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