include/linux/netfilter/ipset/ip_set_ahash.h

   1 #ifndef _IP_SET_AHASH_H
   2 #define _IP_SET_AHASH_H
   3
   4 #include <linux/rcupdate.h>
   5 #include <linux/jhash.h>
   6 #include <linux/netfilter/ipset/ip_set_timeout.h>
   7
   8 #define CONCAT(a, b, c)         a##b##c
   9 #define TOKEN(a, b, c)          CONCAT(a, b, c)
  10
  11 #define type_pf_next            TOKEN(TYPE, PF, _elem)
  12
  13 /* Hashing which uses arrays to resolve clashing. The hash table is resized
  14  * (doubled) when searching becomes too long.
  15  * Internally jhash is used with the assumption that the size of the
  16  * stored data is a multiple of sizeof(u32). If storage supports timeout,
  17  * the timeout field must be the last one in the data structure - that field
  18  * is ignored when computing the hash key.
  19  *
  20  * Readers and resizing
  21  *
  22  * Resizing can be triggered by userspace command only, and those
  23  * are serialized by the nfnl mutex. During resizing the set is
  24  * read-locked, so the only possible concurrent operations are
  25  * the kernel side readers. Those must be protected by proper RCU locking.
  26  */
  27
  28 /* Number of elements to store in an initial array block */
  29 #define AHASH_INIT_SIZE                 4
  30 /* Max number of elements to store in an array block */
  31 #define AHASH_MAX_SIZE                  (3*4)
  32
  33 /* A hash bucket */
  34 struct hbucket {
  35         void *value;            /* the array of the values */
  36         u8 size;                /* size of the array */
  37         u8 pos;                 /* position of the first free entry */
  38 };
  39
  40 /* The hash table: the table size stored here in order to make resizing easy */
  41 struct htable {
  42         u8 htable_bits;         /* size of hash table == 2^htable_bits */
  43         struct hbucket bucket[0]; /* hashtable buckets */
  44 };
  45
  46 #define hbucket(h, i)           (&((h)->bucket[i]))
  47
  48 /* Book-keeping of the prefixes added to the set */
  49 struct ip_set_hash_nets {
  50         u8 cidr;                /* the different cidr values in the set */
  51         u32 nets;               /* number of elements per cidr */
  52 };
  53
  54 /* The generic ip_set hash structure */
  55 struct ip_set_hash {
  56         struct htable *table;   /* the hash table */
  57         u32 maxelem;            /* max elements in the hash */
  58         u32 elements;           /* current element (vs timeout) */
  59         u32 initval;            /* random jhash init value */
  60         u32 timeout;            /* timeout value, if enabled */
  61         struct timer_list gc;   /* garbage collection when timeout enabled */
  62         struct type_pf_next next; /* temporary storage for uadd */
  63 #ifdef IP_SET_HASH_WITH_NETMASK
  64         u8 netmask;             /* netmask value for subnets to store */
  65 #endif
  66 #ifdef IP_SET_HASH_WITH_RBTREE
  67         struct rb_root rbtree;
  68 #endif
  69 #ifdef IP_SET_HASH_WITH_NETS
  70         struct ip_set_hash_nets nets[0]; /* book-keeping of prefixes */
  71 #endif
  72 };
  73
  74 /* Compute htable_bits from the user input parameter hashsize */
  75 static u8
  76 htable_bits(u32 hashsize)
  77 {
  78         /* Assume that hashsize == 2^htable_bits */
  79         u8 bits = fls(hashsize - 1);
  80         if (jhash_size(bits) != hashsize)
  81                 /* Round up to the first 2^n value */
  82                 bits = fls(hashsize);
  83
  84         return bits;
  85 }
  86
  87 #ifdef IP_SET_HASH_WITH_NETS
  88
  89 #define SET_HOST_MASK(family)   (family == AF_INET ? 32 : 128)
  90
  91 /* Network cidr size book keeping when the hash stores different
  92  * sized networks */
  93 static void
  94 add_cidr(struct ip_set_hash *h, u8 cidr, u8 host_mask)
  95 {
  96         u8 i;
  97
  98         ++h->nets[cidr-1].nets;
  99
 100         pr_debug("add_cidr added %u: %u\n", cidr, h->nets[cidr-1].nets);
 101
 102         if (h->nets[cidr-1].nets > 1)
 103                 return;
 104
 105         /* New cidr size */
 106         for (i = 0; i < host_mask && h->nets[i].cidr; i++) {
 107                 /* Add in increasing prefix order, so larger cidr first */
 108                 if (h->nets[i].cidr < cidr)
 109                         swap(h->nets[i].cidr, cidr);
 110         }
 111         if (i < host_mask)
 112                 h->nets[i].cidr = cidr;
 113 }
 114
 115 static void
 116 del_cidr(struct ip_set_hash *h, u8 cidr, u8 host_mask)
 117 {
 118         u8 i;
 119
 120         --h->nets[cidr-1].nets;
 121
 122         pr_debug("del_cidr deleted %u: %u\n", cidr, h->nets[cidr-1].nets);
 123
 124         if (h->nets[cidr-1].nets != 0)
 125                 return;
 126
 127         /* All entries with this cidr size deleted, so cleanup h->cidr[] */
 128         for (i = 0; i < host_mask - 1 && h->nets[i].cidr; i++) {
 129                 if (h->nets[i].cidr == cidr)
 130                         h->nets[i].cidr = cidr = h->nets[i+1].cidr;
 131         }
 132         h->nets[i - 1].cidr = 0;
 133 }
 134 #endif
 135
 136 /* Destroy the hashtable part of the set */
 137 static void
 138 ahash_destroy(struct htable *t)
 139 {
 140         struct hbucket *n;
 141         u32 i;
 142
 143         for (i = 0; i < jhash_size(t->htable_bits); i++) {
 144                 n = hbucket(t, i);
 145                 if (n->size)
 146                         /* FIXME: use slab cache */
 147                         kfree(n->value);
 148         }
 149
 150         ip_set_free(t);
 151 }
 152
 153 /* Calculate the actual memory size of the set data */
 154 static size_t
 155 ahash_memsize(const struct ip_set_hash *h, size_t dsize, u8 host_mask)
 156 {
 157         u32 i;
 158         struct htable *t = h->table;
 159         size_t memsize = sizeof(*h)
 160                          + sizeof(*t)
 161 #ifdef IP_SET_HASH_WITH_NETS
 162                          + sizeof(struct ip_set_hash_nets) * host_mask
 163 #endif
 164                          + jhash_size(t->htable_bits) * sizeof(struct hbucket);
 165
 166         for (i = 0; i < jhash_size(t->htable_bits); i++)
 167                         memsize += t->bucket[i].size * dsize;
 168
 169         return memsize;
 170 }
 171
 172 /* Flush a hash type of set: destroy all elements */
 173 static void
 174 ip_set_hash_flush(struct ip_set *set)
 175 {
 176         struct ip_set_hash *h = set->data;
 177         struct htable *t = h->table;
 178         struct hbucket *n;
 179         u32 i;
 180
 181         for (i = 0; i < jhash_size(t->htable_bits); i++) {
 182                 n = hbucket(t, i);
 183                 if (n->size) {
 184                         n->size = n->pos = 0;
 185                         /* FIXME: use slab cache */
 186                         kfree(n->value);
 187                 }
 188         }
 189 #ifdef IP_SET_HASH_WITH_NETS
 190         memset(h->nets, 0, sizeof(struct ip_set_hash_nets)
 191                            * SET_HOST_MASK(set->family));
 192 #endif
 193         h->elements = 0;
 194 }
 195
 196 /* Destroy a hash type of set */
 197 static void
 198 ip_set_hash_destroy(struct ip_set *set)
 199 {
 200         struct ip_set_hash *h = set->data;
 201
 202         if (with_timeout(h->timeout))
 203                 del_timer_sync(&h->gc);
 204
 205         ahash_destroy(h->table);
 206 #ifdef IP_SET_HASH_WITH_RBTREE
 207         rbtree_destroy(&h->rbtree);
 208 #endif
 209         kfree(h);
 210
 211         set->data = NULL;
 212 }
 213
 214 #define HKEY(data, initval, htable_bits)                                 \
 215 (jhash2((u32 *)(data), sizeof(struct type_pf_elem)/sizeof(u32), initval) \
 216         & jhash_mask(htable_bits))
 217
 218 #endif /* _IP_SET_AHASH_H */
 219
 220 #define CONCAT(a, b, c)         a##b##c
 221 #define TOKEN(a, b, c)          CONCAT(a, b, c)
 222
 223 /* Type/family dependent function prototypes */
 224
 225 #define type_pf_data_equal      TOKEN(TYPE, PF, _data_equal)
 226 #define type_pf_data_isnull     TOKEN(TYPE, PF, _data_isnull)
 227 #define type_pf_data_copy       TOKEN(TYPE, PF, _data_copy)
 228 #define type_pf_data_zero_out   TOKEN(TYPE, PF, _data_zero_out)
 229 #define type_pf_data_netmask    TOKEN(TYPE, PF, _data_netmask)
 230 #define type_pf_data_list       TOKEN(TYPE, PF, _data_list)
 231 #define type_pf_data_tlist      TOKEN(TYPE, PF, _data_tlist)
 232 #define type_pf_data_next       TOKEN(TYPE, PF, _data_next)
 233
 234 #define type_pf_elem            TOKEN(TYPE, PF, _elem)
 235 #define type_pf_telem           TOKEN(TYPE, PF, _telem)
 236 #define type_pf_data_timeout    TOKEN(TYPE, PF, _data_timeout)
 237 #define type_pf_data_expired    TOKEN(TYPE, PF, _data_expired)
 238 #define type_pf_data_timeout_set TOKEN(TYPE, PF, _data_timeout_set)
 239
 240 #define type_pf_elem_add        TOKEN(TYPE, PF, _elem_add)
 241 #define type_pf_add             TOKEN(TYPE, PF, _add)
 242 #define type_pf_del             TOKEN(TYPE, PF, _del)
 243 #define type_pf_test_cidrs      TOKEN(TYPE, PF, _test_cidrs)
 244 #define type_pf_test            TOKEN(TYPE, PF, _test)
 245
 246 #define type_pf_elem_tadd       TOKEN(TYPE, PF, _elem_tadd)
 247 #define type_pf_del_telem       TOKEN(TYPE, PF, _ahash_del_telem)
 248 #define type_pf_expire          TOKEN(TYPE, PF, _expire)
 249 #define type_pf_tadd            TOKEN(TYPE, PF, _tadd)
 250 #define type_pf_tdel            TOKEN(TYPE, PF, _tdel)
 251 #define type_pf_ttest_cidrs     TOKEN(TYPE, PF, _ahash_ttest_cidrs)
 252 #define type_pf_ttest           TOKEN(TYPE, PF, _ahash_ttest)
 253
 254 #define type_pf_resize          TOKEN(TYPE, PF, _resize)
 255 #define type_pf_tresize         TOKEN(TYPE, PF, _tresize)
 256 #define type_pf_flush           ip_set_hash_flush
 257 #define type_pf_destroy         ip_set_hash_destroy
 258 #define type_pf_head            TOKEN(TYPE, PF, _head)
 259 #define type_pf_list            TOKEN(TYPE, PF, _list)
 260 #define type_pf_tlist           TOKEN(TYPE, PF, _tlist)
 261 #define type_pf_same_set        TOKEN(TYPE, PF, _same_set)
 262 #define type_pf_kadt            TOKEN(TYPE, PF, _kadt)
 263 #define type_pf_uadt            TOKEN(TYPE, PF, _uadt)
 264 #define type_pf_gc              TOKEN(TYPE, PF, _gc)
 265 #define type_pf_gc_init         TOKEN(TYPE, PF, _gc_init)
 266 #define type_pf_variant         TOKEN(TYPE, PF, _variant)
 267 #define type_pf_tvariant        TOKEN(TYPE, PF, _tvariant)
 268
 269 /* Flavour without timeout */
 270
 271 /* Get the ith element from the array block n */
 272 #define ahash_data(n, i)        \
 273         ((struct type_pf_elem *)((n)->value) + (i))
 274
 275 /* Add an element to the hash table when resizing the set:
 276  * we spare the maintenance of the internal counters. */
 277 static int
 278 type_pf_elem_add(struct hbucket *n, const struct type_pf_elem *value)
 279 {
 280         if (n->pos >= n->size) {
 281                 void *tmp;
 282
 283                 if (n->size >= AHASH_MAX_SIZE)
 284                         /* Trigger rehashing */
 285                         return -EAGAIN;
 286
 287                 tmp = kzalloc((n->size + AHASH_INIT_SIZE)
 288                               * sizeof(struct type_pf_elem),
 289                               GFP_ATOMIC);
 290                 if (!tmp)
 291                         return -ENOMEM;
 292                 if (n->size) {
 293                         memcpy(tmp, n->value,
 294                                sizeof(struct type_pf_elem) * n->size);
 295                         kfree(n->value);
 296                 }
 297                 n->value = tmp;
 298                 n->size += AHASH_INIT_SIZE;
 299         }
 300         type_pf_data_copy(ahash_data(n, n->pos++), value);
 301         return 0;
 302 }
 303
 304 /* Resize a hash: create a new hash table with doubling the hashsize
 305  * and inserting the elements to it. Repeat until we succeed or
 306  * fail due to memory pressures. */
 307 static int
 308 type_pf_resize(struct ip_set *set, bool retried)
 309 {
 310         struct ip_set_hash *h = set->data;
 311         struct htable *t, *orig = h->table;
 312         u8 htable_bits = orig->htable_bits;
 313         const struct type_pf_elem *data;
 314         struct hbucket *n, *m;
 315         u32 i, j;
 316         int ret;
 317
 318 retry:
 319         ret = 0;
 320         htable_bits++;
 321         pr_debug("attempt to resize set %s from %u to %u, t %p\n",
 322                  set->name, orig->htable_bits, htable_bits, orig);
 323         if (!htable_bits)
 324                 /* In case we have plenty of memory :-) */
 325                 return -IPSET_ERR_HASH_FULL;
 326         t = ip_set_alloc(sizeof(*t)
 327                          + jhash_size(htable_bits) * sizeof(struct hbucket));
 328         if (!t)
 329                 return -ENOMEM;
 330         t->htable_bits = htable_bits;
 331
 332         read_lock_bh(&set->lock);
 333         for (i = 0; i < jhash_size(orig->htable_bits); i++) {
 334                 n = hbucket(orig, i);
 335                 for (j = 0; j < n->pos; j++) {
 336                         data = ahash_data(n, j);
 337                         m = hbucket(t, HKEY(data, h->initval, htable_bits));
 338                         ret = type_pf_elem_add(m, data);
 339                         if (ret < 0) {
 340                                 read_unlock_bh(&set->lock);
 341                                 ahash_destroy(t);
 342                                 if (ret == -EAGAIN)
 343                                         goto retry;
 344                                 return ret;
 345                         }
 346                 }
 347         }
 348
 349         rcu_assign_pointer(h->table, t);
 350         read_unlock_bh(&set->lock);
 351
 352         /* Give time to other readers of the set */
 353         synchronize_rcu_bh();
 354
 355         pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
 356                  orig->htable_bits, orig, t->htable_bits, t);
 357         ahash_destroy(orig);
 358
 359         return 0;
 360 }
 361
 362 static void
 363 type_pf_data_next(struct ip_set_hash *h, const struct type_pf_elem *d);
 364
 365 /* Add an element to a hash and update the internal counters when succeeded,
 366  * otherwise report the proper error code. */
 367 static int
 368 type_pf_add(struct ip_set *set, void *value, u32 timeout, u32 flags)
 369 {
 370         struct ip_set_hash *h = set->data;
 371         struct htable *t;
 372         const struct type_pf_elem *d = value;
 373         struct hbucket *n;
 374         int i, ret = 0;
 375         u32 key;
 376
 377         if (h->elements >= h->maxelem)
 378                 return -IPSET_ERR_HASH_FULL;
 379
 380         rcu_read_lock_bh();
 381         t = rcu_dereference_bh(h->table);
 382         key = HKEY(value, h->initval, t->htable_bits);
 383         n = hbucket(t, key);
 384         for (i = 0; i < n->pos; i++)
 385                 if (type_pf_data_equal(ahash_data(n, i), d)) {
 386                         ret = -IPSET_ERR_EXIST;
 387                         goto out;
 388                 }
 389
 390         ret = type_pf_elem_add(n, value);
 391         if (ret != 0) {
 392                 if (ret == -EAGAIN)
 393                         type_pf_data_next(h, d);
 394                 goto out;
 395         }
 396
 397 #ifdef IP_SET_HASH_WITH_NETS
 398         add_cidr(h, d->cidr, HOST_MASK);
 399 #endif
 400         h->elements++;
 401 out:
 402         rcu_read_unlock_bh();
 403         return ret;
 404 }
 405
 406 /* Delete an element from the hash: swap it with the last element
 407  * and free up space if possible.
 408  */
 409 static int
 410 type_pf_del(struct ip_set *set, void *value, u32 timeout, u32 flags)
 411 {
 412         struct ip_set_hash *h = set->data;
 413         struct htable *t = h->table;
 414         const struct type_pf_elem *d = value;
 415         struct hbucket *n;
 416         int i;
 417         struct type_pf_elem *data;
 418         u32 key;
 419
 420         key = HKEY(value, h->initval, t->htable_bits);
 421         n = hbucket(t, key);
 422         for (i = 0; i < n->pos; i++) {
 423                 data = ahash_data(n, i);
 424                 if (!type_pf_data_equal(data, d))
 425                         continue;
 426                 if (i != n->pos - 1)
 427                         /* Not last one */
 428                         type_pf_data_copy(data, ahash_data(n, n->pos - 1));
 429
 430                 n->pos--;
 431                 h->elements--;
 432 #ifdef IP_SET_HASH_WITH_NETS
 433                 del_cidr(h, d->cidr, HOST_MASK);
 434 #endif
 435                 if (n->pos + AHASH_INIT_SIZE < n->size) {
 436                         void *tmp = kzalloc((n->size - AHASH_INIT_SIZE)
 437                                             * sizeof(struct type_pf_elem),
 438                                             GFP_ATOMIC);
 439                         if (!tmp)
 440                                 return 0;
 441                         n->size -= AHASH_INIT_SIZE;
 442                         memcpy(tmp, n->value,
 443                                n->size * sizeof(struct type_pf_elem));
 444                         kfree(n->value);
 445                         n->value = tmp;
 446                 }
 447                 return 0;
 448         }
 449
 450         return -IPSET_ERR_EXIST;
 451 }
 452
 453 #ifdef IP_SET_HASH_WITH_NETS
 454
 455 /* Special test function which takes into account the different network
 456  * sizes added to the set */
 457 static int
 458 type_pf_test_cidrs(struct ip_set *set, struct type_pf_elem *d, u32 timeout)
 459 {
 460         struct ip_set_hash *h = set->data;
 461         struct htable *t = h->table;
 462         struct hbucket *n;
 463         const struct type_pf_elem *data;
 464         int i, j = 0;
 465         u32 key;
 466         u8 host_mask = SET_HOST_MASK(set->family);
 467
 468         pr_debug("test by nets\n");
 469         for (; j < host_mask && h->nets[j].cidr; j++) {
 470                 type_pf_data_netmask(d, h->nets[j].cidr);
 471                 key = HKEY(d, h->initval, t->htable_bits);
 472                 n = hbucket(t, key);
 473                 for (i = 0; i < n->pos; i++) {
 474                         data = ahash_data(n, i);
 475                         if (type_pf_data_equal(data, d))
 476                                 return 1;
 477                 }
 478         }
 479         return 0;
 480 }
 481 #endif
 482
 483 /* Test whether the element is added to the set */
 484 static int
 485 type_pf_test(struct ip_set *set, void *value, u32 timeout, u32 flags)
 486 {
 487         struct ip_set_hash *h = set->data;
 488         struct htable *t = h->table;
 489         struct type_pf_elem *d = value;
 490         struct hbucket *n;
 491         const struct type_pf_elem *data;
 492         int i;
 493         u32 key;
 494
 495 #ifdef IP_SET_HASH_WITH_NETS
 496         /* If we test an IP address and not a network address,
 497          * try all possible network sizes */
 498         if (d->cidr == SET_HOST_MASK(set->family))
 499                 return type_pf_test_cidrs(set, d, timeout);
 500 #endif
 501
 502         key = HKEY(d, h->initval, t->htable_bits);
 503         n = hbucket(t, key);
 504         for (i = 0; i < n->pos; i++) {
 505                 data = ahash_data(n, i);
 506                 if (type_pf_data_equal(data, d))
 507                         return 1;
 508         }
 509         return 0;
 510 }
 511
 512 /* Reply a HEADER request: fill out the header part of the set */
 513 static int
 514 type_pf_head(struct ip_set *set, struct sk_buff *skb)
 515 {
 516         const struct ip_set_hash *h = set->data;
 517         struct nlattr *nested;
 518         size_t memsize;
 519
 520         read_lock_bh(&set->lock);
 521         memsize = ahash_memsize(h, with_timeout(h->timeout)
 522                                         ? sizeof(struct type_pf_telem)
 523                                         : sizeof(struct type_pf_elem),
 524                                 set->family == AF_INET ? 32 : 128);
 525         read_unlock_bh(&set->lock);
 526
 527         nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
 528         if (!nested)
 529                 goto nla_put_failure;
 530         NLA_PUT_NET32(skb, IPSET_ATTR_HASHSIZE,
 531                       htonl(jhash_size(h->table->htable_bits)));
 532         NLA_PUT_NET32(skb, IPSET_ATTR_MAXELEM, htonl(h->maxelem));
 533 #ifdef IP_SET_HASH_WITH_NETMASK
 534         if (h->netmask != HOST_MASK)
 535                 NLA_PUT_U8(skb, IPSET_ATTR_NETMASK, h->netmask);
 536 #endif
 537         NLA_PUT_NET32(skb, IPSET_ATTR_REFERENCES, htonl(set->ref - 1));
 538         NLA_PUT_NET32(skb, IPSET_ATTR_MEMSIZE, htonl(memsize));
 539         if (with_timeout(h->timeout))
 540                 NLA_PUT_NET32(skb, IPSET_ATTR_TIMEOUT, htonl(h->timeout));
 541         ipset_nest_end(skb, nested);
 542
 543         return 0;
 544 nla_put_failure:
 545         return -EMSGSIZE;
 546 }
 547
 548 /* Reply a LIST/SAVE request: dump the elements of the specified set */
 549 static int
 550 type_pf_list(const struct ip_set *set,
 551              struct sk_buff *skb, struct netlink_callback *cb)
 552 {
 553         const struct ip_set_hash *h = set->data;
 554         const struct htable *t = h->table;
 555         struct nlattr *atd, *nested;
 556         const struct hbucket *n;
 557         const struct type_pf_elem *data;
 558         u32 first = cb->args[2];
 559         /* We assume that one hash bucket fills into one page */
 560         void *incomplete;
 561         int i;
 562
 563         atd = ipset_nest_start(skb, IPSET_ATTR_ADT);
 564         if (!atd)
 565                 return -EMSGSIZE;
 566         pr_debug("list hash set %s\n", set->name);
 567         for (; cb->args[2] < jhash_size(t->htable_bits); cb->args[2]++) {
 568                 incomplete = skb_tail_pointer(skb);
 569                 n = hbucket(t, cb->args[2]);
 570                 pr_debug("cb->args[2]: %lu, t %p n %p\n", cb->args[2], t, n);
 571                 for (i = 0; i < n->pos; i++) {
 572                         data = ahash_data(n, i);
 573                         pr_debug("list hash %lu hbucket %p i %u, data %p\n",
 574                                  cb->args[2], n, i, data);
 575                         nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
 576                         if (!nested) {
 577                                 if (cb->args[2] == first) {
 578                                         nla_nest_cancel(skb, atd);
 579                                         return -EMSGSIZE;
 580                                 } else
 581                                         goto nla_put_failure;
 582                         }
 583                         if (type_pf_data_list(skb, data))
 584                                 goto nla_put_failure;
 585                         ipset_nest_end(skb, nested);
 586                 }
 587         }
 588         ipset_nest_end(skb, atd);
 589         /* Set listing finished */
 590         cb->args[2] = 0;
 591
 592         return 0;
 593
 594 nla_put_failure:
 595         nlmsg_trim(skb, incomplete);
 596         ipset_nest_end(skb, atd);
 597         if (unlikely(first == cb->args[2])) {
 598                 pr_warning("Can't list set %s: one bucket does not fit into "
 599                            "a message. Please report it!\n", set->name);
 600                 cb->args[2] = 0;
 601                 return -EMSGSIZE;
 602         }
 603         return 0;
 604 }
 605
 606 static int
 607 type_pf_kadt(struct ip_set *set, const struct sk_buff * skb,
 608              const struct xt_action_param *par,
 609              enum ipset_adt adt, const struct ip_set_adt_opt *opt);
 610 static int
 611 type_pf_uadt(struct ip_set *set, struct nlattr *tb[],
 612              enum ipset_adt adt, u32 *lineno, u32 flags, bool retried);
 613
 614 static const struct ip_set_type_variant type_pf_variant = {
 615         .kadt   = type_pf_kadt,
 616         .uadt   = type_pf_uadt,
 617         .adt    = {
 618                 [IPSET_ADD] = type_pf_add,
 619                 [IPSET_DEL] = type_pf_del,
 620                 [IPSET_TEST] = type_pf_test,
 621         },
 622         .destroy = type_pf_destroy,
 623         .flush  = type_pf_flush,
 624         .head   = type_pf_head,
 625         .list   = type_pf_list,
 626         .resize = type_pf_resize,
 627         .same_set = type_pf_same_set,
 628 };
 629
 630 /* Flavour with timeout support */
 631
 632 #define ahash_tdata(n, i) \
 633         (struct type_pf_elem *)((struct type_pf_telem *)((n)->value) + (i))
 634
 635 static inline u32
 636 type_pf_data_timeout(const struct type_pf_elem *data)
 637 {
 638         const struct type_pf_telem *tdata =
 639                 (const struct type_pf_telem *) data;
 640
 641         return tdata->timeout;
 642 }
 643
 644 static inline bool
 645 type_pf_data_expired(const struct type_pf_elem *data)
 646 {
 647         const struct type_pf_telem *tdata =
 648                 (const struct type_pf_telem *) data;
 649
 650         return ip_set_timeout_expired(tdata->timeout);
 651 }
 652
 653 static inline void
 654 type_pf_data_timeout_set(struct type_pf_elem *data, u32 timeout)
 655 {
 656         struct type_pf_telem *tdata = (struct type_pf_telem *) data;
 657
 658         tdata->timeout = ip_set_timeout_set(timeout);
 659 }
 660
 661 static int
 662 type_pf_elem_tadd(struct hbucket *n, const struct type_pf_elem *value,
 663                   u32 timeout)
 664 {
 665         struct type_pf_elem *data;
 666
 667         if (n->pos >= n->size) {
 668                 void *tmp;
 669
 670                 if (n->size >= AHASH_MAX_SIZE)
 671                         /* Trigger rehashing */
 672                         return -EAGAIN;
 673
 674                 tmp = kzalloc((n->size + AHASH_INIT_SIZE)
 675                               * sizeof(struct type_pf_telem),
 676                               GFP_ATOMIC);
 677                 if (!tmp)
 678                         return -ENOMEM;
 679                 if (n->size) {
 680                         memcpy(tmp, n->value,
 681                                sizeof(struct type_pf_telem) * n->size);
 682                         kfree(n->value);
 683                 }
 684                 n->value = tmp;
 685                 n->size += AHASH_INIT_SIZE;
 686         }
 687         data = ahash_tdata(n, n->pos++);
 688         type_pf_data_copy(data, value);
 689         type_pf_data_timeout_set(data, timeout);
 690         return 0;
 691 }
 692
 693 /* Delete expired elements from the hashtable */
 694 static void
 695 type_pf_expire(struct ip_set_hash *h)
 696 {
 697         struct htable *t = h->table;
 698         struct hbucket *n;
 699         struct type_pf_elem *data;
 700         u32 i;
 701         int j;
 702
 703         for (i = 0; i < jhash_size(t->htable_bits); i++) {
 704                 n = hbucket(t, i);
 705                 for (j = 0; j < n->pos; j++) {
 706                         data = ahash_tdata(n, j);
 707                         if (type_pf_data_expired(data)) {
 708                                 pr_debug("expired %u/%u\n", i, j);
 709 #ifdef IP_SET_HASH_WITH_NETS
 710                                 del_cidr(h, data->cidr, HOST_MASK);
 711 #endif
 712                                 if (j != n->pos - 1)
 713                                         /* Not last one */
 714                                         type_pf_data_copy(data,
 715                                                 ahash_tdata(n, n->pos - 1));
 716                                 n->pos--;
 717                                 h->elements--;
 718                         }
 719                 }
 720                 if (n->pos + AHASH_INIT_SIZE < n->size) {
 721                         void *tmp = kzalloc((n->size - AHASH_INIT_SIZE)
 722                                             * sizeof(struct type_pf_telem),
 723                                             GFP_ATOMIC);
 724                         if (!tmp)
 725                                 /* Still try to delete expired elements */
 726                                 continue;
 727                         n->size -= AHASH_INIT_SIZE;
 728                         memcpy(tmp, n->value,
 729                                n->size * sizeof(struct type_pf_telem));
 730                         kfree(n->value);
 731                         n->value = tmp;
 732                 }
 733         }
 734 }
 735
 736 static int
 737 type_pf_tresize(struct ip_set *set, bool retried)
 738 {
 739         struct ip_set_hash *h = set->data;
 740         struct htable *t, *orig = h->table;
 741         u8 htable_bits = orig->htable_bits;
 742         const struct type_pf_elem *data;
 743         struct hbucket *n, *m;
 744         u32 i, j;
 745         int ret;
 746
 747         /* Try to cleanup once */
 748         if (!retried) {
 749                 i = h->elements;
 750                 write_lock_bh(&set->lock);
 751                 type_pf_expire(set->data);
 752                 write_unlock_bh(&set->lock);
 753                 if (h->elements <  i)
 754                         return 0;
 755         }
 756
 757 retry:
 758         ret = 0;
 759         htable_bits++;
 760         if (!htable_bits)
 761                 /* In case we have plenty of memory :-) */
 762                 return -IPSET_ERR_HASH_FULL;
 763         t = ip_set_alloc(sizeof(*t)
 764                          + jhash_size(htable_bits) * sizeof(struct hbucket));
 765         if (!t)
 766                 return -ENOMEM;
 767         t->htable_bits = htable_bits;
 768
 769         read_lock_bh(&set->lock);
 770         for (i = 0; i < jhash_size(orig->htable_bits); i++) {
 771                 n = hbucket(orig, i);
 772                 for (j = 0; j < n->pos; j++) {
 773                         data = ahash_tdata(n, j);
 774                         m = hbucket(t, HKEY(data, h->initval, htable_bits));
 775                         ret = type_pf_elem_tadd(m, data,
 776                                                 type_pf_data_timeout(data));
 777                         if (ret < 0) {
 778                                 read_unlock_bh(&set->lock);
 779                                 ahash_destroy(t);
 780                                 if (ret == -EAGAIN)
 781                                         goto retry;
 782                                 return ret;
 783                         }
 784                 }
 785         }
 786
 787         rcu_assign_pointer(h->table, t);
 788         read_unlock_bh(&set->lock);
 789
 790         /* Give time to other readers of the set */
 791         synchronize_rcu_bh();
 792
 793         ahash_destroy(orig);
 794
 795         return 0;
 796 }
 797
 798 static int
 799 type_pf_tadd(struct ip_set *set, void *value, u32 timeout, u32 flags)
 800 {
 801         struct ip_set_hash *h = set->data;
 802         struct htable *t = h->table;
 803         const struct type_pf_elem *d = value;
 804         struct hbucket *n;
 805         struct type_pf_elem *data;
 806         int ret = 0, i, j = AHASH_MAX_SIZE + 1;
 807         bool flag_exist = flags & IPSET_FLAG_EXIST;
 808         u32 key;
 809
 810         if (h->elements >= h->maxelem)
 811                 /* FIXME: when set is full, we slow down here */
 812                 type_pf_expire(h);
 813         if (h->elements >= h->maxelem)
 814                 return -IPSET_ERR_HASH_FULL;
 815
 816         rcu_read_lock_bh();
 817         t = rcu_dereference_bh(h->table);
 818         key = HKEY(d, h->initval, t->htable_bits);
 819         n = hbucket(t, key);
 820         for (i = 0; i < n->pos; i++) {
 821                 data = ahash_tdata(n, i);
 822                 if (type_pf_data_equal(data, d)) {
 823                         if (type_pf_data_expired(data) || flag_exist)
 824                                 j = i;
 825                         else {
 826                                 ret = -IPSET_ERR_EXIST;
 827                                 goto out;
 828                         }
 829                 } else if (j == AHASH_MAX_SIZE + 1 &&
 830                            type_pf_data_expired(data))
 831                         j = i;
 832         }
 833         if (j != AHASH_MAX_SIZE + 1) {
 834                 data = ahash_tdata(n, j);
 835 #ifdef IP_SET_HASH_WITH_NETS
 836                 del_cidr(h, data->cidr, HOST_MASK);
 837                 add_cidr(h, d->cidr, HOST_MASK);
 838 #endif
 839                 type_pf_data_copy(data, d);
 840                 type_pf_data_timeout_set(data, timeout);
 841                 goto out;
 842         }
 843         ret = type_pf_elem_tadd(n, d, timeout);
 844         if (ret != 0) {
 845                 if (ret == -EAGAIN)
 846                         type_pf_data_next(h, d);
 847                 goto out;
 848         }
 849
 850 #ifdef IP_SET_HASH_WITH_NETS
 851         add_cidr(h, d->cidr, HOST_MASK);
 852 #endif
 853         h->elements++;
 854 out:
 855         rcu_read_unlock_bh();
 856         return ret;
 857 }
 858
 859 static int
 860 type_pf_tdel(struct ip_set *set, void *value, u32 timeout, u32 flags)
 861 {
 862         struct ip_set_hash *h = set->data;
 863         struct htable *t = h->table;
 864         const struct type_pf_elem *d = value;
 865         struct hbucket *n;
 866         int i;
 867         struct type_pf_elem *data;
 868         u32 key;
 869
 870         key = HKEY(value, h->initval, t->htable_bits);
 871         n = hbucket(t, key);
 872         for (i = 0; i < n->pos; i++) {
 873                 data = ahash_tdata(n, i);
 874                 if (!type_pf_data_equal(data, d))
 875                         continue;
 876                 if (type_pf_data_expired(data))
 877                         return -IPSET_ERR_EXIST;
 878                 if (i != n->pos - 1)
 879                         /* Not last one */
 880                         type_pf_data_copy(data, ahash_tdata(n, n->pos - 1));
 881
 882                 n->pos--;
 883                 h->elements--;
 884 #ifdef IP_SET_HASH_WITH_NETS
 885                 del_cidr(h, d->cidr, HOST_MASK);
 886 #endif
 887                 if (n->pos + AHASH_INIT_SIZE < n->size) {
 888                         void *tmp = kzalloc((n->size - AHASH_INIT_SIZE)
 889                                             * sizeof(struct type_pf_telem),
 890                                             GFP_ATOMIC);
 891                         if (!tmp)
 892                                 return 0;
 893                         n->size -= AHASH_INIT_SIZE;
 894                         memcpy(tmp, n->value,
 895                                n->size * sizeof(struct type_pf_telem));
 896                         kfree(n->value);
 897                         n->value = tmp;
 898                 }
 899                 return 0;
 900         }
 901
 902         return -IPSET_ERR_EXIST;
 903 }
 904
 905 #ifdef IP_SET_HASH_WITH_NETS
 906 static int
 907 type_pf_ttest_cidrs(struct ip_set *set, struct type_pf_elem *d, u32 timeout)
 908 {
 909         struct ip_set_hash *h = set->data;
 910         struct htable *t = h->table;
 911         struct type_pf_elem *data;
 912         struct hbucket *n;
 913         int i, j = 0;
 914         u32 key;
 915         u8 host_mask = SET_HOST_MASK(set->family);
 916
 917         for (; j < host_mask && h->nets[j].cidr; j++) {
 918                 type_pf_data_netmask(d, h->nets[j].cidr);
 919                 key = HKEY(d, h->initval, t->htable_bits);
 920                 n = hbucket(t, key);
 921                 for (i = 0; i < n->pos; i++) {
 922                         data = ahash_tdata(n, i);
 923                         if (type_pf_data_equal(data, d))
 924                                 return !type_pf_data_expired(data);
 925                 }
 926         }
 927         return 0;
 928 }
 929 #endif
 930
 931 static int
 932 type_pf_ttest(struct ip_set *set, void *value, u32 timeout, u32 flags)
 933 {
 934         struct ip_set_hash *h = set->data;
 935         struct htable *t = h->table;
 936         struct type_pf_elem *data, *d = value;
 937         struct hbucket *n;
 938         int i;
 939         u32 key;
 940
 941 #ifdef IP_SET_HASH_WITH_NETS
 942         if (d->cidr == SET_HOST_MASK(set->family))
 943                 return type_pf_ttest_cidrs(set, d, timeout);
 944 #endif
 945         key = HKEY(d, h->initval, t->htable_bits);
 946         n = hbucket(t, key);
 947         for (i = 0; i < n->pos; i++) {
 948                 data = ahash_tdata(n, i);
 949                 if (type_pf_data_equal(data, d))
 950                         return !type_pf_data_expired(data);
 951         }
 952         return 0;
 953 }
 954
 955 static int
 956 type_pf_tlist(const struct ip_set *set,
 957               struct sk_buff *skb, struct netlink_callback *cb)
 958 {
 959         const struct ip_set_hash *h = set->data;
 960         const struct htable *t = h->table;
 961         struct nlattr *atd, *nested;
 962         const struct hbucket *n;
 963         const struct type_pf_elem *data;
 964         u32 first = cb->args[2];
 965         /* We assume that one hash bucket fills into one page */
 966         void *incomplete;
 967         int i;
 968
 969         atd = ipset_nest_start(skb, IPSET_ATTR_ADT);
 970         if (!atd)
 971                 return -EMSGSIZE;
 972         for (; cb->args[2] < jhash_size(t->htable_bits); cb->args[2]++) {
 973                 incomplete = skb_tail_pointer(skb);
 974                 n = hbucket(t, cb->args[2]);
 975                 for (i = 0; i < n->pos; i++) {
 976                         data = ahash_tdata(n, i);
 977                         pr_debug("list %p %u\n", n, i);
 978                         if (type_pf_data_expired(data))
 979                                 continue;
 980                         pr_debug("do list %p %u\n", n, i);
 981                         nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
 982                         if (!nested) {
 983                                 if (cb->args[2] == first) {
 984                                         nla_nest_cancel(skb, atd);
 985                                         return -EMSGSIZE;
 986                                 } else
 987                                         goto nla_put_failure;
 988                         }
 989                         if (type_pf_data_tlist(skb, data))
 990                                 goto nla_put_failure;
 991                         ipset_nest_end(skb, nested);
 992                 }
 993         }
 994         ipset_nest_end(skb, atd);
 995         /* Set listing finished */
 996         cb->args[2] = 0;
 997
 998         return 0;
 999
1000 nla_put_failure:
1001         nlmsg_trim(skb, incomplete);
1002         ipset_nest_end(skb, atd);
1003         if (unlikely(first == cb->args[2])) {
1004                 pr_warning("Can't list set %s: one bucket does not fit into "
1005                            "a message. Please report it!\n", set->name);
1006                 cb->args[2] = 0;
1007                 return -EMSGSIZE;
1008         }
1009         return 0;
1010 }
1011
1012 static const struct ip_set_type_variant type_pf_tvariant = {
1013         .kadt   = type_pf_kadt,
1014         .uadt   = type_pf_uadt,
1015         .adt    = {
1016                 [IPSET_ADD] = type_pf_tadd,
1017                 [IPSET_DEL] = type_pf_tdel,
1018                 [IPSET_TEST] = type_pf_ttest,
1019         },
1020         .destroy = type_pf_destroy,
1021         .flush  = type_pf_flush,
1022         .head   = type_pf_head,
1023         .list   = type_pf_tlist,
1024         .resize = type_pf_tresize,
1025         .same_set = type_pf_same_set,
1026 };
1027
1028 static void
1029 type_pf_gc(unsigned long ul_set)
1030 {
1031         struct ip_set *set = (struct ip_set *) ul_set;
1032         struct ip_set_hash *h = set->data;
1033
1034         pr_debug("called\n");
1035         write_lock_bh(&set->lock);
1036         type_pf_expire(h);
1037         write_unlock_bh(&set->lock);
1038
1039         h->gc.expires = jiffies + IPSET_GC_PERIOD(h->timeout) * HZ;
1040         add_timer(&h->gc);
1041 }
1042
1043 static void
1044 type_pf_gc_init(struct ip_set *set)
1045 {
1046         struct ip_set_hash *h = set->data;
1047
1048         init_timer(&h->gc);
1049         h->gc.data = (unsigned long) set;
1050         h->gc.function = type_pf_gc;
1051         h->gc.expires = jiffies + IPSET_GC_PERIOD(h->timeout) * HZ;
1052         add_timer(&h->gc);
1053         pr_debug("gc initialized, run in every %u\n",
1054                  IPSET_GC_PERIOD(h->timeout));
1055 }
1056
1057 #undef type_pf_data_equal
1058 #undef type_pf_data_isnull
1059 #undef type_pf_data_copy
1060 #undef type_pf_data_zero_out
1061 #undef type_pf_data_list
1062 #undef type_pf_data_tlist
1063
1064 #undef type_pf_elem
1065 #undef type_pf_telem
1066 #undef type_pf_data_timeout
1067 #undef type_pf_data_expired
1068 #undef type_pf_data_netmask
1069 #undef type_pf_data_timeout_set
1070
1071 #undef type_pf_elem_add
1072 #undef type_pf_add
1073 #undef type_pf_del
1074 #undef type_pf_test_cidrs
1075 #undef type_pf_test
1076
1077 #undef type_pf_elem_tadd
1078 #undef type_pf_expire
1079 #undef type_pf_tadd
1080 #undef type_pf_tdel
1081 #undef type_pf_ttest_cidrs
1082 #undef type_pf_ttest
1083
1084 #undef type_pf_resize
1085 #undef type_pf_tresize
1086 #undef type_pf_flush
1087 #undef type_pf_destroy
1088 #undef type_pf_head
1089 #undef type_pf_list
1090 #undef type_pf_tlist
1091 #undef type_pf_same_set
1092 #undef type_pf_kadt
1093 #undef type_pf_uadt
1094 #undef type_pf_gc
1095 #undef type_pf_gc_init
1096 #undef type_pf_variant
1097 #undef type_pf_tvariant