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/config.h>
  13 #include <linux/kernel.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/netdevice.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 <asm/semaphore.h>
  25
  26 #include <linux/netfilter_arp/arp_tables.h>
  27
  28 MODULE_LICENSE("GPL");
  29 MODULE_AUTHOR("David S. Miller <davem@redhat.com>");
  30 MODULE_DESCRIPTION("arptables core");
  31
  32 /*#define DEBUG_ARP_TABLES*/
  33 /*#define DEBUG_ARP_TABLES_USER*/
  34
  35 #ifdef DEBUG_ARP_TABLES
  36 #define dprintf(format, args...)  printk(format , ## args)
  37 #else
  38 #define dprintf(format, args...)
  39 #endif
  40
  41 #ifdef DEBUG_ARP_TABLES_USER
  42 #define duprintf(format, args...) printk(format , ## args)
  43 #else
  44 #define duprintf(format, args...)
  45 #endif
  46
  47 #ifdef CONFIG_NETFILTER_DEBUG
  48 #define ARP_NF_ASSERT(x)                                        \
  49 do {                                                            \
  50         if (!(x))                                               \
  51                 printk("ARP_NF_ASSERT: %s:%s:%u\n",             \
  52                        __FUNCTION__, __FILE__, __LINE__);       \
  53 } while(0)
  54 #else
  55 #define ARP_NF_ASSERT(x)
  56 #endif
  57 #define SMP_ALIGN(x) (((x) + SMP_CACHE_BYTES-1) & ~(SMP_CACHE_BYTES-1))
  58
  59 static DECLARE_MUTEX(arpt_mutex);
  60
  61 #define ASSERT_READ_LOCK(x) ARP_NF_ASSERT(down_trylock(&arpt_mutex) != 0)
  62 #define ASSERT_WRITE_LOCK(x) ARP_NF_ASSERT(down_trylock(&arpt_mutex) != 0)
  63 #include <linux/netfilter_ipv4/listhelp.h>
  64
  65 struct arpt_table_info {
  66         unsigned int size;
  67         unsigned int number;
  68         unsigned int initial_entries;
  69         unsigned int hook_entry[NF_ARP_NUMHOOKS];
  70         unsigned int underflow[NF_ARP_NUMHOOKS];
  71         void *entries[NR_CPUS];
  72 };
  73
  74 static LIST_HEAD(arpt_target);
  75 static LIST_HEAD(arpt_tables);
  76 #define SET_COUNTER(c,b,p) do { (c).bcnt = (b); (c).pcnt = (p); } while(0)
  77 #define ADD_COUNTER(c,b,p) do { (c).bcnt += (b); (c).pcnt += (p); } while(0)
  78
  79 static inline int arp_devaddr_compare(const struct arpt_devaddr_info *ap,
  80                                       char *hdr_addr, int len)
  81 {
  82         int i, ret;
  83
  84         if (len > ARPT_DEV_ADDR_LEN_MAX)
  85                 len = ARPT_DEV_ADDR_LEN_MAX;
  86
  87         ret = 0;
  88         for (i = 0; i < len; i++)
  89                 ret |= (hdr_addr[i] ^ ap->addr[i]) & ap->mask[i];
  90
  91         return (ret != 0);
  92 }
  93
  94 /* Returns whether packet matches rule or not. */
  95 static inline int arp_packet_match(const struct arphdr *arphdr,
  96                                    struct net_device *dev,
  97                                    const char *indev,
  98                                    const char *outdev,
  99                                    const struct arpt_arp *arpinfo)
 100 {
 101         char *arpptr = (char *)(arphdr + 1);
 102         char *src_devaddr, *tgt_devaddr;
 103         u32 src_ipaddr, tgt_ipaddr;
 104         int i, ret;
 105
 106 #define FWINV(bool,invflg) ((bool) ^ !!(arpinfo->invflags & invflg))
 107
 108         if (FWINV((arphdr->ar_op & arpinfo->arpop_mask) != arpinfo->arpop,
 109                   ARPT_INV_ARPOP)) {
 110                 dprintf("ARP operation field mismatch.\n");
 111                 dprintf("ar_op: %04x info->arpop: %04x info->arpop_mask: %04x\n",
 112                         arphdr->ar_op, arpinfo->arpop, arpinfo->arpop_mask);
 113                 return 0;
 114         }
 115
 116         if (FWINV((arphdr->ar_hrd & arpinfo->arhrd_mask) != arpinfo->arhrd,
 117                   ARPT_INV_ARPHRD)) {
 118                 dprintf("ARP hardware address format mismatch.\n");
 119                 dprintf("ar_hrd: %04x info->arhrd: %04x info->arhrd_mask: %04x\n",
 120                         arphdr->ar_hrd, arpinfo->arhrd, arpinfo->arhrd_mask);
 121                 return 0;
 122         }
 123
 124         if (FWINV((arphdr->ar_pro & arpinfo->arpro_mask) != arpinfo->arpro,
 125                   ARPT_INV_ARPPRO)) {
 126                 dprintf("ARP protocol address format mismatch.\n");
 127                 dprintf("ar_pro: %04x info->arpro: %04x info->arpro_mask: %04x\n",
 128                         arphdr->ar_pro, arpinfo->arpro, arpinfo->arpro_mask);
 129                 return 0;
 130         }
 131
 132         if (FWINV((arphdr->ar_hln & arpinfo->arhln_mask) != arpinfo->arhln,
 133                   ARPT_INV_ARPHLN)) {
 134                 dprintf("ARP hardware address length mismatch.\n");
 135                 dprintf("ar_hln: %02x info->arhln: %02x info->arhln_mask: %02x\n",
 136                         arphdr->ar_hln, arpinfo->arhln, arpinfo->arhln_mask);
 137                 return 0;
 138         }
 139
 140         src_devaddr = arpptr;
 141         arpptr += dev->addr_len;
 142         memcpy(&src_ipaddr, arpptr, sizeof(u32));
 143         arpptr += sizeof(u32);
 144         tgt_devaddr = arpptr;
 145         arpptr += dev->addr_len;
 146         memcpy(&tgt_ipaddr, arpptr, sizeof(u32));
 147
 148         if (FWINV(arp_devaddr_compare(&arpinfo->src_devaddr, src_devaddr, dev->addr_len),
 149                   ARPT_INV_SRCDEVADDR) ||
 150             FWINV(arp_devaddr_compare(&arpinfo->tgt_devaddr, tgt_devaddr, dev->addr_len),
 151                   ARPT_INV_TGTDEVADDR)) {
 152                 dprintf("Source or target device address mismatch.\n");
 153
 154                 return 0;
 155         }
 156
 157         if (FWINV((src_ipaddr & arpinfo->smsk.s_addr) != arpinfo->src.s_addr,
 158                   ARPT_INV_SRCIP) ||
 159             FWINV(((tgt_ipaddr & arpinfo->tmsk.s_addr) != arpinfo->tgt.s_addr),
 160                   ARPT_INV_TGTIP)) {
 161                 dprintf("Source or target IP address mismatch.\n");
 162
 163                 dprintf("SRC: %u.%u.%u.%u. Mask: %u.%u.%u.%u. Target: %u.%u.%u.%u.%s\n",
 164                         NIPQUAD(src_ipaddr),
 165                         NIPQUAD(arpinfo->smsk.s_addr),
 166                         NIPQUAD(arpinfo->src.s_addr),
 167                         arpinfo->invflags & ARPT_INV_SRCIP ? " (INV)" : "");
 168                 dprintf("TGT: %u.%u.%u.%u Mask: %u.%u.%u.%u Target: %u.%u.%u.%u.%s\n",
 169                         NIPQUAD(tgt_ipaddr),
 170                         NIPQUAD(arpinfo->tmsk.s_addr),
 171                         NIPQUAD(arpinfo->tgt.s_addr),
 172                         arpinfo->invflags & ARPT_INV_TGTIP ? " (INV)" : "");
 173                 return 0;
 174         }
 175
 176         /* Look for ifname matches.  */
 177         for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
 178                 ret |= (indev[i] ^ arpinfo->iniface[i])
 179                         & arpinfo->iniface_mask[i];
 180         }
 181
 182         if (FWINV(ret != 0, ARPT_INV_VIA_IN)) {
 183                 dprintf("VIA in mismatch (%s vs %s).%s\n",
 184                         indev, arpinfo->iniface,
 185                         arpinfo->invflags&ARPT_INV_VIA_IN ?" (INV)":"");
 186                 return 0;
 187         }
 188
 189         for (i = 0, ret = 0; i < IFNAMSIZ/sizeof(unsigned long); i++) {
 190                 unsigned long odev;
 191                 memcpy(&odev, outdev + i*sizeof(unsigned long),
 192                        sizeof(unsigned long));
 193                 ret |= (odev
 194                         ^ ((const unsigned long *)arpinfo->outiface)[i])
 195                         & ((const unsigned long *)arpinfo->outiface_mask)[i];
 196         }
 197
 198         if (FWINV(ret != 0, ARPT_INV_VIA_OUT)) {
 199                 dprintf("VIA out mismatch (%s vs %s).%s\n",
 200                         outdev, arpinfo->outiface,
 201                         arpinfo->invflags&ARPT_INV_VIA_OUT ?" (INV)":"");
 202                 return 0;
 203         }
 204
 205         return 1;
 206 }
 207
 208 static inline int arp_checkentry(const struct arpt_arp *arp)
 209 {
 210         if (arp->flags & ~ARPT_F_MASK) {
 211                 duprintf("Unknown flag bits set: %08X\n",
 212                          arp->flags & ~ARPT_F_MASK);
 213                 return 0;
 214         }
 215         if (arp->invflags & ~ARPT_INV_MASK) {
 216                 duprintf("Unknown invflag bits set: %08X\n",
 217                          arp->invflags & ~ARPT_INV_MASK);
 218                 return 0;
 219         }
 220
 221         return 1;
 222 }
 223
 224 static unsigned int arpt_error(struct sk_buff **pskb,
 225                                unsigned int hooknum,
 226                                const struct net_device *in,
 227                                const struct net_device *out,
 228                                const void *targinfo,
 229                                void *userinfo)
 230 {
 231         if (net_ratelimit())
 232                 printk("arp_tables: error: '%s'\n", (char *)targinfo);
 233
 234         return NF_DROP;
 235 }
 236
 237 static inline struct arpt_entry *get_entry(void *base, unsigned int offset)
 238 {
 239         return (struct arpt_entry *)(base + offset);
 240 }
 241
 242 unsigned int arpt_do_table(struct sk_buff **pskb,
 243                            unsigned int hook,
 244                            const struct net_device *in,
 245                            const struct net_device *out,
 246                            struct arpt_table *table,
 247                            void *userdata)
 248 {
 249         static const char nulldevname[IFNAMSIZ];
 250         unsigned int verdict = NF_DROP;
 251         struct arphdr *arp;
 252         int hotdrop = 0;
 253         struct arpt_entry *e, *back;
 254         const char *indev, *outdev;
 255         void *table_base;
 256
 257         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
 258         if (!pskb_may_pull((*pskb), (sizeof(struct arphdr) +
 259                                      (2 * (*pskb)->dev->addr_len) +
 260                                      (2 * sizeof(u32)))))
 261                 return NF_DROP;
 262
 263         indev = in ? in->name : nulldevname;
 264         outdev = out ? out->name : nulldevname;
 265
 266         read_lock_bh(&table->lock);
 267         table_base = (void *)table->private->entries[smp_processor_id()];
 268         e = get_entry(table_base, table->private->hook_entry[hook]);
 269         back = get_entry(table_base, table->private->underflow[hook]);
 270
 271         arp = (*pskb)->nh.arph;
 272         do {
 273                 if (arp_packet_match(arp, (*pskb)->dev, indev, outdev, &e->arp)) {
 274                         struct arpt_entry_target *t;
 275                         int hdr_len;
 276
 277                         hdr_len = sizeof(*arp) + (2 * sizeof(struct in_addr)) +
 278                                 (2 * (*pskb)->dev->addr_len);
 279                         ADD_COUNTER(e->counters, hdr_len, 1);
 280
 281                         t = arpt_get_target(e);
 282
 283                         /* Standard target? */
 284                         if (!t->u.kernel.target->target) {
 285                                 int v;
 286
 287                                 v = ((struct arpt_standard_target *)t)->verdict;
 288                                 if (v < 0) {
 289                                         /* Pop from stack? */
 290                                         if (v != ARPT_RETURN) {
 291                                                 verdict = (unsigned)(-v) - 1;
 292                                                 break;
 293                                         }
 294                                         e = back;
 295                                         back = get_entry(table_base,
 296                                                          back->comefrom);
 297                                         continue;
 298                                 }
 299                                 if (table_base + v
 300                                     != (void *)e + e->next_offset) {
 301                                         /* Save old back ptr in next entry */
 302                                         struct arpt_entry *next
 303                                                 = (void *)e + e->next_offset;
 304                                         next->comefrom =
 305                                                 (void *)back - table_base;
 306
 307                                         /* set back pointer to next entry */
 308                                         back = next;
 309                                 }
 310
 311                                 e = get_entry(table_base, v);
 312                         } else {
 313                                 /* Targets which reenter must return
 314                                  * abs. verdicts
 315                                  */
 316                                 verdict = t->u.kernel.target->target(pskb,
 317                                                                      hook,
 318                                                                      in, out,
 319                                                                      t->data,
 320                                                                      userdata);
 321
 322                                 /* Target might have changed stuff. */
 323                                 arp = (*pskb)->nh.arph;
 324
 325                                 if (verdict == ARPT_CONTINUE)
 326                                         e = (void *)e + e->next_offset;
 327                                 else
 328                                         /* Verdict */
 329                                         break;
 330                         }
 331                 } else {
 332                         e = (void *)e + e->next_offset;
 333                 }
 334         } while (!hotdrop);
 335         read_unlock_bh(&table->lock);
 336
 337         if (hotdrop)
 338                 return NF_DROP;
 339         else
 340                 return verdict;
 341 }
 342
 343 /*
 344  * These are weird, but module loading must not be done with mutex
 345  * held (since they will register), and we have to have a single
 346  * function to use try_then_request_module().
 347  */
 348
 349 /* Find table by name, grabs mutex & ref.  Returns ERR_PTR() on error. */
 350 static inline struct arpt_table *find_table_lock(const char *name)
 351 {
 352         struct arpt_table *t;
 353
 354         if (down_interruptible(&arpt_mutex) != 0)
 355                 return ERR_PTR(-EINTR);
 356
 357         list_for_each_entry(t, &arpt_tables, list)
 358                 if (strcmp(t->name, name) == 0 && try_module_get(t->me))
 359                         return t;
 360         up(&arpt_mutex);
 361         return NULL;
 362 }
 363
 364
 365 /* Find target, grabs ref.  Returns ERR_PTR() on error. */
 366 static inline struct arpt_target *find_target(const char *name, u8 revision)
 367 {
 368         struct arpt_target *t;
 369         int err = 0;
 370
 371         if (down_interruptible(&arpt_mutex) != 0)
 372                 return ERR_PTR(-EINTR);
 373
 374         list_for_each_entry(t, &arpt_target, list) {
 375                 if (strcmp(t->name, name) == 0) {
 376                         if (t->revision == revision) {
 377                                 if (try_module_get(t->me)) {
 378                                         up(&arpt_mutex);
 379                                         return t;
 380                                 }
 381                         } else
 382                                 err = -EPROTOTYPE; /* Found something. */
 383                 }
 384         }
 385         up(&arpt_mutex);
 386         return ERR_PTR(err);
 387 }
 388
 389 struct arpt_target *arpt_find_target(const char *name, u8 revision)
 390 {
 391         struct arpt_target *target;
 392
 393         target = try_then_request_module(find_target(name, revision),
 394                                          "arpt_%s", name);
 395         if (IS_ERR(target) || !target)
 396                 return NULL;
 397         return target;
 398 }
 399
 400 static int target_revfn(const char *name, u8 revision, int *bestp)
 401 {
 402         struct arpt_target *t;
 403         int have_rev = 0;
 404
 405         list_for_each_entry(t, &arpt_target, list) {
 406                 if (strcmp(t->name, name) == 0) {
 407                         if (t->revision > *bestp)
 408                                 *bestp = t->revision;
 409                         if (t->revision == revision)
 410                                 have_rev =1;
 411                 }
 412         }
 413         return have_rev;
 414 }
 415
 416 /* Returns true or false (if no such extension at all) */
 417 static inline int find_revision(const char *name, u8 revision,
 418                                 int (*revfn)(const char *, u8, int *),
 419                                 int *err)
 420 {
 421         int have_rev, best = -1;
 422
 423         if (down_interruptible(&arpt_mutex) != 0) {
 424                 *err = -EINTR;
 425                 return 1;
 426         }
 427         have_rev = revfn(name, revision, &best);
 428         up(&arpt_mutex);
 429
 430         /* Nothing at all?  Return 0 to try loading module. */
 431         if (best == -1) {
 432                 *err = -ENOENT;
 433                 return 0;
 434         }
 435
 436         *err = best;
 437         if (!have_rev)
 438                 *err = -EPROTONOSUPPORT;
 439         return 1;
 440 }
 441
 442
 443 /* All zeroes == unconditional rule. */
 444 static inline int unconditional(const struct arpt_arp *arp)
 445 {
 446         unsigned int i;
 447
 448         for (i = 0; i < sizeof(*arp)/sizeof(__u32); i++)
 449                 if (((__u32 *)arp)[i])
 450                         return 0;
 451
 452         return 1;
 453 }
 454
 455 /* Figures out from what hook each rule can be called: returns 0 if
 456  * there are loops.  Puts hook bitmask in comefrom.
 457  */
 458 static int mark_source_chains(struct arpt_table_info *newinfo,
 459                               unsigned int valid_hooks, void *entry0)
 460 {
 461         unsigned int hook;
 462
 463         /* No recursion; use packet counter to save back ptrs (reset
 464          * to 0 as we leave), and comefrom to save source hook bitmask.
 465          */
 466         for (hook = 0; hook < NF_ARP_NUMHOOKS; hook++) {
 467                 unsigned int pos = newinfo->hook_entry[hook];
 468                 struct arpt_entry *e
 469                         = (struct arpt_entry *)(entry0 + pos);
 470
 471                 if (!(valid_hooks & (1 << hook)))
 472                         continue;
 473
 474                 /* Set initial back pointer. */
 475                 e->counters.pcnt = pos;
 476
 477                 for (;;) {
 478                         struct arpt_standard_target *t
 479                                 = (void *)arpt_get_target(e);
 480
 481                         if (e->comefrom & (1 << NF_ARP_NUMHOOKS)) {
 482                                 printk("arptables: loop hook %u pos %u %08X.\n",
 483                                        hook, pos, e->comefrom);
 484                                 return 0;
 485                         }
 486                         e->comefrom
 487                                 |= ((1 << hook) | (1 << NF_ARP_NUMHOOKS));
 488
 489                         /* Unconditional return/END. */
 490                         if (e->target_offset == sizeof(struct arpt_entry)
 491                             && (strcmp(t->target.u.user.name,
 492                                        ARPT_STANDARD_TARGET) == 0)
 493                             && t->verdict < 0
 494                             && unconditional(&e->arp)) {
 495                                 unsigned int oldpos, size;
 496
 497                                 /* Return: backtrack through the last
 498                                  * big jump.
 499                                  */
 500                                 do {
 501                                         e->comefrom ^= (1<<NF_ARP_NUMHOOKS);
 502                                         oldpos = pos;
 503                                         pos = e->counters.pcnt;
 504                                         e->counters.pcnt = 0;
 505
 506                                         /* We're at the start. */
 507                                         if (pos == oldpos)
 508                                                 goto next;
 509
 510                                         e = (struct arpt_entry *)
 511                                                 (entry0 + pos);
 512                                 } while (oldpos == pos + e->next_offset);
 513
 514                                 /* Move along one */
 515                                 size = e->next_offset;
 516                                 e = (struct arpt_entry *)
 517                                         (entry0 + pos + size);
 518                                 e->counters.pcnt = pos;
 519                                 pos += size;
 520                         } else {
 521                                 int newpos = t->verdict;
 522
 523                                 if (strcmp(t->target.u.user.name,
 524                                            ARPT_STANDARD_TARGET) == 0
 525                                     && newpos >= 0) {
 526                                         /* This a jump; chase it. */
 527                                         duprintf("Jump rule %u -> %u\n",
 528                                                  pos, newpos);
 529                                 } else {
 530                                         /* ... this is a fallthru */
 531                                         newpos = pos + e->next_offset;
 532                                 }
 533                                 e = (struct arpt_entry *)
 534                                         (entry0 + newpos);
 535                                 e->counters.pcnt = pos;
 536                                 pos = newpos;
 537                         }
 538                 }
 539                 next:
 540                 duprintf("Finished chain %u\n", hook);
 541         }
 542         return 1;
 543 }
 544
 545 static inline int standard_check(const struct arpt_entry_target *t,
 546                                  unsigned int max_offset)
 547 {
 548         struct arpt_standard_target *targ = (void *)t;
 549
 550         /* Check standard info. */
 551         if (t->u.target_size
 552             != ARPT_ALIGN(sizeof(struct arpt_standard_target))) {
 553                 duprintf("arpt_standard_check: target size %u != %Zu\n",
 554                          t->u.target_size,
 555                          ARPT_ALIGN(sizeof(struct arpt_standard_target)));
 556                 return 0;
 557         }
 558
 559         if (targ->verdict >= 0
 560             && targ->verdict > max_offset - sizeof(struct arpt_entry)) {
 561                 duprintf("arpt_standard_check: bad verdict (%i)\n",
 562                          targ->verdict);
 563                 return 0;
 564         }
 565
 566         if (targ->verdict < -NF_MAX_VERDICT - 1) {
 567                 duprintf("arpt_standard_check: bad negative verdict (%i)\n",
 568                          targ->verdict);
 569                 return 0;
 570         }
 571         return 1;
 572 }
 573
 574 static struct arpt_target arpt_standard_target;
 575
 576 static inline int check_entry(struct arpt_entry *e, const char *name, unsigned int size,
 577                               unsigned int *i)
 578 {
 579         struct arpt_entry_target *t;
 580         struct arpt_target *target;
 581         int ret;
 582
 583         if (!arp_checkentry(&e->arp)) {
 584                 duprintf("arp_tables: arp check failed %p %s.\n", e, name);
 585                 return -EINVAL;
 586         }
 587
 588         t = arpt_get_target(e);
 589         target = try_then_request_module(find_target(t->u.user.name,
 590                                                      t->u.user.revision),
 591                                          "arpt_%s", t->u.user.name);
 592         if (IS_ERR(target) || !target) {
 593                 duprintf("check_entry: `%s' not found\n", t->u.user.name);
 594                 ret = target ? PTR_ERR(target) : -ENOENT;
 595                 goto out;
 596         }
 597         t->u.kernel.target = target;
 598
 599         if (t->u.kernel.target == &arpt_standard_target) {
 600                 if (!standard_check(t, size)) {
 601                         ret = -EINVAL;
 602                         goto out;
 603                 }
 604         } else if (t->u.kernel.target->checkentry
 605                    && !t->u.kernel.target->checkentry(name, e, t->data,
 606                                                       t->u.target_size
 607                                                       - sizeof(*t),
 608                                                       e->comefrom)) {
 609                 module_put(t->u.kernel.target->me);
 610                 duprintf("arp_tables: check failed for `%s'.\n",
 611                          t->u.kernel.target->name);
 612                 ret = -EINVAL;
 613                 goto out;
 614         }
 615
 616         (*i)++;
 617         return 0;
 618
 619 out:
 620         return ret;
 621 }
 622
 623 static inline int check_entry_size_and_hooks(struct arpt_entry *e,
 624                                              struct arpt_table_info *newinfo,
 625                                              unsigned char *base,
 626                                              unsigned char *limit,
 627                                              const unsigned int *hook_entries,
 628                                              const unsigned int *underflows,
 629                                              unsigned int *i)
 630 {
 631         unsigned int h;
 632
 633         if ((unsigned long)e % __alignof__(struct arpt_entry) != 0
 634             || (unsigned char *)e + sizeof(struct arpt_entry) >= limit) {
 635                 duprintf("Bad offset %p\n", e);
 636                 return -EINVAL;
 637         }
 638
 639         if (e->next_offset
 640             < sizeof(struct arpt_entry) + sizeof(struct arpt_entry_target)) {
 641                 duprintf("checking: element %p size %u\n",
 642                          e, e->next_offset);
 643                 return -EINVAL;
 644         }
 645
 646         /* Check hooks & underflows */
 647         for (h = 0; h < NF_ARP_NUMHOOKS; h++) {
 648                 if ((unsigned char *)e - base == hook_entries[h])
 649                         newinfo->hook_entry[h] = hook_entries[h];
 650                 if ((unsigned char *)e - base == underflows[h])
 651                         newinfo->underflow[h] = underflows[h];
 652         }
 653
 654         /* FIXME: underflows must be unconditional, standard verdicts
 655            < 0 (not ARPT_RETURN). --RR */
 656
 657         /* Clear counters and comefrom */
 658         e->counters = ((struct arpt_counters) { 0, 0 });
 659         e->comefrom = 0;
 660
 661         (*i)++;
 662         return 0;
 663 }
 664
 665 static inline int cleanup_entry(struct arpt_entry *e, unsigned int *i)
 666 {
 667         struct arpt_entry_target *t;
 668
 669         if (i && (*i)-- == 0)
 670                 return 1;
 671
 672         t = arpt_get_target(e);
 673         if (t->u.kernel.target->destroy)
 674                 t->u.kernel.target->destroy(t->data,
 675                                             t->u.target_size - sizeof(*t));
 676         module_put(t->u.kernel.target->me);
 677         return 0;
 678 }
 679
 680 /* Checks and translates the user-supplied table segment (held in
 681  * newinfo).
 682  */
 683 static int translate_table(const char *name,
 684                            unsigned int valid_hooks,
 685                            struct arpt_table_info *newinfo,
 686                            void *entry0,
 687                            unsigned int size,
 688                            unsigned int number,
 689                            const unsigned int *hook_entries,
 690                            const unsigned int *underflows)
 691 {
 692         unsigned int i;
 693         int ret;
 694
 695         newinfo->size = size;
 696         newinfo->number = number;
 697
 698         /* Init all hooks to impossible value. */
 699         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
 700                 newinfo->hook_entry[i] = 0xFFFFFFFF;
 701                 newinfo->underflow[i] = 0xFFFFFFFF;
 702         }
 703
 704         duprintf("translate_table: size %u\n", newinfo->size);
 705         i = 0;
 706
 707         /* Walk through entries, checking offsets. */
 708         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 709                                  check_entry_size_and_hooks,
 710                                  newinfo,
 711                                  entry0,
 712                                  entry0 + size,
 713                                  hook_entries, underflows, &i);
 714         duprintf("translate_table: ARPT_ENTRY_ITERATE gives %d\n", ret);
 715         if (ret != 0)
 716                 return ret;
 717
 718         if (i != number) {
 719                 duprintf("translate_table: %u not %u entries\n",
 720                          i, number);
 721                 return -EINVAL;
 722         }
 723
 724         /* Check hooks all assigned */
 725         for (i = 0; i < NF_ARP_NUMHOOKS; i++) {
 726                 /* Only hooks which are valid */
 727                 if (!(valid_hooks & (1 << i)))
 728                         continue;
 729                 if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
 730                         duprintf("Invalid hook entry %u %u\n",
 731                                  i, hook_entries[i]);
 732                         return -EINVAL;
 733                 }
 734                 if (newinfo->underflow[i] == 0xFFFFFFFF) {
 735                         duprintf("Invalid underflow %u %u\n",
 736                                  i, underflows[i]);
 737                         return -EINVAL;
 738                 }
 739         }
 740
 741         if (!mark_source_chains(newinfo, valid_hooks, entry0)) {
 742                 duprintf("Looping hook\n");
 743                 return -ELOOP;
 744         }
 745
 746         /* Finally, each sanity check must pass */
 747         i = 0;
 748         ret = ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 749                                  check_entry, name, size, &i);
 750
 751         if (ret != 0) {
 752                 ARPT_ENTRY_ITERATE(entry0, newinfo->size,
 753                                    cleanup_entry, &i);
 754                 return ret;
 755         }
 756
 757         /* And one copy for every other CPU */
 758         for_each_cpu(i) {
 759                 if (newinfo->entries[i] && newinfo->entries[i] != entry0)
 760                         memcpy(newinfo->entries[i], entry0, newinfo->size);
 761         }
 762
 763         return ret;
 764 }
 765
 766 static struct arpt_table_info *replace_table(struct arpt_table *table,
 767                                              unsigned int num_counters,
 768                                              struct arpt_table_info *newinfo,
 769                                              int *error)
 770 {
 771         struct arpt_table_info *oldinfo;
 772
 773         /* Do the substitution. */
 774         write_lock_bh(&table->lock);
 775         /* Check inside lock: is the old number correct? */
 776         if (num_counters != table->private->number) {
 777                 duprintf("num_counters != table->private->number (%u/%u)\n",
 778                          num_counters, table->private->number);
 779                 write_unlock_bh(&table->lock);
 780                 *error = -EAGAIN;
 781                 return NULL;
 782         }
 783         oldinfo = table->private;
 784         table->private = newinfo;
 785         newinfo->initial_entries = oldinfo->initial_entries;
 786         write_unlock_bh(&table->lock);
 787
 788         return oldinfo;
 789 }
 790
 791 /* Gets counters. */
 792 static inline int add_entry_to_counter(const struct arpt_entry *e,
 793                                        struct arpt_counters total[],
 794                                        unsigned int *i)
 795 {
 796         ADD_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
 797
 798         (*i)++;
 799         return 0;
 800 }
 801
 802 static inline int set_entry_to_counter(const struct arpt_entry *e,
 803                                        struct arpt_counters total[],
 804                                        unsigned int *i)
 805 {
 806         SET_COUNTER(total[*i], e->counters.bcnt, e->counters.pcnt);
 807
 808         (*i)++;
 809         return 0;
 810 }
 811
 812 static void get_counters(const struct arpt_table_info *t,
 813                          struct arpt_counters counters[])
 814 {
 815         unsigned int cpu;
 816         unsigned int i;
 817         unsigned int curcpu;
 818
 819         /* Instead of clearing (by a previous call to memset())
 820          * the counters and using adds, we set the counters
 821          * with data used by 'current' CPU
 822          * We dont care about preemption here.
 823          */
 824         curcpu = raw_smp_processor_id();
 825
 826         i = 0;
 827         ARPT_ENTRY_ITERATE(t->entries[curcpu],
 828                            t->size,
 829                            set_entry_to_counter,
 830                            counters,
 831                            &i);
 832
 833         for_each_cpu(cpu) {
 834                 if (cpu == curcpu)
 835                         continue;
 836                 i = 0;
 837                 ARPT_ENTRY_ITERATE(t->entries[cpu],
 838                                    t->size,
 839                                    add_entry_to_counter,
 840                                    counters,
 841                                    &i);
 842         }
 843 }
 844
 845 static int copy_entries_to_user(unsigned int total_size,
 846                                 struct arpt_table *table,
 847                                 void __user *userptr)
 848 {
 849         unsigned int off, num, countersize;
 850         struct arpt_entry *e;
 851         struct arpt_counters *counters;
 852         int ret = 0;
 853         void *loc_cpu_entry;
 854
 855         /* We need atomic snapshot of counters: rest doesn't change
 856          * (other than comefrom, which userspace doesn't care
 857          * about).
 858          */
 859         countersize = sizeof(struct arpt_counters) * table->private->number;
 860         counters = vmalloc(countersize);
 861
 862         if (counters == NULL)
 863                 return -ENOMEM;
 864
 865         /* First, sum counters... */
 866         write_lock_bh(&table->lock);
 867         get_counters(table->private, counters);
 868         write_unlock_bh(&table->lock);
 869
 870         loc_cpu_entry = table->private->entries[raw_smp_processor_id()];
 871         /* ... then copy entire thing ... */
 872         if (copy_to_user(userptr, loc_cpu_entry, total_size) != 0) {
 873                 ret = -EFAULT;
 874                 goto free_counters;
 875         }
 876
 877         /* FIXME: use iterator macros --RR */
 878         /* ... then go back and fix counters and names */
 879         for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
 880                 struct arpt_entry_target *t;
 881
 882                 e = (struct arpt_entry *)(loc_cpu_entry + off);
 883                 if (copy_to_user(userptr + off
 884                                  + offsetof(struct arpt_entry, counters),
 885                                  &counters[num],
 886                                  sizeof(counters[num])) != 0) {
 887                         ret = -EFAULT;
 888                         goto free_counters;
 889                 }
 890
 891                 t = arpt_get_target(e);
 892                 if (copy_to_user(userptr + off + e->target_offset
 893                                  + offsetof(struct arpt_entry_target,
 894                                             u.user.name),
 895                                  t->u.kernel.target->name,
 896                                  strlen(t->u.kernel.target->name)+1) != 0) {
 897                         ret = -EFAULT;
 898                         goto free_counters;
 899                 }
 900         }
 901
 902  free_counters:
 903         vfree(counters);
 904         return ret;
 905 }
 906
 907 static int get_entries(const struct arpt_get_entries *entries,
 908                        struct arpt_get_entries __user *uptr)
 909 {
 910         int ret;
 911         struct arpt_table *t;
 912
 913         t = find_table_lock(entries->name);
 914         if (t || !IS_ERR(t)) {
 915                 duprintf("t->private->number = %u\n",
 916                          t->private->number);
 917                 if (entries->size == t->private->size)
 918                         ret = copy_entries_to_user(t->private->size,
 919                                                    t, uptr->entrytable);
 920                 else {
 921                         duprintf("get_entries: I've got %u not %u!\n",
 922                                  t->private->size,
 923                                  entries->size);
 924                         ret = -EINVAL;
 925                 }
 926                 module_put(t->me);
 927                 up(&arpt_mutex);
 928         } else
 929                 ret = t ? PTR_ERR(t) : -ENOENT;
 930
 931         return ret;
 932 }
 933
 934 static void free_table_info(struct arpt_table_info *info)
 935 {
 936         int cpu;
 937         for_each_cpu(cpu) {
 938                 if (info->size <= PAGE_SIZE)
 939                         kfree(info->entries[cpu]);
 940                 else
 941                         vfree(info->entries[cpu]);
 942         }
 943         kfree(info);
 944 }
 945
 946 static struct arpt_table_info *alloc_table_info(unsigned int size)
 947 {
 948         struct arpt_table_info *newinfo;
 949         int cpu;
 950
 951         newinfo = kzalloc(sizeof(struct arpt_table_info), GFP_KERNEL);
 952         if (!newinfo)
 953                 return NULL;
 954
 955         newinfo->size = size;
 956
 957         for_each_cpu(cpu) {
 958                 if (size <= PAGE_SIZE)
 959                         newinfo->entries[cpu] = kmalloc_node(size,
 960                                                         GFP_KERNEL,
 961                                                         cpu_to_node(cpu));
 962                 else
 963                         newinfo->entries[cpu] = vmalloc_node(size,
 964                                                              cpu_to_node(cpu));
 965
 966                 if (newinfo->entries[cpu] == NULL) {
 967                         free_table_info(newinfo);
 968                         return NULL;
 969                 }
 970         }
 971
 972         return newinfo;
 973 }
 974
 975 static int do_replace(void __user *user, unsigned int len)
 976 {
 977         int ret;
 978         struct arpt_replace tmp;
 979         struct arpt_table *t;
 980         struct arpt_table_info *newinfo, *oldinfo;
 981         struct arpt_counters *counters;
 982         void *loc_cpu_entry, *loc_cpu_old_entry;
 983
 984         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
 985                 return -EFAULT;
 986
 987         /* Hack: Causes ipchains to give correct error msg --RR */
 988         if (len != sizeof(tmp) + tmp.size)
 989                 return -ENOPROTOOPT;
 990
 991         /* Pedantry: prevent them from hitting BUG() in vmalloc.c --RR */
 992         if ((SMP_ALIGN(tmp.size) >> PAGE_SHIFT) + 2 > num_physpages)
 993                 return -ENOMEM;
 994
 995         newinfo = alloc_table_info(tmp.size);
 996         if (!newinfo)
 997                 return -ENOMEM;
 998
 999         /* choose the copy that is on our node/cpu */
1000         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1001         if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
1002                            tmp.size) != 0) {
1003                 ret = -EFAULT;
1004                 goto free_newinfo;
1005         }
1006
1007         counters = vmalloc(tmp.num_counters * sizeof(struct arpt_counters));
1008         if (!counters) {
1009                 ret = -ENOMEM;
1010                 goto free_newinfo;
1011         }
1012
1013         ret = translate_table(tmp.name, tmp.valid_hooks,
1014                               newinfo, loc_cpu_entry, tmp.size, tmp.num_entries,
1015                               tmp.hook_entry, tmp.underflow);
1016         if (ret != 0)
1017                 goto free_newinfo_counters;
1018
1019         duprintf("arp_tables: Translated table\n");
1020
1021         t = try_then_request_module(find_table_lock(tmp.name),
1022                                     "arptable_%s", tmp.name);
1023         if (!t || IS_ERR(t)) {
1024                 ret = t ? PTR_ERR(t) : -ENOENT;
1025                 goto free_newinfo_counters_untrans;
1026         }
1027
1028         /* You lied! */
1029         if (tmp.valid_hooks != t->valid_hooks) {
1030                 duprintf("Valid hook crap: %08X vs %08X\n",
1031                          tmp.valid_hooks, t->valid_hooks);
1032                 ret = -EINVAL;
1033                 goto put_module;
1034         }
1035
1036         oldinfo = replace_table(t, tmp.num_counters, newinfo, &ret);
1037         if (!oldinfo)
1038                 goto put_module;
1039
1040         /* Update module usage count based on number of rules */
1041         duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
1042                 oldinfo->number, oldinfo->initial_entries, newinfo->number);
1043         if ((oldinfo->number > oldinfo->initial_entries) ||
1044             (newinfo->number <= oldinfo->initial_entries))
1045                 module_put(t->me);
1046         if ((oldinfo->number > oldinfo->initial_entries) &&
1047             (newinfo->number <= oldinfo->initial_entries))
1048                 module_put(t->me);
1049
1050         /* Get the old counters. */
1051         get_counters(oldinfo, counters);
1052         /* Decrease module usage counts and free resource */
1053         loc_cpu_old_entry = oldinfo->entries[raw_smp_processor_id()];
1054         ARPT_ENTRY_ITERATE(loc_cpu_old_entry, oldinfo->size, cleanup_entry,NULL);
1055
1056         free_table_info(oldinfo);
1057         if (copy_to_user(tmp.counters, counters,
1058                          sizeof(struct arpt_counters) * tmp.num_counters) != 0)
1059                 ret = -EFAULT;
1060         vfree(counters);
1061         up(&arpt_mutex);
1062         return ret;
1063
1064  put_module:
1065         module_put(t->me);
1066         up(&arpt_mutex);
1067  free_newinfo_counters_untrans:
1068         ARPT_ENTRY_ITERATE(loc_cpu_entry, newinfo->size, cleanup_entry, NULL);
1069  free_newinfo_counters:
1070         vfree(counters);
1071  free_newinfo:
1072         free_table_info(newinfo);
1073         return ret;
1074 }
1075
1076 /* We're lazy, and add to the first CPU; overflow works its fey magic
1077  * and everything is OK.
1078  */
1079 static inline int add_counter_to_entry(struct arpt_entry *e,
1080                                        const struct arpt_counters addme[],
1081                                        unsigned int *i)
1082 {
1083
1084         ADD_COUNTER(e->counters, addme[*i].bcnt, addme[*i].pcnt);
1085
1086         (*i)++;
1087         return 0;
1088 }
1089
1090 static int do_add_counters(void __user *user, unsigned int len)
1091 {
1092         unsigned int i;
1093         struct arpt_counters_info tmp, *paddc;
1094         struct arpt_table *t;
1095         int ret = 0;
1096         void *loc_cpu_entry;
1097
1098         if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
1099                 return -EFAULT;
1100
1101         if (len != sizeof(tmp) + tmp.num_counters*sizeof(struct arpt_counters))
1102                 return -EINVAL;
1103
1104         paddc = vmalloc(len);
1105         if (!paddc)
1106                 return -ENOMEM;
1107
1108         if (copy_from_user(paddc, user, len) != 0) {
1109                 ret = -EFAULT;
1110                 goto free;
1111         }
1112
1113         t = find_table_lock(tmp.name);
1114         if (!t || IS_ERR(t)) {
1115                 ret = t ? PTR_ERR(t) : -ENOENT;
1116                 goto free;
1117         }
1118
1119         write_lock_bh(&t->lock);
1120         if (t->private->number != paddc->num_counters) {
1121                 ret = -EINVAL;
1122                 goto unlock_up_free;
1123         }
1124
1125         i = 0;
1126         /* Choose the copy that is on our node */
1127         loc_cpu_entry = t->private->entries[smp_processor_id()];
1128         ARPT_ENTRY_ITERATE(loc_cpu_entry,
1129                            t->private->size,
1130                            add_counter_to_entry,
1131                            paddc->counters,
1132                            &i);
1133  unlock_up_free:
1134         write_unlock_bh(&t->lock);
1135         up(&arpt_mutex);
1136         module_put(t->me);
1137  free:
1138         vfree(paddc);
1139
1140         return ret;
1141 }
1142
1143 static int do_arpt_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
1144 {
1145         int ret;
1146
1147         if (!capable(CAP_NET_ADMIN))
1148                 return -EPERM;
1149
1150         switch (cmd) {
1151         case ARPT_SO_SET_REPLACE:
1152                 ret = do_replace(user, len);
1153                 break;
1154
1155         case ARPT_SO_SET_ADD_COUNTERS:
1156                 ret = do_add_counters(user, len);
1157                 break;
1158
1159         default:
1160                 duprintf("do_arpt_set_ctl:  unknown request %i\n", cmd);
1161                 ret = -EINVAL;
1162         }
1163
1164         return ret;
1165 }
1166
1167 static int do_arpt_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
1168 {
1169         int ret;
1170
1171         if (!capable(CAP_NET_ADMIN))
1172                 return -EPERM;
1173
1174         switch (cmd) {
1175         case ARPT_SO_GET_INFO: {
1176                 char name[ARPT_TABLE_MAXNAMELEN];
1177                 struct arpt_table *t;
1178
1179                 if (*len != sizeof(struct arpt_getinfo)) {
1180                         duprintf("length %u != %Zu\n", *len,
1181                                  sizeof(struct arpt_getinfo));
1182                         ret = -EINVAL;
1183                         break;
1184                 }
1185
1186                 if (copy_from_user(name, user, sizeof(name)) != 0) {
1187                         ret = -EFAULT;
1188                         break;
1189                 }
1190                 name[ARPT_TABLE_MAXNAMELEN-1] = '\0';
1191
1192                 t = try_then_request_module(find_table_lock(name),
1193                                             "arptable_%s", name);
1194                 if (t && !IS_ERR(t)) {
1195                         struct arpt_getinfo info;
1196
1197                         info.valid_hooks = t->valid_hooks;
1198                         memcpy(info.hook_entry, t->private->hook_entry,
1199                                sizeof(info.hook_entry));
1200                         memcpy(info.underflow, t->private->underflow,
1201                                sizeof(info.underflow));
1202                         info.num_entries = t->private->number;
1203                         info.size = t->private->size;
1204                         strcpy(info.name, name);
1205
1206                         if (copy_to_user(user, &info, *len) != 0)
1207                                 ret = -EFAULT;
1208                         else
1209                                 ret = 0;
1210                         up(&arpt_mutex);
1211                         module_put(t->me);
1212                 } else
1213                         ret = t ? PTR_ERR(t) : -ENOENT;
1214         }
1215         break;
1216
1217         case ARPT_SO_GET_ENTRIES: {
1218                 struct arpt_get_entries get;
1219
1220                 if (*len < sizeof(get)) {
1221                         duprintf("get_entries: %u < %Zu\n", *len, sizeof(get));
1222                         ret = -EINVAL;
1223                 } else if (copy_from_user(&get, user, sizeof(get)) != 0) {
1224                         ret = -EFAULT;
1225                 } else if (*len != sizeof(struct arpt_get_entries) + get.size) {
1226                         duprintf("get_entries: %u != %Zu\n", *len,
1227                                  sizeof(struct arpt_get_entries) + get.size);
1228                         ret = -EINVAL;
1229                 } else
1230                         ret = get_entries(&get, user);
1231                 break;
1232         }
1233
1234         case ARPT_SO_GET_REVISION_TARGET: {
1235                 struct arpt_get_revision rev;
1236
1237                 if (*len != sizeof(rev)) {
1238                         ret = -EINVAL;
1239                         break;
1240                 }
1241                 if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
1242                         ret = -EFAULT;
1243                         break;
1244                 }
1245
1246                 try_then_request_module(find_revision(rev.name, rev.revision,
1247                                                       target_revfn, &ret),
1248                                         "arpt_%s", rev.name);
1249                 break;
1250         }
1251
1252         default:
1253                 duprintf("do_arpt_get_ctl: unknown request %i\n", cmd);
1254                 ret = -EINVAL;
1255         }
1256
1257         return ret;
1258 }
1259
1260 /* Registration hooks for targets. */
1261 int arpt_register_target(struct arpt_target *target)
1262 {
1263         int ret;
1264
1265         ret = down_interruptible(&arpt_mutex);
1266         if (ret != 0)
1267                 return ret;
1268
1269         list_add(&target->list, &arpt_target);
1270         up(&arpt_mutex);
1271
1272         return ret;
1273 }
1274
1275 void arpt_unregister_target(struct arpt_target *target)
1276 {
1277         down(&arpt_mutex);
1278         LIST_DELETE(&arpt_target, target);
1279         up(&arpt_mutex);
1280 }
1281
1282 int arpt_register_table(struct arpt_table *table,
1283                         const struct arpt_replace *repl)
1284 {
1285         int ret;
1286         struct arpt_table_info *newinfo;
1287         static struct arpt_table_info bootstrap
1288                 = { 0, 0, 0, { 0 }, { 0 }, { } };
1289         void *loc_cpu_entry;
1290
1291         newinfo = alloc_table_info(repl->size);
1292         if (!newinfo) {
1293                 ret = -ENOMEM;
1294                 return ret;
1295         }
1296
1297         /* choose the copy on our node/cpu */
1298         loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
1299         memcpy(loc_cpu_entry, repl->entries, repl->size);
1300
1301         ret = translate_table(table->name, table->valid_hooks,
1302                               newinfo, loc_cpu_entry, repl->size,
1303                               repl->num_entries,
1304                               repl->hook_entry,
1305                               repl->underflow);
1306         duprintf("arpt_register_table: translate table gives %d\n", ret);
1307         if (ret != 0) {
1308                 free_table_info(newinfo);
1309                 return ret;
1310         }
1311
1312         ret = down_interruptible(&arpt_mutex);
1313         if (ret != 0) {
1314                 free_table_info(newinfo);
1315                 return ret;
1316         }
1317
1318         /* Don't autoload: we'd eat our tail... */
1319         if (list_named_find(&arpt_tables, table->name)) {
1320                 ret = -EEXIST;
1321                 goto free_unlock;
1322         }
1323
1324         /* Simplifies replace_table code. */
1325         table->private = &bootstrap;
1326         if (!replace_table(table, 0, newinfo, &ret))
1327                 goto free_unlock;
1328
1329         duprintf("table->private->number = %u\n",
1330                  table->private->number);
1331
1332         /* save number of initial entries */
1333         table->private->initial_entries = table->private->number;
1334
1335         rwlock_init(&table->lock);
1336         list_prepend(&arpt_tables, table);
1337
1338  unlock:
1339         up(&arpt_mutex);
1340         return ret;
1341
1342  free_unlock:
1343         free_table_info(newinfo);
1344         goto unlock;
1345 }
1346
1347 void arpt_unregister_table(struct arpt_table *table)
1348 {
1349         void *loc_cpu_entry;
1350
1351         down(&arpt_mutex);
1352         LIST_DELETE(&arpt_tables, table);
1353         up(&arpt_mutex);
1354
1355         /* Decrease module usage counts and free resources */
1356         loc_cpu_entry = table->private->entries[raw_smp_processor_id()];
1357         ARPT_ENTRY_ITERATE(loc_cpu_entry, table->private->size,
1358                            cleanup_entry, NULL);
1359         free_table_info(table->private);
1360 }
1361
1362 /* The built-in targets: standard (NULL) and error. */
1363 static struct arpt_target arpt_standard_target = {
1364         .name           = ARPT_STANDARD_TARGET,
1365 };
1366
1367 static struct arpt_target arpt_error_target = {
1368         .name           = ARPT_ERROR_TARGET,
1369         .target         = arpt_error,
1370 };
1371
1372 static struct nf_sockopt_ops arpt_sockopts = {
1373         .pf             = PF_INET,
1374         .set_optmin     = ARPT_BASE_CTL,
1375         .set_optmax     = ARPT_SO_SET_MAX+1,
1376         .set            = do_arpt_set_ctl,
1377         .get_optmin     = ARPT_BASE_CTL,
1378         .get_optmax     = ARPT_SO_GET_MAX+1,
1379         .get            = do_arpt_get_ctl,
1380 };
1381
1382 #ifdef CONFIG_PROC_FS
1383 static inline int print_name(const struct arpt_table *t,
1384                              off_t start_offset, char *buffer, int length,
1385                              off_t *pos, unsigned int *count)
1386 {
1387         if ((*count)++ >= start_offset) {
1388                 unsigned int namelen;
1389
1390                 namelen = sprintf(buffer + *pos, "%s\n", t->name);
1391                 if (*pos + namelen > length) {
1392                         /* Stop iterating */
1393                         return 1;
1394                 }
1395                 *pos += namelen;
1396         }
1397         return 0;
1398 }
1399
1400 static int arpt_get_tables(char *buffer, char **start, off_t offset, int length)
1401 {
1402         off_t pos = 0;
1403         unsigned int count = 0;
1404
1405         if (down_interruptible(&arpt_mutex) != 0)
1406                 return 0;
1407
1408         LIST_FIND(&arpt_tables, print_name, struct arpt_table *,
1409                   offset, buffer, length, &pos, &count);
1410
1411         up(&arpt_mutex);
1412
1413         /* `start' hack - see fs/proc/generic.c line ~105 */
1414         *start=(char *)((unsigned long)count-offset);
1415         return pos;
1416 }
1417 #endif /*CONFIG_PROC_FS*/
1418
1419 static int __init init(void)
1420 {
1421         int ret;
1422
1423         /* Noone else will be downing sem now, so we won't sleep */
1424         down(&arpt_mutex);
1425         list_append(&arpt_target, &arpt_standard_target);
1426         list_append(&arpt_target, &arpt_error_target);
1427         up(&arpt_mutex);
1428
1429         /* Register setsockopt */
1430         ret = nf_register_sockopt(&arpt_sockopts);
1431         if (ret < 0) {
1432                 duprintf("Unable to register sockopts.\n");
1433                 return ret;
1434         }
1435
1436 #ifdef CONFIG_PROC_FS
1437         {
1438                 struct proc_dir_entry *proc;
1439
1440                 proc = proc_net_create("arp_tables_names", 0, arpt_get_tables);
1441                 if (!proc) {
1442                         nf_unregister_sockopt(&arpt_sockopts);
1443                         return -ENOMEM;
1444                 }
1445                 proc->owner = THIS_MODULE;
1446         }
1447 #endif
1448
1449         printk("arp_tables: (C) 2002 David S. Miller\n");
1450         return 0;
1451 }
1452
1453 static void __exit fini(void)
1454 {
1455         nf_unregister_sockopt(&arpt_sockopts);
1456 #ifdef CONFIG_PROC_FS
1457         proc_net_remove("arp_tables_names");
1458 #endif
1459 }
1460
1461 EXPORT_SYMBOL(arpt_register_table);
1462 EXPORT_SYMBOL(arpt_unregister_table);
1463 EXPORT_SYMBOL(arpt_do_table);
1464 EXPORT_SYMBOL(arpt_register_target);
1465 EXPORT_SYMBOL(arpt_unregister_target);
1466
1467 module_init(init);
1468 module_exit(fini);