                 numa_node_id());
 809         if (!counters) {
 810                 ret = -ENOMEM;
 811                 goto free_newinfo;
 812         }
 813
 814         ret = translate_table(tmp.name, tmp.valid_hooks,
 815                               newinfo, loc_cpu_entry, tmp.size, tmp.num_entries,
 816                               tmp.hook_entry, tmp.underflow);
 817         if (ret != 0)
 818                 goto free_newinfo_counters;
 819
 820         duprintf("arp_tables: Translated table\n");
 821
 822         t = try_then_request_module(xt_find_table_lock(NF_ARP, tmp.name),
 823                                     "arptable_%s", tmp.name);
 824         if (!t || IS_ERR(t)) {
 825                 ret = t ? PTR_ERR(t) : -ENOENT;
 826                 goto free_newinfo_counters_untrans;
 827         }
 828
 829         /* You lied! */
 830         if (tmp.valid_hooks != t->valid_hooks) {
 831                 duprintf("Valid hook crap: %08X vs %08X\n",
 832                          tmp.valid_hooks, t->valid_hooks);
 833                 ret = -EINVAL;
 834                 goto put_module;
 835         }
 836
 837         oldinfo = xt_replace_table(t, tmp.num_counters, newinfo, &ret);
 838         if (!oldinfo)
 839                 goto put_module;
 840
 841         /* Update module usage count based on number of rules */
 842         duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
 843                 oldinfo->number, oldinfo->initial_entries, newinfo->number);
 844         if ((oldinfo->number > oldinfo->initial_entries) ||
 845             (newinfo->number <= oldinfo->initial_entries))
 846                 module_put(t->me);
 847         if ((oldinfo->number > oldinfo->initial_entries) &&
 848             (newinfo->number <= oldinfo->initial_entries))
 849                 module_put(t->me);
 850
 851         /* Get the old counters. */
 852         get_counters(oldinfo, counters);
 853         /* Decrease module usage counts and free resource */
 854         loc_cpu_old_entry = oldinfo->entries[raw_smp_processor_id()];
 855         ARPT_ENTRY_ITERATE(loc_cpu_old_entry, oldinfo->size, cleanup_entry,NULL);
 856
 857         xt_free_table_info(oldinfo);
 858         if (copy_to_user(tmp.counters, counters,
 859                          sizeof(struct xt_counters) * tmp.num_counters) != 0)
 860                 ret = -EFAULT;
 861         vfree(counters);
 862         xt_table_unlock(t);
 863         return ret;
 864
 865  put_module:
 866         module_put(t->me);
 867         xt_table_unlock(t);
 868  free_newinfo_counters_untrans:
 869         ARPT_ENTRY_ITERATE(loc_cpu_entry, newinfo->size, cleanup_entry, NULL);
 870  free_newinfo_counters:
 871         vfree(counters);
 872  free_newinfo:
 873         xt_free_table_info(newinfo);
 874         return ret;
 875 }
 876
 877 /* We're lazy, and add to the first CPU; overflow works its fey magic
 878  * and everything is OK.
 879  */
 880 static inline int add_counter_to_entry(struct arpt_entry *e,
 881                                        const struct xt_counters addme[],
 882                                        unsigned int *i)
 883 {
 884
 885         ADD_COUNTER(e->counters, addme[*i].bcnt, addme[*i].pcnt);
 886
 887         (*i)++;
 888         return 0;
 889 }
 890
 891 static int do_add_counters(void __user *user, unsigned int len)
 892 {
 893         unsigned int i;
 894         struct xt_counters_info tmp, *paddc;
 895         struct arpt_table *t;
 896         struct xt_table_info *private;
 897         int ret = 0;
 898         void *loc_cpu_entry;
 899
 900         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
 901                 return -EFAULT;
 902
 903         if (len != sizeof(tmp) + tmp.num_counters*sizeof(struct xt_counters))
 904                 return -EINVAL;
 905
 906         paddc = vmalloc_node(len, numa_node_id());
 907         if (!paddc)
 908                 return -ENOMEM;
 909
 910         if (copy_from_user(paddc, user, len) != 0) {
 911                 ret = -EFAULT;
 912                 goto free;
 913         }
 914
 915         t = xt_find_table_lock(NF_ARP, tmp.name);
 916         if (!t || IS_ERR(t)) {
 917                 ret = t ? PTR_ERR(t) : -ENOENT;
 918                 goto free;
 919         }
 920
 921         write_lock_bh(&t->lock);
 922         private = t->private;
 923         if (private->number != tmp.num_counters) {
 924                 ret = -EINVAL;
 925                 goto unlock_up_free;
 926         }
 927
 928         i = 0;
 929         /* Choose the copy that is on our node */
 930         loc_cpu_entry = private->entries[smp_processor_id()];
 931         ARPT_ENTRY_ITERATE(loc_cpu_entry,
 932                            private->size,
 933                            add_counter_to_entry,
 934                            paddc->counters,
 935                            &i);
 936  unlock_up_free:
 937         write_unlock_bh(&t->lock);
 938         xt_table_unlock(t);
 939         module_put(t->me);
 940  free:
 941         vfree(paddc);
 942
 943         return ret;
 944 }
 945
 946 static int do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
 947 {
 948         int ret;
 949
 950         if (!capable(CAP_NET_ADMIN))
 951                 return -EPERM;
 952
 953         switch (cmd) {
 954         case ARPT_SO_SET_REPLACE:
 955                 ret = do_replace(user, len);
 956                 break;
 957
 958         case ARPT_SO_SET_ADD_COUNTERS:
 959                 ret = do_add_counters(user, len);
 960                 break;
 961
 962         default:
 963                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
 964                 ret = -EINVAL;
 965         }
 966
 967         return ret;
 968 }
 969
 970 static int do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
 971 {
 972         int ret;
 973
 974         if (!capable(CAP_NET_ADMIN))
 975                 return -EPERM;
 976
 977         switch (cmd) {
 978         case ARPT_SO_GET_INFO: {
 979                 char name[ARPT_TABLE_MAXNAMELEN];
 980                 struct arpt_table *t;
 981
 982                 if (*len != sizeof(struct arpt_getinfo)) {
 983                         duprintf("length %u != %Zu\n", *len,
 984                                  sizeof(struct arpt_getinfo));
 985                         ret = -EINVAL;
 986                         break;
 987                 }
 988
 989                 if (copy_from_user(name, user, sizeof(name)) != 0) {
 990                         ret = -EFAULT;
 991                         break;
 992                 }
 993                 name[ARPT_TABLE_MAXNAMELEN-1] = '\0';
 994
 995                 t = try_then_request_module(xt_find_table_lock(NF_ARP, name),
 996                                             "arptable_%s", name);
 997                 if (t && !IS_ERR(t)) {
 998                         struct arpt_getinfo info;
 999                         struct xt_table_info *private = t->private;
1000
1001                         info.valid_hooks = t->valid_hooks;
1002                         memcpy(info.hook_entry, private->hook_entry,
1003                                sizeof(info.hook_entry));
1004                         memcpy(info.underflow, private->underflow,
1005                                sizeof(info.underflow));
1006                         info.num_entries = private->number;
1007                         info.size = private->size;
1008                         strcpy(info.name, name);
1009
1010                         if (copy_to_user(user, &info, *len) != 0)
1011                                 ret = -EFAULT;
1012                         else
1013                                 ret = 0;
1014                         xt_table_unlock(t);
1015                         module_put(t->me);
1016                 } else
1017                         ret = t ? PTR_ERR(t) : -ENOENT;
1018         }
1019         break;
1020
1021         case ARPT_SO_GET_ENTRIES: {
1022                 struct arpt_get_entries get;
1023
1024                 if (*len < sizeof(get)) {
1025                         duprintf("get_entries: %u < %Zu\n", *len, sizeof(get));
1026                         ret = -EINVAL;
1027                 } else if (copy_from_user(&get, user, sizeof(get)) != 0) {
1028                         ret = -EFAULT;
1029                 } else if (*len != sizeof(struct arpt_get_entries) + get.size) {
1030                         duprintf("get_entries: %u != %Zu\n", *len,
1031                                  sizeof(struct arpt_get_entries) + get.size);
1032                         ret = -EINVAL;
1033                 } else
1034                         ret = get_entries(&get, user);
1035                 break;
1036         }
1037
1038         case ARPT_SO_GET_REVISION_TARGET: {
1039                 struct xt_get_revision rev;
1040
1041                 if (*len != sizeof(rev)) {
1042                         ret = -EINVAL;
1043                         break;
1044                 }
1045                 if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
1046                         ret = -EFAULT;
1047                         break;
1048                 }
1049
1050                 try_then_request_module(xt_find_revision(NF_ARP, rev.name,
1051                                                          rev.revision, 1, &ret),
1052                                         "arpt_%s", rev.name);
1053                 break;
1054         }
1055
1056         default:
1057                 duprintf("do_arpt_get_ctl: unknown request %i\n", cmd);
1058                 ret = -EINVAL;
1059         }
1060
1061         return ret;
1062 }
1063
1064 int arpt_register_table(struct arpt_table *table,
1065                         const struct arpt_replace *repl)
1066 {
1067         int ret;
1068         struct xt_table_info *newinfo;
1069         struct xt_table_info bootstrap
1070                 = { 0, 0, 0, { 0 }, { 0 }, { } };
1071         void *loc_cpu_entry;
1072
1073         newinfo = xt_alloc_table_info(repl->size);
1074         if (!newinfo) {
1075                 ret = -ENOMEM;
1076                 return ret;
1077         }
1078
1079         /* choose the copy on our node/cpu */
1080         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1081         memcpy(loc_cpu_entry, repl->entries, repl->size);
1082
1083         ret = translate_table(table->name, table->valid_hooks,
1084                               newinfo, loc_cpu_entry, repl->size,
1085                               repl->num_entries,
1086                               repl->hook_entry,
1087                               repl->underflow);
1088
1089         duprintf("arpt_register_table: translate table gives %d\n", ret);
1090         if (ret != 0) {
1091                 xt_free_table_info(newinfo);
1092                 return ret;
1093         }
1094
1095         ret = xt_register_table(table, &bootstrap, newinfo);
1096         if (ret != 0) {
1097                 xt_free_table_info(newinfo);
1098                 return ret;
1099         }
1100
1101         return 0;
1102 }
1103
1104 void arpt_unregister_table(struct arpt_table *table)
1105 {
1106         struct xt_table_info *private;
1107         void *loc_cpu_entry;
1108
1109         private = xt_unregister_table(table);
1110
1111         /* Decrease module usage counts and free resources */
1112         loc_cpu_entry = private->entries[raw_smp_processor_id()];
1113         ARPT_ENTRY_ITERATE(loc_cpu_entry, private->size,
1114                            cleanup_entry, NULL);
1115         xt_free_table_info(private);
1116 }
1117
1118 /* The built-in targets: standard (NULL) and error. */
1119 static struct arpt_target arpt_standard_target __read_mostly = {
1120         .name           = ARPT_STANDARD_TARGET,
1121         .targetsize     = sizeof(int),
1122         .family         = NF_ARP,
1123 };
1124
1125 static struct arpt_target arpt_error_target __read_mostly = {
1126         .name           = ARPT_ERROR_TARGET,
1127         .target         = arpt_error,
1128         .targetsize     = ARPT_FUNCTION_MAXNAMELEN,
1129         .family         = NF_ARP,
1130 };
1131
1132 static struct nf_sockopt_ops arpt_sockopts = {
1133         .pf             = PF_INET,
1134         .set_optmin     = ARPT_BASE_CTL,
1135         .set_optmax     = ARPT_SO_SET_MAX+1,
1136         .set            = do_arpt_set_ctl,
1137         .get_optmin     = ARPT_BASE_CTL,
1138         .get_optmax     = ARPT_SO_GET_MAX+1,
1139         .get            = do_arpt_get_ctl,
1140         .owner          = THIS_MODULE,
1141 };
1142
1143 static int __init arp_tables_init(void)
1144 {
1145         int ret;
1146
1147         ret = xt_proto_init(NF_ARP);
1148         if (ret < 0)
1149                 goto err1;
1150
1151         /* Noone else will be downing sem now, so we won't sleep */
1152         ret = xt_register_target(&arpt_standard_target);
1153         if (ret < 0)
1154                 goto err2;
1155         ret = xt_register_target(&arpt_error_target);
1156         if (ret < 0)
1157                 goto err3;
1158
1159         /* Register setsockopt */
1160         ret = nf_register_sockopt(&arpt_sockopts);
1161         if (ret < 0)
1162                 goto err4;
1163
1164         printk(KERN_INFO "arp_tables: (C) 2002 David S. Miller\n");
1165         return 0;
1166
1167 err4:
1168         xt_unregister_target(&arpt_error_target);
1169 err3:
1170         xt_unregister_target(&arpt_standard_target);
1171 err2:
1172         xt_proto_fini(NF_ARP);
1173 err1:
1174         return ret;
1175 }
1176
1177 static void __exit arp_tables_fini(void)
1178 {
1179         nf_unregister_sockopt(&arpt_sockopts);
1180         xt_unregister_target(&arpt_error_target);
1181         xt_unregister_target(&arpt_standard_target);
1182         xt_proto_fini(NF_ARP);
1183 }
1184
1185 EXPORT_SYMBOL(arpt_register_table);
1186 EXPORT_SYMBOL(arpt_unregister_table);
1187 EXPORT_SYMBOL(arpt_do_table);
1188
1189 module_init(arp_tables_init);
1190 module_exit(arp_tables_fini);