net/ipv4/netfilter/arp_tables.c

   1 /*
   2  * Packet matching code for ARP packets.
   3  *
   4  * Based heavily, if not almost entirely, upon ip_tables.c framework.
   5  *
   6  * Some ARP specific bits are:
   7  *
   8  * Copyright (C) 2002 David S. Miller (davem@redhat.com)
   9  *
  10  */
  11
  12 #include <linux/config.h>
  13 #include <linux/kernel.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/capability.h>
  17 #include <linux/if_arp.h>
  18 #include <linux/kmod.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/module.h>
  22 #include <linux/init.h>
  23
  24 #include <asm/uaccess.h>
  25 #include <linux/mutex.h>
  26
  27 #include <linux/netfilter/x_tables.h>
  28 #include <linux/netfilter_arp/arp_tables.h>
  29
  30 MODULE_LICENSE("GPL");
  31 MODULE_AUTHOR("David S. Miller <davem@redhat.com>");
  32 MODULE_DESCRIPTION("arptables core");
  33
  34 /*#define DEBUG_ARP_TABLES*/
  35 /*#define DEBUG_ARP_TABLES_USER*/
  36
  37 #ifdef DEBUG_ARP_TABLES
  38 #define dprintf(format, args...)  printk(format , ## args)
  39 #else
  40 #define dprintf(format, args...)
  41 #endif
  42
  43 #ifdef DEBUG_ARP_TABLES_USER
  44 #define duprintf(format, args...) printk(format , ## args)
  45 #else
  46 #define duprintf(format, args...)
  47 #endif
  48
  49 #ifdef CONFIG_NETFILTER_DEBUG
  50 #define ARP_NF_ASSERT(x)                                        \
  51 do {                                                            \
  52         if (!(x))                                               \
  53                 printk("ARP_NF_ASSERT: %s:%s:%u\n",             \
  54                        __FUNCTION__, __FILE__, __LINE__);       \
  55 } while(0)
  56 #else
  57 #define ARP_NF_ASSERT(x)
  58 #endif
  59
  60 #include <linux/netfilter_ipv4/listhelp.h>
  61
  62 static inline int arp_devaddr_compare(const struct arpt_devaddr_info *ap,
  63                                       char *hdr_addr, int len)
  64 {
  65         int i, ret;
  66
  67         if (len > ARPT_DEV_ADDR_LEN_MAX)
  68                 len = ARPT_DEV_ADDR_LEN_MAX;
  69
  70         ret = 0;
  71         for (i = 0; i < len; i++)
  72                 ret |= (hdr_addr[i] ^ ap->addr[i]) & ap->mask[i];
  73
  74         return (ret != 0);
  75 }
  76
  77 /* Returns whether packet matches rule or not. */
  78 static inline int arp_packet_match(const struct arphdr *arphdr,
  79                                    struct net_device *dev,
  80                                    const char *indev,
  81                                    const char *outdev,
  82                                    const struct arpt_arp *arpinfo)
  83 {
  84         char *arpptr = (char *)(arphdr + 1);
  85         char *src_devaddr, *tgt_devaddr;
  86         u32 src_ipaddr, tgt_ipaddr;
  87         int i, ret;
  88
  89 #define FWINV(bool,invflg) ((bool) ^ !!(arpinfo->invflags & invflg))
  90
  91         if (FWINV((arphdr->ar_op & arpinfo->arpop_mask) != arpinfo->arpop,
  92                   ARPT_INV_ARPOP)) {
  93                 dprintf("ARP operation field mismatch.\n");
  94                 dprintf("ar_op: %04x info->arpop: %04x info->arpop_mask: %04x\n",
  95                         arphdr->ar_op, arpinfo->arpop, arpinfo->arpop_mask);
  96                 return 0;
  97         }
  98
  99         if (FWINV((arphdr->ar_hrd & arpinfo->arhrd_mask) != arpinfo->arhrd,
 100                   ARPT_INV_ARPHRD)) {
 101                 dprintf("ARP hardware address format mismatch.\n");
 102                 dprintf("ar_hrd: %04x info->arhrd: %04x info->arhrd_mask: %04x\n",
 103                         arphdr->ar_hrd, arpinfo->arhrd, arpinfo->arhrd_mask);
 104                 return 0;
 105         }
 106
 107         if (FWINV((arphdr->ar_pro & arpinfo->arpro_mask) != arpinfo->arpro,
 108                   ARPT_INV_ARPPRO)) {
 109                 dprintf("ARP protocol address format mismatch.\n");
 110                 dprintf("ar_pro: %04x info->arpro: %04x info->arpro_mask: %04x\n",
 111                         arphdr->ar_pro, arpinfo->arpro, arpinfo->arpro_mask);
 112                 return 0;
 113         }
 114
 115         if (FWINV((arphdr->ar_hln & arpinfo->arhln_mask) != arpinfo->arhln,
 116                   ARPT_INV_ARPHLN)) {
 117                 dprintf("ARP hardware address length mismatch.\n");
 118                 dprintf("ar_hln: %02x info->arhln: %02x info->arhln_mask: %02x\n",
 119                         arphdr->ar_hln, arpinfo->arhln, arpinfo->arhln_mask);
 120                 return 0;
 121         }
 122
 123         src_devaddr = arpptr;
 124         arpptr += dev->addr_len;
 125         memcpy(&src_ipaddr, arpptr, sizeof(u32));
 126         arpptr += sizeof(u32);
 127         tgt_devaddr = arpptr;
 128         arpptr += dev->addr_len;
 129         memcpy(&tgt_ipaddr, arpptr, sizeof(u32));
 130
 131         if (FWINV(arp_devaddr_compare(&arpinfo->src_devaddr, src_devaddr, dev->addr_len),
 132                   ARPT_INV_SRCDEVADDR) ||
 133             FWINV(arp_devaddr_compare(&arpinfo->tgt_devaddr, tgt_devaddr, dev->addr_len),
 134                   ARPT_INV_TGTDEVADDR)) {
 135                 dprintf("Source or target device address mismatch.\n");
 136
 137                 return 0;
 138         }
 139
 140         if (FWINV((src_ipaddr & arpinfo->smsk.s_addr) != arpinfo->src.s_addr,
 141                   ARPT_INV_SRCIP) ||
 142             FWINV(((tgt_ipaddr & arpinfo->tmsk.s_addr) != arpinfo->tgt.s_addr),
 143                   ARPT_INV_TGTIP)) {
 144                 dprintf("Source or target IP address mismatch.\n");
 145
 146                 dprintf("SRC: %u.%u.%u.%u. Mask: %u.%u.%u.%u. Target: %u.%u.%u.%u.%s\n",
 147                         NIPQUAD(src_ipaddr),
 148                         NIPQUAD(arpinfo->smsk.s_addr),
 149                         NIPQUAD(arpinfo->src.s_addr),
 150                         arpinfo->invflags & ARPT_INV_SRCIP ? " (INV)" : "");
 151                 dprintf("TGT: %u.%u.%u.%u Mask: %u.%u.%u.%u Target: %u.%u.%u.%u.%s\n",
 152                         NIPQUAD(tgt_ipaddr),
 153                         NIPQUAD(arpinfo->tmsk.s_addr),
 154                         NIPQUAD(arpinfo->tgt.s_addr),
 155                         arpinfo->invflags & ARPT_INV_TGTIP ? " (INV)" : "");
 156                 return 0;
 157         }
 158
 159         /* Look for ifname matches.  */
 160         for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
 161                 ret |= (indev[i] ^ arpinfo->iniface[i])
 162                         & arpinfo->iniface_mask[i];
 163         }
 164
 165         if (FWINV(ret != 0, ARPT_INV_VIA_IN)) {
 166                 dprintf("VIA in mismatch (%s vs %s).%s\n",
 167                         indev, arpinfo->iniface,
 168                         arpinfo->invflags&ARPT_INV_VIA_IN ?" (INV)":"");
 169                 return 0;
 170         }
 171
 172         for (i = 0, ret = 0; i < IFNAMSIZ/sizeof(unsigned long); i++) {
 173                 unsigned long odev;
 174                 memcpy(&odev, outdev + i*sizeof(unsigned long),
 175                        sizeof(unsigned long));
 176                 ret |= (odev
 177                         ^ ((const unsigned long *)arpinfo->outiface)[i])
 178                         & ((const unsigned long *)arpinfo->outiface_mask)[i];
 179         }
 180
 181         if (FWINV(ret != 0, ARPT_INV_VIA_OUT)) {
 182                 dprintf("VIA out mismatch (%s vs %s).%s\n",
 183                         outdev, arpinfo->outiface,
 184                         arpinfo->invflags&ARPT_INV_VIA_OUT ?" (INV)":"");
 185                 return 0;
 186         }
 187
 188         return 1;
 189 }
 190
 191 static inline int arp_checkentry(const struct arpt_arp *arp)
 192 {
 193         if (arp->flags & ~ARPT_F_MASK) {
 194                 duprintf("Unknown flag bits set: %08X\n",
 195                          arp->flags & ~ARPT_F_MASK);
 196                 return 0;
 197         }
 198         if (arp->invflags & ~ARPT_INV_MASK) {
 199                 duprintf("Unknown invflag bits set: %08X\n",
 200                          arp->invflags & ~ARPT_INV_MASK);
 201                 return 0;
 202         }
 203
 204         return 1;
 205 }
 206
 207 static unsigned int arpt_error(struct sk_buff **pskb,
 208                                const struct net_device *in,
 209                                const struct net_device *out,
 210                                unsigned int hooknum,
 211                                const struct xt_target *target,
 212                                const void *targinfo,
 213                                void *userinfo)
 214 {
 215         if (net_ratelimit())
 216                 printk("arp_tables: error: '%s'\n", (char *)targinfo);
 217
 218         return NF_DROP;
 219 }
 220
 221 static inline struct arpt_entry *get_entry(void *base, unsigned int offset)
 222 {
 223         return (struct arpt_entry *)(base + offset);
 224 }
 225
 226 unsigned int arpt_do_table(struct sk_buff **pskb,
 227                            unsigned int hook,
 228                            const struct net_device *in,
 229                            const struct net_device *out,
 230                            struct arpt_table *table,
 231                            void *userdata)
 232 {
 233         static const char nulldevname[IFNAMSIZ];
 234         unsigned int verdict = NF_DROP;
 235         struct arphdr *arp;
 236         int hotdrop = 0;
 237         struct arpt_entry *e, *back;
 238         const char *indev, *outdev;
 239         void *table_base;
 240         struct xt_table_info *private;
 241
 242         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
 243         if (!pskb_may_pull((*pskb), (sizeof(struct arphdr) +
 244                                      (2 * (*pskb)->dev->addr_len) +
 245                                      (2 * sizeof(u32)))))
 246                 return NF_DROP;
 247
 248         indev = in ? in->name : nulldevname;
 249         outdev = out ? out->name : nulldevname;
 250
 251         read_lock_bh(&table->lock);
 252         private = table->private;
 253         table_base = (void *)private->entries[smp_processor_id()];
 254         e = get_entry(table_base, private->hook_entry[hook]);
 255         back = get_entry(table_base, private->underflow[hook]);
 256
 257         arp = (*pskb)->nh.arph;
 258         do {
 259                 if (arp_packet_match(arp, (*pskb)->dev, indev, outdev, &e->arp)) {
 260                         struct arpt_entry_target *t;
 261                         int hdr_len;
 262
 263                         hdr_len = sizeof(*arp) + (2 * sizeof(struct in_addr)) +
 264                                 (2 * (*pskb)->dev->addr_len);
 265                         ADD_COUNTER(e->counters, hdr_len, 1);
 266
 267                         t = arpt_get_target(e);
 268
 269                         /* Standard target? */
 270                         if (!t->u.kernel.target->target) {
 271                                 int v;
 272
 273                                 v = ((struct arpt_standard_target *)t)->verdict;
 274                                 if (v < 0) {
 275                                         /* Pop from stack? */
 276                                         if (v != ARPT_RETURN) {
 277                                                 verdict = (unsigned)(-v) - 1;
 278                                                 break;
 279                                         }
 280                                         e = back;
 281                                         back = get_entry(table_base,
 282                                                          back->comefrom);
 283                                         continue;
 284                                 }
 285                                 if (table_base + v
 286                                     != (void *)e + e->next_offset) {
 287                                         /* Save old back ptr in next entry */
 288                                         struct arpt_entry *next
 289                                                 = (void *)e + e->next_offset;
 290                                         next->comefrom =
 291                                                 (void *)back - table_base;
 292
 293                                         /* set back pointer to next entry */
 294                                         back = next;
 295                                 }
 296
 297                                 e = get_entry(table_base, v);
 298                         } else {
 299                                 /* Targets which reenter must return
 300                                  * abs. verdicts
 301                                  */
 302                                 verdict = t->u.kernel.target->target(pskb,
 303                                                                      in, out,
 304                                                                      hook,
 305                                                                      t->u.kernel.target,
 306                                                                      t->data,
 307                                                                      userdata);
 308
 309                                 /* Target might have changed stuff. */
 310                                 arp = (*pskb)->nh.arph;
 311
 312                                 if (verdict == ARPT_CONTINUE)
 313                                         e = (void *)e + e->next_offset;
 314                                 else
 315                                         /* Verdict */
 316                                         break;
 317                         }
 318                 } else {
 319                         e = (void *)e + e->next_offset;
 320                 }
 321         } while (!hotdrop);
 322         read_unlock_bh(&table->lock);
 323
 324         if (hotdrop)
 325                 return NF_DROP;
 326         else
 327                 return verdict;
 328 }
 329
 330 /* All zeroes == unconditional rule. */
 331 static inline int unconditional(const struct arpt_arp *arp)
 332 {
 333         unsigned int i;
 334
 335         for (i = 0; i < sizeof(*arp)/sizeof(__u32); i++)
 336                 if (((__u32 *)arp)[i])
 337                         return 0;
 338
 339         return 1;
 340 }
 341
 342 /* Figures out from what hook each rule can be called: returns 0 if
 343  * there are loops.  Puts hook bitmask in comefrom.
 344  */
 345 static int mark_source_chains(struct xt_table_info *newinfo,
 346                               unsigned int valid_hooks, void *entry0)
 347 {
 348         unsigned int hook;
 349
 350         /* No recursion; use packet counter to save back ptrs (reset
 351          * to 0 as we leave), and comefrom to save source hook bitmask.
 352          */
 353         for (hook = 0; hook < NF_ARP_NUMHOOKS; hook++) {
 354                 unsigned int pos = newinfo->hook_entry[hook];
 355                 struct arpt_entry *e
 356                         = (struct arpt_entry *)(entry0 + pos);
 357
 358                 if (!(valid_hooks & (1 << hook)))
 359                         continue;
 360
 361                 /* Set initial back pointer. */
 362                 e->counters.pcnt = pos;
 363
 364                 for (;;) {
 365                         struct arpt_standard_target *t
 366                                 = (void *)arpt_get_target(e);
 367
 368                         if (e->comefrom & (1 << NF_ARP_NUMHOOKS)) {
 369                                 printk("arptables: loop hook %u pos %u %08X.\n",
 370                                        hook, pos, e->comefrom);
 371                                 return 0;
 372                         }
 373                         e->comefrom
 374                                 |= ((1 << hook) | (1 << NF_ARP_NUMHOOKS));
 375
 376                         /* Unconditional return/END. */
 377                         if (e->target_offset == sizeof(struct arpt_entry)
 378                             && (strcmp(t->target.u.user.name,
 379                                        ARPT_STANDARD_TARGET) == 0)
 380                             && t->verdict < 0
 381                             && unconditional(&e->arp)) {
 382                                 unsigned int oldpos, size;
 383
 384                                 /* Return: backtrack through the last
 385                                  * big jump.
 386                                  */
 387                                 do {
 388                                         e->comefrom ^= (1<<NF_ARP_NUMHOOKS);
 389                                         oldpos = pos;
 390                                         pos = e->counters.pcnt;
 391                                         e->counters.pcnt = 0;
 392
 393                                         /* We're at the start. */
 394                                         if (pos == oldpos)
 395                                                 goto next;
 396
 397                                         e = (struct arpt_entry *)
 398                                                 (entry0 + pos);
 399                                 } while (oldpos == pos + e->next_offset);
 400
 401                                 /* Move along one */
 402                                 size = e->next_offset;
 403                                 e = (struct arpt_entry *)
 404                                         (entry0 + pos + size);
 405                                 e->counters.pcnt = pos;
 406                                 pos += size;
 407                         } else {
 408                                 int newpos = t->verdict;
 409
 410                                 if (strcmp(t->target.u.user.name,
 411                                            ARPT_STANDARD_TARGET) == 0
 412                                     && newpos >= 0) {
 413                                         /* This a jump; chase it. */
 414                                         duprintf("Jump rule %u -> %u\n",
 415                                                  pos, newpos);
 416                                 } else {
 417                                         /* ... this is a fallthru */
 418                                         newpos = pos + e->next_offset;
 419                                 }
 420                                 e = (struct arpt_entry *)
 421                                         (entry0 + newpos);
 422                                 e->counters.pcnt = pos;
 423                                 pos = newpos;
 424                         }
 425                 }
 426                 next:
 427                 duprintf("Finished chain %u\n", hook);
 428         }
 429         return 1;
 430 }
 431
 432 static inline int standard_check(const struct arpt_entry_target *t,
 433                                  unsigned int max_offset)
 434 {
 435         struct arpt_standard_target *targ = (void *)t;
 436
 437         /* Check standard info. */
 438         if (t->u.target_size
 439             != ARPT_ALIGN(sizeof(struct arpt_standard_target))) {
 440                 duprintf("arpt_standard_check: target size %u != %Zu\n",
 441                          t->u.target_size,
 442                          ARPT_ALIGN(sizeof(struct arpt_standard_target)));
 443                 return 0;
 444         }
 445
 446         if (targ->verdict >= 0
 447             && targ->verdict > max_offset - sizeof(struct arpt_entry)) {
 448                 duprintf("arpt_standard_check: bad verdict (%i)\n",
 449                          targ->verdict);
 450                 return 0;
 451         }
 452
 453         if (targ->verdict < -NF_MAX_VERDICT - 1) {
 454                 duprintf("arpt_standard_check: bad negative verdict (%i)\n",
 455                          targ->verdict);
 456                 return 0;
 457         }
 458         return 1;
 459 }
 460
 461 static struct arpt_target arpt_standard_target;
 462
 463 static inline int check_entry(struct arpt_entry *e, const char *name, unsigned int size,
 464                               unsigned int *i)
 465 {
 466         struct arpt_entry_target *t;
 467         struct arpt_target *target;
 468         int ret;
 469
 470         if (!arp_checkentry(&e->arp)) {
 471                 duprintf("arp_tables: arp check failed %p %s.\n", e, name);
 472                 return -EINVAL;
 473         }
 474
 475         t = arpt_get_target(e);
 476         target = try_then_request_module(xt_find_target(NF_ARP, t->u.user.name,
 477                                                         t->u.user.revision),
 478                                          "arpt_%s", t->u.user.name);
 479         if (IS_ERR(target) || !target) {
 480                 duprintf("check_entry: `%s' not found\n", t->u.user.name);
 481                 ret = target ? PTR_ERR(target) : -ENOENT;
 482                 goto out;
 483         }
 484         t->u.kernel.target = target;
 485
 486         ret = xt_check_target(target, NF_ARP, t->u.target_size - sizeof(*t),
 487                               name, e->comefrom, 0, 0);
 488         if (ret)
 489                 goto err;
 490
 491         if (t->u.kernel.target == &arpt_standard_target) {
 492                 if (!standard_check(t, size)) {
 493                         ret = -EINVAL;
 494                         goto out;
 495                 }
 496         } else if (t->u.kernel.target->checkentry
 497                    && !t->u.kernel.target->checkentry(name, e, target, t->data,
 498                                                       t->u.target_size
 499                                                       - sizeof(*t),
 500                                                       e->comefrom)) {
 501                 duprintf("arp_tables: check failed for `%s'.\n",
 502                          t->u.kernel.target->name);
 503                 ret = -EINVAL;
 504                 goto err;
 505         }
 506
 507         (*i)++;
 508         return 0;
 509 err:
 510         module_put(t->u.kernel.target->me);
 511 out:
 512         return ret;
 513 }
 514
 515 static inline int check_entry_size_and_hooks(struct arpt_entry *e,
 516                                              struct xt_table_info *newinfo,
 517                                              unsigned char *base,
 518                                              unsigned char *limit,
 519                                              const unsigned int *hook_entries,
 520                                              const unsigned int *underflows,
 521                                              unsigned int *i)
 522 {
 523         unsigned int h;
 524
 525         if ((unsigned long)e % __alignof__(struct arpt_entry) != 0
 526             || (unsigned char *)e + sizeof(struct arpt_entry) >= limit) {
 527                 duprintf("Bad offset %p\n", e);
 528                 return -EINVAL;
 529         }
 530
 531         if (e->next_offset
 532             < sizeof(struct arpt_entry) + sizeof(struct arpt_entry_target)) {
 533                 duprintf("checking: element %p size %u\n",
 534                          e, e->next_offset);
 535                 return -EINVAL;
 536         }
 537
 538         /* Check hooks & underflows */
 539         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
 540                 if ((unsigned char *)e - base == hook_entries[h])
 541                         newinfo->hook_entry[h] = hook_entries[h];
 542                 if ((unsigned char *)e - base == underflows[h])
 543                         newinfo->underflow[h] = underflows[h];
 544         }
 545
 546         /* FIXME: underflows must be unconditional, standard verdicts
 547            < 0 (not ARPT_RETURN). --RR */
 548
 549         /* Clear counters and comefrom */
 550         e->counters = ((struct xt_counters) { 0, 0 });
 551         e->comefrom = 0;
 552
 553         (*i)++;
 554         return 0;
 555 }
 556
 557 static inline int cleanup_entry(struct arpt_entry *e, unsigned int *i)
 558 {
 559         struct arpt_entry_target *t;
 560
 561         if (i && (*i)-- == 0)
 562                 return 1;
 563
 564         t = arpt_get_target(e);
 565         if (t->u.kernel.target->destroy)
 566                 t->u.kernel.target->destroy(t->u.kernel.target, t->data,
 567                                             t->u.target_size - sizeof(*t));
 568         module_put(t->u.kernel.target->me);
 569         return 0;
 570 }
 571
 572 /* Checks and translates the user-supplied table segment (held in
 573  * newinfo).
 574  */
 575 static int translate_table(const char *name,
 576                            unsigned int valid_hooks,
 577                            struct xt_table_info *newinfo,
 578                            void *entry0,
 579                            unsigned int size,
 580                            unsigned int number,
 581                            const unsigned int *hook_entries,
 582                            const unsigned int *underflows)
 583 {
 584         unsigned int i;
 585         int ret;
 586
 587         newinfo->size = size;
 588         newinfo->number = number;
 589
 590         /* Init all hooks to impossible value. */
 591         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
 592                 newinfo->hook_entry[i] = 0xFFFFFFFF;
 593                 newinfo->underflow[i] = 0xFFFFFFFF;
 594         }
 595
 596         duprintf("translate_table: size %u\n", newinfo->size);
 597         i = 0;
 598
 599         /* Walk through entries, checking offsets. */
 600         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 601                                  check_entry_size_and_hooks,
 602                                  newinfo,
 603                                  entry0,
 604                                  entry0 + size,
 605                                  hook_entries, underflows, &i);
 606         duprintf("translate_table: ARPT_ENTRY_ITERATE gives %d\n", ret);
 607         if (ret != 0)
 608                 return ret;
 609
 610         if (i != number) {
 611                 duprintf("translate_table: %u not %u entries\n",
 612                          i, number);
 613                 return -EINVAL;
 614         }
 615
 616         /* Check hooks all assigned */
 617         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
 618                 /* Only hooks which are valid */
 619                 if (!(valid_hooks & (1 << i)))
 620                         continue;
 621                 if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
 622                         duprintf("Invalid hook entry %u %u\n",
 623                                  i, hook_entries[i]);
 624                         return -EINVAL;
 625                 }
 626                 if (newinfo->underflow[i] == 0xFFFFFFFF) {
 627                         duprintf("Invalid underflow %u %u\n",
 628                                  i, underflows[i]);
 629                         return -EINVAL;
 630                 }
 631         }
 632
 633         if (!mark_source_chains(newinfo, valid_hooks, entry0)) {
 634                 duprintf("Looping hook\n");
 635                 return -ELOOP;
 636         }
 637
 638         /* Finally, each sanity check must pass */
 639         i = 0;
 640         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 641                                  check_entry, name, size, &i);
 642
 643         if (ret != 0) {
 644                 ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 645                                    cleanup_entry, &i);
 646                 return ret;
 647         }
 648
 649         /* And one copy for every other CPU */
 650         for_each_possible_cpu(i) {
 651                 if (newinfo->entries[i] && newinfo->entries[i] != entry0)
 652                         memcpy(newinfo->entries[i], entry0, newinfo->size);
 653         }
 654
 655         return ret;
 656 }
 657
 658 /* Gets counters. */
 659 static inline int add_entry_to_counter(const struct arpt_entry *e,
 660                                        struct xt_counters total[],
 661                                        unsigned int *i)
 662 {
 663         ADD_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
 664
 665         (*i)++;
 666         return 0;
 667 }
 668
 669 static inline int set_entry_to_counter(const struct arpt_entry *e,
 670                                        struct xt_counters total[],
 671                                        unsigned int *i)
 672 {
 673         SET_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
 674
 675         (*i)++;
 676         return 0;
 677 }
 678
 679 static void get_counters(const struct xt_table_info *t,
 680                          struct xt_counters counters[])
 681 {
 682         unsigned int cpu;
 683         unsigned int i;
 684         unsigned int curcpu;
 685
 686         /* Instead of clearing (by a previous call to memset())
 687          * the counters and using adds, we set the counters
 688          * with data used by 'current' CPU
 689          * We dont care about preemption here.
 690          */
 691         curcpu = raw_smp_processor_id();
 692
 693         i = 0;
 694         ARPT_ENTRY_ITERATE(t->entries[curcpu],
 695                            t->size,
 696                            set_entry_to_counter,
 697                            counters,
 698                            &i);
 699
 700         for_each_possible_cpu(cpu) {
 701                 if (cpu == curcpu)
 702                         continue;
 703                 i = 0;
 704                 ARPT_ENTRY_ITERATE(t->entries[cpu],
 705                                    t->size,
 706                                    add_entry_to_counter,
 707                                    counters,
 708                                    &i);
 709         }
 710 }
 711
 712 static int copy_entries_to_user(unsigned int total_size,
 713                                 struct arpt_table *table,
 714                                 void __user *userptr)
 715 {
 716         unsigned int off, num, countersize;
 717         struct arpt_entry *e;
 718         struct xt_counters *counters;
 719         struct xt_table_info *private = table->private;
 720         int ret = 0;
 721         void *loc_cpu_entry;
 722
 723         /* We need atomic snapshot of counters: rest doesn't change
 724          * (other than comefrom, which userspace doesn't care
 725          * about).
 726          */
 727         countersize = sizeof(struct xt_counters) * private->number;
 728         counters = vmalloc_node(countersize, numa_node_id());
 729
 730         if (counters == NULL)
 731                 return -ENOMEM;
 732
 733         /* First, sum counters... */
 734         write_lock_bh(&table->lock);
 735         get_counters(private, counters);
 736         write_unlock_bh(&table->lock);
 737
 738         loc_cpu_entry = private->entries[raw_smp_processor_id()];
 739         /* ... then copy entire thing ... */
 740         if (copy_to_user(userptr, loc_cpu_entry, total_size) != 0) {
 741                 ret = -EFAULT;
 742                 goto free_counters;
 743         }
 744
 745         /* FIXME: use iterator macros --RR */
 746         /* ... then go back and fix counters and names */
 747         for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
 748                 struct arpt_entry_target *t;
 749
 750                 e = (struct arpt_entry *)(loc_cpu_entry + off);
 751                 if (copy_to_user(userptr + off
 752                                  + offsetof(struct arpt_entry, counters),
 753                                  &counters[num],
 754                                  sizeof(counters[num])) != 0) {
 755                         ret = -EFAULT;
 756                         goto free_counters;
 757                 }
 758
 759                 t = arpt_get_target(e);
 760                 if (copy_to_user(userptr + off + e->target_offset
 761                                  + offsetof(struct arpt_entry_target,
 762                                             u.user.name),
 763                                  t->u.kernel.target->name,
 764                                  strlen(t->u.kernel.target->name)+1) != 0) {
 765                         ret = -EFAULT;
 766                         goto free_counters;
 767                 }
 768         }
 769
 770  free_counters:
 771         vfree(counters);
 772         return ret;
 773 }
 774
 775 static int get_entries(const struct arpt_get_entries *entries,
 776                        struct arpt_get_entries __user *uptr)
 777 {
 778         int ret;
 779         struct arpt_table *t;
 780
 781         t = xt_find_table_lock(NF_ARP, entries->name);
 782         if (t && !IS_ERR(t)) {
 783                 struct xt_table_info *private = t->private;
 784                 duprintf("t->private->number = %u\n",
 785                          private->number);
 786                 if (entries->size == private->size)
 787                         ret = copy_entries_to_user(private->size,
 788                                                    t, uptr->entrytable);
 789                 else {
 790                         duprintf("get_entries: I've got %u not %u!\n",
 791                                  private->size, entries->size);
 792                         ret = -EINVAL;
 793                 }
 794                 module_put(t->me);
 795                 xt_table_unlock(t);
 796         } else
 797                 ret = t ? PTR_ERR(t) : -ENOENT;
 798
 799         return ret;
 800 }
 801
 802 static int do_replace(void __user *user, unsigned int len)
 803 {
 804         int ret;
 805         struct arpt_replace tmp;
 806         struct arpt_table *t;
 807         struct xt_table_info *newinfo, *oldinfo;
 808         struct xt_counters *counters;
 809         void *loc_cpu_entry, *loc_cpu_old_entry;
 810
 811         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
 812                 return -EFAULT;
 813
 814         /* Hack: Causes ipchains to give correct error msg --RR */
 815         if (len != sizeof(tmp) + tmp.size)
 816                 return -ENOPROTOOPT;
 817
 818         /* overflow check */
 819         if (tmp.size >= (INT_MAX - sizeof(struct xt_table_info)) / NR_CPUS -
 820                         SMP_CACHE_BYTES)
 821                 return -ENOMEM;
 822         if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
 823                 return -ENOMEM;
 824
 825         newinfo = xt_alloc_table_info(tmp.size);
 826         if (!newinfo)
 827                 return -ENOMEM;
 828
 829         /* choose the copy that is on our node/cpu */
 830         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
 831         if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
 832                            tmp.size) != 0) {
 833                 ret = -EFAULT;
 834                 goto free_newinfo;
 835         }
 836
 837         counters = vmalloc(tmp.num_counters * sizeof(struct xt_counters));
 838         if (!counters) {
 839                 ret = -ENOMEM;
 840                 goto free_newinfo;
 841         }
 842
 843         ret = translate_table(tmp.name, tmp.valid_hooks,
 844                               newinfo, loc_cpu_entry, tmp.size, tmp.num_entries,
 845                               tmp.hook_entry, tmp.underflow);
 846         if (ret != 0)
 847                 goto free_newinfo_counters;
 848
 849         duprintf("arp_tables: Translated table\n");
 850
 851         t = try_then_request_module(xt_find_table_lock(NF_ARP, tmp.name),
 852                                     "arptable_%s", tmp.name);
 853         if (!t || IS_ERR(t)) {
 854                 ret = t ? PTR_ERR(t) : -ENOENT;
 855                 goto free_newinfo_counters_untrans;
 856         }
 857
 858         /* You lied! */
 859         if (tmp.valid_hooks != t->valid_hooks) {
 860                 duprintf("Valid hook crap: %08X vs %08X\n",
 861                          tmp.valid_hooks, t->valid_hooks);
 862                 ret = -EINVAL;
 863                 goto put_module;
 864         }
 865
 866         oldinfo = xt_replace_table(t, tmp.num_counters, newinfo, &ret);
 867         if (!oldinfo)
 868                 goto put_module;
 869
 870         /* Update module usage count based on number of rules */
 871         duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
 872                 oldinfo->number, oldinfo->initial_entries, newinfo->number);
 873         if ((oldinfo->number > oldinfo->initial_entries) ||
 874             (newinfo->number <= oldinfo->initial_entries))
 875                 module_put(t->me);
 876         if ((oldinfo->number > oldinfo->initial_entries) &&
 877             (newinfo->number <= oldinfo->initial_entries))
 878                 module_put(t->me);
 879
 880         /* Get the old counters. */
 881         get_counters(oldinfo, counters);
 882         /* Decrease module usage counts and free resource */
 883         loc_cpu_old_entry = oldinfo->entries[raw_smp_processor_id()];
 884         ARPT_ENTRY_ITERATE(loc_cpu_old_entry, oldinfo->size, cleanup_entry,NULL);
 885
 886         xt_free_table_info(oldinfo);
 887         if (copy_to_user(tmp.counters, counters,
 888                          sizeof(struct xt_counters) * tmp.num_counters) != 0)
 889                 ret = -EFAULT;
 890         vfree(counters);
 891         xt_table_unlock(t);
 892         return ret;
 893
 894  put_module:
 895         module_put(t->me);
 896         xt_table_unlock(t);
 897  free_newinfo_counters_untrans:
 898         ARPT_ENTRY_ITERATE(loc_cpu_entry, newinfo->size, cleanup_entry, NULL);
 899  free_newinfo_counters:
 900         vfree(counters);
 901  free_newinfo:
 902         xt_free_table_info(newinfo);
 903         return ret;
 904 }
 905
 906 /* We're lazy, and add to the first CPU; overflow works its fey magic
 907  * and everything is OK.
 908  */
 909 static inline int add_counter_to_entry(struct arpt_entry *e,
 910                                        const struct xt_counters addme[],
 911                                        unsigned int *i)
 912 {
 913
 914         ADD_COUNTER(e->counters, addme[*i].bcnt, addme[*i].pcnt);
 915
 916         (*i)++;
 917         return 0;
 918 }
 919
 920 static int do_add_counters(void __user *user, unsigned int len)
 921 {
 922         unsigned int i;
 923         struct xt_counters_info tmp, *paddc;
 924         struct arpt_table *t;
 925         struct xt_table_info *private;
 926         int ret = 0;
 927         void *loc_cpu_entry;
 928
 929         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
 930                 return -EFAULT;
 931
 932         if (len != sizeof(tmp) + tmp.num_counters*sizeof(struct xt_counters))
 933                 return -EINVAL;
 934
 935         paddc = vmalloc(len);
 936         if (!paddc)
 937                 return -ENOMEM;
 938
 939         if (copy_from_user(paddc, user, len) != 0) {
 940                 ret = -EFAULT;
 941                 goto free;
 942         }
 943
 944         t = xt_find_table_lock(NF_ARP, tmp.name);
 945         if (!t || IS_ERR(t)) {
 946                 ret = t ? PTR_ERR(t) : -ENOENT;
 947                 goto free;
 948         }
 949
 950         write_lock_bh(&t->lock);
 951         private = t->private;
 952         if (private->number != tmp.num_counters) {
 953                 ret = -EINVAL;
 954                 goto unlock_up_free;
 955         }
 956
 957         i = 0;
 958         /* Choose the copy that is on our node */
 959         loc_cpu_entry = private->entries[smp_processor_id()];
 960         ARPT_ENTRY_ITERATE(loc_cpu_entry,
 961                            private->size,
 962                            add_counter_to_entry,
 963                            paddc->counters,
 964                            &i);
 965  unlock_up_free:
 966         write_unlock_bh(&t->lock);
 967         xt_table_unlock(t);
 968         module_put(t->me);
 969  free:
 970         vfree(paddc);
 971
 972         return ret;
 973 }
 974
 975 static int do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
 976 {
 977         int ret;
 978
 979         if (!capable(CAP_NET_ADMIN))
 980                 return -EPERM;
 981
 982         switch (cmd) {
 983         case ARPT_SO_SET_REPLACE:
 984                 ret = do_replace(user, len);
 985                 break;
 986
 987         case ARPT_SO_SET_ADD_COUNTERS:
 988                 ret = do_add_counters(user, len);
 989                 break;
 990
 991         default:
 992                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
 993                 ret = -EINVAL;
 994         }
 995
 996         return ret;
 997 }
 998
 999 static int do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
1000 {
1001         int ret;
1002
1003         if (!capable(CAP_NET_ADMIN))
1004                 return -EPERM;
1005
1006         switch (cmd) {
1007         case ARPT_SO_GET_INFO: {
1008                 char name[ARPT_TABLE_MAXNAMELEN];
1009                 struct arpt_table *t;
1010
1011                 if (*len != sizeof(struct arpt_getinfo)) {
1012                         duprintf("length %u != %Zu\n", *len,
1013                                  sizeof(struct arpt_getinfo));
1014                         ret = -EINVAL;
1015                         break;
1016                 }
1017
1018                 if (copy_from_user(name, user, sizeof(name)) != 0) {
1019                         ret = -EFAULT;
1020                         break;
1021                 }
1022                 name[ARPT_TABLE_MAXNAMELEN-1] = '\0';
1023
1024                 t = try_then_request_module(xt_find_table_lock(NF_ARP, name),
1025                                             "arptable_%s", name);
1026                 if (t && !IS_ERR(t)) {
1027                         struct arpt_getinfo info;
1028                         struct xt_table_info *private = t->private;
1029
1030                         info.valid_hooks = t->valid_hooks;
1031                         memcpy(info.hook_entry, private->hook_entry,
1032                                sizeof(info.hook_entry));
1033                         memcpy(info.underflow, private->underflow,
1034                                sizeof(info.underflow));
1035                         info.num_entries = private->number;
1036                         info.size = private->size;
1037                         strcpy(info.name, name);
1038
1039                         if (copy_to_user(user, &info, *len) != 0)
1040                                 ret = -EFAULT;
1041                         else
1042                                 ret = 0;
1043                         xt_table_unlock(t);
1044                         module_put(t->me);
1045                 } else
1046                         ret = t ? PTR_ERR(t) : -ENOENT;
1047         }
1048         break;
1049
1050         case ARPT_SO_GET_ENTRIES: {
1051                 struct arpt_get_entries get;
1052
1053                 if (*len < sizeof(get)) {
1054                         duprintf("get_entries: %u < %Zu\n", *len, sizeof(get));
1055                         ret = -EINVAL;
1056                 } else if (copy_from_user(&get, user, sizeof(get)) != 0) {
1057                         ret = -EFAULT;
1058                 } else if (*len != sizeof(struct arpt_get_entries) + get.size) {
1059                         duprintf("get_entries: %u != %Zu\n", *len,
1060                                  sizeof(struct arpt_get_entries) + get.size);
1061                         ret = -EINVAL;
1062                 } else
1063                         ret = get_entries(&get, user);
1064                 break;
1065         }
1066
1067         case ARPT_SO_GET_REVISION_TARGET: {
1068                 struct xt_get_revision rev;
1069
1070                 if (*len != sizeof(rev)) {
1071                         ret = -EINVAL;
1072                         break;
1073                 }
1074                 if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
1075                         ret = -EFAULT;
1076                         break;
1077                 }
1078
1079                 try_then_request_module(xt_find_revision(NF_ARP, rev.name,
1080                                                          rev.revision, 1, &ret),
1081                                         "arpt_%s", rev.name);
1082                 break;
1083         }
1084
1085         default:
1086                 duprintf("do_arpt_get_ctl: unknown request %i\n", cmd);
1087                 ret = -EINVAL;
1088         }
1089
1090         return ret;
1091 }
1092
1093 int arpt_register_table(struct arpt_table *table,
1094                         const struct arpt_replace *repl)
1095 {
1096         int ret;
1097         struct xt_table_info *newinfo;
1098         static struct xt_table_info bootstrap
1099                 = { 0, 0, 0, { 0 }, { 0 }, { } };
1100         void *loc_cpu_entry;
1101
1102         newinfo = xt_alloc_table_info(repl->size);
1103         if (!newinfo) {
1104                 ret = -ENOMEM;
1105                 return ret;
1106         }
1107
1108         /* choose the copy on our node/cpu */
1109         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1110         memcpy(loc_cpu_entry, repl->entries, repl->size);
1111
1112         ret = translate_table(table->name, table->valid_hooks,
1113                               newinfo, loc_cpu_entry, repl->size,
1114                               repl->num_entries,
1115                               repl->hook_entry,
1116                               repl->underflow);
1117
1118         duprintf("arpt_register_table: translate table gives %d\n", ret);
1119         if (ret != 0) {
1120                 xt_free_table_info(newinfo);
1121                 return ret;
1122         }
1123
1124         if (xt_register_table(table, &bootstrap, newinfo) != 0) {
1125                 xt_free_table_info(newinfo);
1126                 return ret;
1127         }
1128
1129         return 0;
1130 }
1131
1132 void arpt_unregister_table(struct arpt_table *table)
1133 {
1134         struct xt_table_info *private;
1135         void *loc_cpu_entry;
1136
1137         private = xt_unregister_table(table);
1138
1139         /* Decrease module usage counts and free resources */
1140         loc_cpu_entry = private->entries[raw_smp_processor_id()];
1141         ARPT_ENTRY_ITERATE(loc_cpu_entry, private->size,
1142                            cleanup_entry, NULL);
1143         xt_free_table_info(private);
1144 }
1145
1146 /* The built-in targets: standard (NULL) and error. */
1147 static struct arpt_target arpt_standard_target = {
1148         .name           = ARPT_STANDARD_TARGET,
1149         .targetsize     = sizeof(int),
1150         .family         = NF_ARP,
1151 };
1152
1153 static struct arpt_target arpt_error_target = {
1154         .name           = ARPT_ERROR_TARGET,
1155         .target         = arpt_error,
1156         .targetsize     = ARPT_FUNCTION_MAXNAMELEN,
1157         .family         = NF_ARP,
1158 };
1159
1160 static struct nf_sockopt_ops arpt_sockopts = {
1161         .pf             = PF_INET,
1162         .set_optmin     = ARPT_BASE_CTL,
1163         .set_optmax     = ARPT_SO_SET_MAX+1,
1164         .set            = do_arpt_set_ctl,
1165         .get_optmin     = ARPT_BASE_CTL,
1166         .get_optmax     = ARPT_SO_GET_MAX+1,
1167         .get            = do_arpt_get_ctl,
1168 };
1169
1170 static int __init arp_tables_init(void)
1171 {
1172         int ret;
1173
1174         xt_proto_init(NF_ARP);
1175
1176         /* Noone else will be downing sem now, so we won't sleep */
1177         xt_register_target(&arpt_standard_target);
1178         xt_register_target(&arpt_error_target);
1179
1180         /* Register setsockopt */
1181         ret = nf_register_sockopt(&arpt_sockopts);
1182         if (ret < 0) {
1183                 duprintf("Unable to register sockopts.\n");
1184                 return ret;
1185         }
1186
1187         printk("arp_tables: (C) 2002 David S. Miller\n");
1188         return 0;
1189 }
1190
1191 static void __exit arp_tables_fini(void)
1192 {
1193         nf_unregister_sockopt(&arpt_sockopts);
1194         xt_proto_fini(NF_ARP);
1195 }
1196
1197 EXPORT_SYMBOL(arpt_register_table);
1198 EXPORT_SYMBOL(arpt_unregister_table);
1199 EXPORT_SYMBOL(arpt_do_table);
1200
1201 module_init(arp_tables_init);
1202 module_exit(arp_tables_fini);