net/netfilter/ipset/ip_set_hash

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
tune_ahash_max
htable_size
htable_bits
mtype_add_cidr
mtype_del_cidr
mtype_ahash_memsize
mtype_ext_cleanup
mtype_flush
mtype_ahash_destroy
mtype_destroy
mtype_same_set
mtype_gc_do
mtype_gc
mtype_gc_init
mtype_resize
mtype_ext_size
mtype_add
mtype_del
mtype_data_match
mtype_test_cidrs
mtype_test
mtype_head
mtype_uref
mtype_list
   1 /* SPDX-License-Identifier: GPL-2.0-only */
   2 /* Copyright (C) 2013 Jozsef Kadlecsik <kadlec@netfilter.org> */
   3 
   4 #ifndef _IP_SET_HASH_GEN_H
   5 #define _IP_SET_HASH_GEN_H
   6 
   7 #include <linux/rcupdate.h>
   8 #include <linux/jhash.h>
   9 #include <linux/types.h>
  10 #include <linux/netfilter/nfnetlink.h>
  11 #include <linux/netfilter/ipset/ip_set.h>
  12 
  13 #define __ipset_dereference(p)          \
  14         rcu_dereference_protected(p, 1)
  15 #define ipset_dereference_nfnl(p)       \
  16         rcu_dereference_protected(p,    \
  17                 lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET))
  18 #define ipset_dereference_set(p, set)   \
  19         rcu_dereference_protected(p,    \
  20                 lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET) || \
  21                 lockdep_is_held(&(set)->lock))
  22 #define ipset_dereference_bh_nfnl(p)    \
  23         rcu_dereference_bh_check(p,     \
  24                 lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET))
  25 
  26 /* Hashing which uses arrays to resolve clashing. The hash table is resized
  27  * (doubled) when searching becomes too long.
  28  * Internally jhash is used with the assumption that the size of the
  29  * stored data is a multiple of sizeof(u32).
  30  *
  31  * Readers and resizing
  32  *
  33  * Resizing can be triggered by userspace command only, and those
  34  * are serialized by the nfnl mutex. During resizing the set is
  35  * read-locked, so the only possible concurrent operations are
  36  * the kernel side readers. Those must be protected by proper RCU locking.
  37  */
  38 
  39 /* Number of elements to store in an initial array block */
  40 #define AHASH_INIT_SIZE                 4
  41 /* Max number of elements to store in an array block */
  42 #define AHASH_MAX_SIZE                  (3 * AHASH_INIT_SIZE)
  43 /* Max muber of elements in the array block when tuned */
  44 #define AHASH_MAX_TUNED                 64
  45 
  46 /* Max number of elements can be tuned */
  47 #ifdef IP_SET_HASH_WITH_MULTI
  48 #define AHASH_MAX(h)                    ((h)->ahash_max)
  49 
  50 static inline u8
  51 tune_ahash_max(u8 curr, u32 multi)
  52 {
  53         u32 n;
  54 
  55         if (multi < curr)
  56                 return curr;
  57 
  58         n = curr + AHASH_INIT_SIZE;
  59         /* Currently, at listing one hash bucket must fit into a message.
  60          * Therefore we have a hard limit here.
  61          */
  62         return n > curr && n <= AHASH_MAX_TUNED ? n : curr;
  63 }
  64 
  65 #define TUNE_AHASH_MAX(h, multi)        \
  66         ((h)->ahash_max = tune_ahash_max((h)->ahash_max, multi))
  67 #else
  68 #define AHASH_MAX(h)                    AHASH_MAX_SIZE
  69 #define TUNE_AHASH_MAX(h, multi)
  70 #endif
  71 
  72 /* A hash bucket */
  73 struct hbucket {
  74         struct rcu_head rcu;    /* for call_rcu */
  75         /* Which positions are used in the array */
  76         DECLARE_BITMAP(used, AHASH_MAX_TUNED);
  77         u8 size;                /* size of the array */
  78         u8 pos;                 /* position of the first free entry */
  79         unsigned char value[0]  /* the array of the values */
  80                 __aligned(__alignof__(u64));
  81 };
  82 
  83 /* Region size for locking == 2^HTABLE_REGION_BITS */
  84 #define HTABLE_REGION_BITS      10
  85 #define ahash_numof_locks(htable_bits)          \
  86         ((htable_bits) < HTABLE_REGION_BITS ? 1 \
  87                 : jhash_size((htable_bits) - HTABLE_REGION_BITS))
  88 #define ahash_sizeof_regions(htable_bits)               \
  89         (ahash_numof_locks(htable_bits) * sizeof(struct ip_set_region))
  90 #define ahash_region(n, htable_bits)            \
  91         ((n) % ahash_numof_locks(htable_bits))
  92 #define ahash_bucket_start(h,  htable_bits)     \
  93         ((htable_bits) < HTABLE_REGION_BITS ? 0 \
  94                 : (h) * jhash_size(HTABLE_REGION_BITS))
  95 #define ahash_bucket_end(h,  htable_bits)       \
  96         ((htable_bits) < HTABLE_REGION_BITS ? jhash_size(htable_bits)   \
  97                 : ((h) + 1) * jhash_size(HTABLE_REGION_BITS))
  98 
  99 struct htable_gc {
 100         struct delayed_work dwork;
 101         struct ip_set *set;     /* Set the gc belongs to */
 102         u32 region;             /* Last gc run position */
 103 };
 104 
 105 /* The hash table: the table size stored here in order to make resizing easy */
 106 struct htable {
 107         atomic_t ref;           /* References for resizing */
 108         atomic_t uref;          /* References for dumping and gc */
 109         u8 htable_bits;         /* size of hash table == 2^htable_bits */
 110         u32 maxelem;            /* Maxelem per region */
 111         struct ip_set_region *hregion;  /* Region locks and ext sizes */
 112         struct hbucket __rcu *bucket[0]; /* hashtable buckets */
 113 };
 114 
 115 #define hbucket(h, i)           ((h)->bucket[i])
 116 #define ext_size(n, dsize)      \
 117         (sizeof(struct hbucket) + (n) * (dsize))
 118 
 119 #ifndef IPSET_NET_COUNT
 120 #define IPSET_NET_COUNT         1
 121 #endif
 122 
 123 /* Book-keeping of the prefixes added to the set */
 124 struct net_prefixes {
 125         u32 nets[IPSET_NET_COUNT]; /* number of elements for this cidr */
 126         u8 cidr[IPSET_NET_COUNT];  /* the cidr value */
 127 };
 128 
 129 /* Compute the hash table size */
 130 static size_t
 131 htable_size(u8 hbits)
 132 {
 133         size_t hsize;
 134 
 135         /* We must fit both into u32 in jhash and size_t */
 136         if (hbits > 31)
 137                 return 0;
 138         hsize = jhash_size(hbits);
 139         if ((((size_t)-1) - sizeof(struct htable)) / sizeof(struct hbucket *)
 140             < hsize)
 141                 return 0;
 142 
 143         return hsize * sizeof(struct hbucket *) + sizeof(struct htable);
 144 }
 145 
 146 /* Compute htable_bits from the user input parameter hashsize */
 147 static u8
 148 htable_bits(u32 hashsize)
 149 {
 150         /* Assume that hashsize == 2^htable_bits */
 151         u8 bits = fls(hashsize - 1);
 152 
 153         if (jhash_size(bits) != hashsize)
 154                 /* Round up to the first 2^n value */
 155                 bits = fls(hashsize);
 156 
 157         return bits;
 158 }
 159 
 160 #ifdef IP_SET_HASH_WITH_NETS
 161 #if IPSET_NET_COUNT > 1
 162 #define __CIDR(cidr, i)         (cidr[i])
 163 #else
 164 #define __CIDR(cidr, i)         (cidr)
 165 #endif
 166 
 167 /* cidr + 1 is stored in net_prefixes to support /0 */
 168 #define NCIDR_PUT(cidr)         ((cidr) + 1)
 169 #define NCIDR_GET(cidr)         ((cidr) - 1)
 170 
 171 #ifdef IP_SET_HASH_WITH_NETS_PACKED
 172 /* When cidr is packed with nomatch, cidr - 1 is stored in the data entry */
 173 #define DCIDR_PUT(cidr)         ((cidr) - 1)
 174 #define DCIDR_GET(cidr, i)      (__CIDR(cidr, i) + 1)
 175 #else
 176 #define DCIDR_PUT(cidr)         (cidr)
 177 #define DCIDR_GET(cidr, i)      __CIDR(cidr, i)
 178 #endif
 179 
 180 #define INIT_CIDR(cidr, host_mask)      \
 181         DCIDR_PUT(((cidr) ? NCIDR_GET(cidr) : host_mask))
 182 
 183 #ifdef IP_SET_HASH_WITH_NET0
 184 /* cidr from 0 to HOST_MASK value and c = cidr + 1 */
 185 #define NLEN                    (HOST_MASK + 1)
 186 #define CIDR_POS(c)             ((c) - 1)
 187 #else
 188 /* cidr from 1 to HOST_MASK value and c = cidr + 1 */
 189 #define NLEN                    HOST_MASK
 190 #define CIDR_POS(c)             ((c) - 2)
 191 #endif
 192 
 193 #else
 194 #define NLEN                    0
 195 #endif /* IP_SET_HASH_WITH_NETS */
 196 
 197 #define SET_ELEM_EXPIRED(set, d)        \
 198         (SET_WITH_TIMEOUT(set) &&       \
 199          ip_set_timeout_expired(ext_timeout(d, set)))
 200 
 201 #endif /* _IP_SET_HASH_GEN_H */
 202 
 203 #ifndef MTYPE
 204 #error "MTYPE is not defined!"
 205 #endif
 206 
 207 #ifndef HTYPE
 208 #error "HTYPE is not defined!"
 209 #endif
 210 
 211 #ifndef HOST_MASK
 212 #error "HOST_MASK is not defined!"
 213 #endif
 214 
 215 /* Family dependent templates */
 216 
 217 #undef ahash_data
 218 #undef mtype_data_equal
 219 #undef mtype_do_data_match
 220 #undef mtype_data_set_flags
 221 #undef mtype_data_reset_elem
 222 #undef mtype_data_reset_flags
 223 #undef mtype_data_netmask
 224 #undef mtype_data_list
 225 #undef mtype_data_next
 226 #undef mtype_elem
 227 
 228 #undef mtype_ahash_destroy
 229 #undef mtype_ext_cleanup
 230 #undef mtype_add_cidr
 231 #undef mtype_del_cidr
 232 #undef mtype_ahash_memsize
 233 #undef mtype_flush
 234 #undef mtype_destroy
 235 #undef mtype_same_set
 236 #undef mtype_kadt
 237 #undef mtype_uadt
 238 
 239 #undef mtype_add
 240 #undef mtype_del
 241 #undef mtype_test_cidrs
 242 #undef mtype_test
 243 #undef mtype_uref
 244 #undef mtype_resize
 245 #undef mtype_ext_size
 246 #undef mtype_resize_ad
 247 #undef mtype_head
 248 #undef mtype_list
 249 #undef mtype_gc_do
 250 #undef mtype_gc
 251 #undef mtype_gc_init
 252 #undef mtype_variant
 253 #undef mtype_data_match
 254 
 255 #undef htype
 256 #undef HKEY
 257 
 258 #define mtype_data_equal        IPSET_TOKEN(MTYPE, _data_equal)
 259 #ifdef IP_SET_HASH_WITH_NETS
 260 #define mtype_do_data_match     IPSET_TOKEN(MTYPE, _do_data_match)
 261 #else
 262 #define mtype_do_data_match(d)  1
 263 #endif
 264 #define mtype_data_set_flags    IPSET_TOKEN(MTYPE, _data_set_flags)
 265 #define mtype_data_reset_elem   IPSET_TOKEN(MTYPE, _data_reset_elem)
 266 #define mtype_data_reset_flags  IPSET_TOKEN(MTYPE, _data_reset_flags)
 267 #define mtype_data_netmask      IPSET_TOKEN(MTYPE, _data_netmask)
 268 #define mtype_data_list         IPSET_TOKEN(MTYPE, _data_list)
 269 #define mtype_data_next         IPSET_TOKEN(MTYPE, _data_next)
 270 #define mtype_elem              IPSET_TOKEN(MTYPE, _elem)
 271 
 272 #define mtype_ahash_destroy     IPSET_TOKEN(MTYPE, _ahash_destroy)
 273 #define mtype_ext_cleanup       IPSET_TOKEN(MTYPE, _ext_cleanup)
 274 #define mtype_add_cidr          IPSET_TOKEN(MTYPE, _add_cidr)
 275 #define mtype_del_cidr          IPSET_TOKEN(MTYPE, _del_cidr)
 276 #define mtype_ahash_memsize     IPSET_TOKEN(MTYPE, _ahash_memsize)
 277 #define mtype_flush             IPSET_TOKEN(MTYPE, _flush)
 278 #define mtype_destroy           IPSET_TOKEN(MTYPE, _destroy)
 279 #define mtype_same_set          IPSET_TOKEN(MTYPE, _same_set)
 280 #define mtype_kadt              IPSET_TOKEN(MTYPE, _kadt)
 281 #define mtype_uadt              IPSET_TOKEN(MTYPE, _uadt)
 282 
 283 #define mtype_add               IPSET_TOKEN(MTYPE, _add)
 284 #define mtype_del               IPSET_TOKEN(MTYPE, _del)
 285 #define mtype_test_cidrs        IPSET_TOKEN(MTYPE, _test_cidrs)
 286 #define mtype_test              IPSET_TOKEN(MTYPE, _test)
 287 #define mtype_uref              IPSET_TOKEN(MTYPE, _uref)
 288 #define mtype_resize            IPSET_TOKEN(MTYPE, _resize)
 289 #define mtype_ext_size          IPSET_TOKEN(MTYPE, _ext_size)
 290 #define mtype_resize_ad         IPSET_TOKEN(MTYPE, _resize_ad)
 291 #define mtype_head              IPSET_TOKEN(MTYPE, _head)
 292 #define mtype_list              IPSET_TOKEN(MTYPE, _list)
 293 #define mtype_gc_do             IPSET_TOKEN(MTYPE, _gc_do)
 294 #define mtype_gc                IPSET_TOKEN(MTYPE, _gc)
 295 #define mtype_gc_init           IPSET_TOKEN(MTYPE, _gc_init)
 296 #define mtype_variant           IPSET_TOKEN(MTYPE, _variant)
 297 #define mtype_data_match        IPSET_TOKEN(MTYPE, _data_match)
 298 
 299 #ifndef HKEY_DATALEN
 300 #define HKEY_DATALEN            sizeof(struct mtype_elem)
 301 #endif
 302 
 303 #define htype                   MTYPE
 304 
 305 #define HKEY(data, initval, htable_bits)                        \
 306 ({                                                              \
 307         const u32 *__k = (const u32 *)data;                     \
 308         u32 __l = HKEY_DATALEN / sizeof(u32);                   \
 309                                                                 \
 310         BUILD_BUG_ON(HKEY_DATALEN % sizeof(u32) != 0);          \
 311                                                                 \
 312         jhash2(__k, __l, initval) & jhash_mask(htable_bits);    \
 313 })
 314 
 315 /* The generic hash structure */
 316 struct htype {
 317         struct htable __rcu *table; /* the hash table */
 318         struct htable_gc gc;    /* gc workqueue */
 319         u32 maxelem;            /* max elements in the hash */
 320         u32 initval;            /* random jhash init value */
 321 #ifdef IP_SET_HASH_WITH_MARKMASK
 322         u32 markmask;           /* markmask value for mark mask to store */
 323 #endif
 324 #ifdef IP_SET_HASH_WITH_MULTI
 325         u8 ahash_max;           /* max elements in an array block */
 326 #endif
 327 #ifdef IP_SET_HASH_WITH_NETMASK
 328         u8 netmask;             /* netmask value for subnets to store */
 329 #endif
 330         struct list_head ad;    /* Resize add|del backlist */
 331         struct mtype_elem next; /* temporary storage for uadd */
 332 #ifdef IP_SET_HASH_WITH_NETS
 333         struct net_prefixes nets[NLEN]; /* book-keeping of prefixes */
 334 #endif
 335 };
 336 
 337 /* ADD|DEL entries saved during resize */
 338 struct mtype_resize_ad {
 339         struct list_head list;
 340         enum ipset_adt ad;      /* ADD|DEL element */
 341         struct mtype_elem d;    /* Element value */
 342         struct ip_set_ext ext;  /* Extensions for ADD */
 343         struct ip_set_ext mext; /* Target extensions for ADD */
 344         u32 flags;              /* Flags for ADD */
 345 };
 346 
 347 #ifdef IP_SET_HASH_WITH_NETS
 348 /* Network cidr size book keeping when the hash stores different
 349  * sized networks. cidr == real cidr + 1 to support /0.
 350  */
 351 static void
 352 mtype_add_cidr(struct ip_set *set, struct htype *h, u8 cidr, u8 n)
 353 {
 354         int i, j;
 355 
 356         spin_lock_bh(&set->lock);
 357         /* Add in increasing prefix order, so larger cidr first */
 358         for (i = 0, j = -1; i < NLEN && h->nets[i].cidr[n]; i++) {
 359                 if (j != -1) {
 360                         continue;
 361                 } else if (h->nets[i].cidr[n] < cidr) {
 362                         j = i;
 363                 } else if (h->nets[i].cidr[n] == cidr) {
 364                         h->nets[CIDR_POS(cidr)].nets[n]++;
 365                         goto unlock;
 366                 }
 367         }
 368         if (j != -1) {
 369                 for (; i > j; i--)
 370                         h->nets[i].cidr[n] = h->nets[i - 1].cidr[n];
 371         }
 372         h->nets[i].cidr[n] = cidr;
 373         h->nets[CIDR_POS(cidr)].nets[n] = 1;
 374 unlock:
 375         spin_unlock_bh(&set->lock);
 376 }
 377 
 378 static void
 379 mtype_del_cidr(struct ip_set *set, struct htype *h, u8 cidr, u8 n)
 380 {
 381         u8 i, j, net_end = NLEN - 1;
 382 
 383         spin_lock_bh(&set->lock);
 384         for (i = 0; i < NLEN; i++) {
 385                 if (h->nets[i].cidr[n] != cidr)
 386                         continue;
 387                 h->nets[CIDR_POS(cidr)].nets[n]--;
 388                 if (h->nets[CIDR_POS(cidr)].nets[n] > 0)
 389                         goto unlock;
 390                 for (j = i; j < net_end && h->nets[j].cidr[n]; j++)
 391                         h->nets[j].cidr[n] = h->nets[j + 1].cidr[n];
 392                 h->nets[j].cidr[n] = 0;
 393                 goto unlock;
 394         }
 395 unlock:
 396         spin_unlock_bh(&set->lock);
 397 }
 398 #endif
 399 
 400 /* Calculate the actual memory size of the set data */
 401 static size_t
 402 mtype_ahash_memsize(const struct htype *h, const struct htable *t)
 403 {
 404         return sizeof(*h) + sizeof(*t) + ahash_sizeof_regions(t->htable_bits);
 405 }
 406 
 407 /* Get the ith element from the array block n */
 408 #define ahash_data(n, i, dsize) \
 409         ((struct mtype_elem *)((n)->value + ((i) * (dsize))))
 410 
 411 static void
 412 mtype_ext_cleanup(struct ip_set *set, struct hbucket *n)
 413 {
 414         int i;
 415 
 416         for (i = 0; i < n->pos; i++)
 417                 if (test_bit(i, n->used))
 418                         ip_set_ext_destroy(set, ahash_data(n, i, set->dsize));
 419 }
 420 
 421 /* Flush a hash type of set: destroy all elements */
 422 static void
 423 mtype_flush(struct ip_set *set)
 424 {
 425         struct htype *h = set->data;
 426         struct htable *t;
 427         struct hbucket *n;
 428         u32 r, i;
 429 
 430         t = ipset_dereference_nfnl(h->table);
 431         for (r = 0; r < ahash_numof_locks(t->htable_bits); r++) {
 432                 spin_lock_bh(&t->hregion[r].lock);
 433                 for (i = ahash_bucket_start(r, t->htable_bits);
 434                      i < ahash_bucket_end(r, t->htable_bits); i++) {
 435                         n = __ipset_dereference(hbucket(t, i));
 436                         if (!n)
 437                                 continue;
 438                         if (set->extensions & IPSET_EXT_DESTROY)
 439                                 mtype_ext_cleanup(set, n);
 440                         /* FIXME: use slab cache */
 441                         rcu_assign_pointer(hbucket(t, i), NULL);
 442                         kfree_rcu(n, rcu);
 443                 }
 444                 t->hregion[r].ext_size = 0;
 445                 t->hregion[r].elements = 0;
 446                 spin_unlock_bh(&t->hregion[r].lock);
 447         }
 448 #ifdef IP_SET_HASH_WITH_NETS
 449         memset(h->nets, 0, sizeof(h->nets));
 450 #endif
 451 }
 452 
 453 /* Destroy the hashtable part of the set */
 454 static void
 455 mtype_ahash_destroy(struct ip_set *set, struct htable *t, bool ext_destroy)
 456 {
 457         struct hbucket *n;
 458         u32 i;
 459 
 460         for (i = 0; i < jhash_size(t->htable_bits); i++) {
 461                 n = __ipset_dereference(hbucket(t, i));
 462                 if (!n)
 463                         continue;
 464                 if (set->extensions & IPSET_EXT_DESTROY && ext_destroy)
 465                         mtype_ext_cleanup(set, n);
 466                 /* FIXME: use slab cache */
 467                 kfree(n);
 468         }
 469 
 470         ip_set_free(t->hregion);
 471         ip_set_free(t);
 472 }
 473 
 474 /* Destroy a hash type of set */
 475 static void
 476 mtype_destroy(struct ip_set *set)
 477 {
 478         struct htype *h = set->data;
 479         struct list_head *l, *lt;
 480 
 481         if (SET_WITH_TIMEOUT(set))
 482                 cancel_delayed_work_sync(&h->gc.dwork);
 483 
 484         mtype_ahash_destroy(set, ipset_dereference_nfnl(h->table), true);
 485         list_for_each_safe(l, lt, &h->ad) {
 486                 list_del(l);
 487                 kfree(l);
 488         }
 489         kfree(h);
 490 
 491         set->data = NULL;
 492 }
 493 
 494 static bool
 495 mtype_same_set(const struct ip_set *a, const struct ip_set *b)
 496 {
 497         const struct htype *x = a->data;
 498         const struct htype *y = b->data;
 499 
 500         /* Resizing changes htable_bits, so we ignore it */
 501         return x->maxelem == y->maxelem &&
 502                a->timeout == b->timeout &&
 503 #ifdef IP_SET_HASH_WITH_NETMASK
 504                x->netmask == y->netmask &&
 505 #endif
 506 #ifdef IP_SET_HASH_WITH_MARKMASK
 507                x->markmask == y->markmask &&
 508 #endif
 509                a->extensions == b->extensions;
 510 }
 511 
 512 static void
 513 mtype_gc_do(struct ip_set *set, struct htype *h, struct htable *t, u32 r)
 514 {
 515         struct hbucket *n, *tmp;
 516         struct mtype_elem *data;
 517         u32 i, j, d;
 518         size_t dsize = set->dsize;
 519 #ifdef IP_SET_HASH_WITH_NETS
 520         u8 k;
 521 #endif
 522         u8 htable_bits = t->htable_bits;
 523 
 524         spin_lock_bh(&t->hregion[r].lock);
 525         for (i = ahash_bucket_start(r, htable_bits);
 526              i < ahash_bucket_end(r, htable_bits); i++) {
 527                 n = __ipset_dereference(hbucket(t, i));
 528                 if (!n)
 529                         continue;
 530                 for (j = 0, d = 0; j < n->pos; j++) {
 531                         if (!test_bit(j, n->used)) {
 532                                 d++;
 533                                 continue;
 534                         }
 535                         data = ahash_data(n, j, dsize);
 536                         if (!ip_set_timeout_expired(ext_timeout(data, set)))
 537                                 continue;
 538                         pr_debug("expired %u/%u\n", i, j);
 539                         clear_bit(j, n->used);
 540                         smp_mb__after_atomic();
 541 #ifdef IP_SET_HASH_WITH_NETS
 542                         for (k = 0; k < IPSET_NET_COUNT; k++)
 543                                 mtype_del_cidr(set, h,
 544                                         NCIDR_PUT(DCIDR_GET(data->cidr, k)),
 545                                         k);
 546 #endif
 547                         t->hregion[r].elements--;
 548                         ip_set_ext_destroy(set, data);
 549                         d++;
 550                 }
 551                 if (d >= AHASH_INIT_SIZE) {
 552                         if (d >= n->size) {
 553                                 t->hregion[r].ext_size -=
 554                                         ext_size(n->size, dsize);
 555                                 rcu_assign_pointer(hbucket(t, i), NULL);
 556                                 kfree_rcu(n, rcu);
 557                                 continue;
 558                         }
 559                         tmp = kzalloc(sizeof(*tmp) +
 560                                 (n->size - AHASH_INIT_SIZE) * dsize,
 561                                 GFP_ATOMIC);
 562                         if (!tmp)
 563                                 /* Still try to delete expired elements. */
 564                                 continue;
 565                         tmp->size = n->size - AHASH_INIT_SIZE;
 566                         for (j = 0, d = 0; j < n->pos; j++) {
 567                                 if (!test_bit(j, n->used))
 568                                         continue;
 569                                 data = ahash_data(n, j, dsize);
 570                                 memcpy(tmp->value + d * dsize,
 571                                        data, dsize);
 572                                 set_bit(d, tmp->used);
 573                                 d++;
 574                         }
 575                         tmp->pos = d;
 576                         t->hregion[r].ext_size -=
 577                                 ext_size(AHASH_INIT_SIZE, dsize);
 578                         rcu_assign_pointer(hbucket(t, i), tmp);
 579                         kfree_rcu(n, rcu);
 580                 }
 581         }
 582         spin_unlock_bh(&t->hregion[r].lock);
 583 }
 584 
 585 static void
 586 mtype_gc(struct work_struct *work)
 587 {
 588         struct htable_gc *gc;
 589         struct ip_set *set;
 590         struct htype *h;
 591         struct htable *t;
 592         u32 r, numof_locks;
 593         unsigned int next_run;
 594 
 595         gc = container_of(work, struct htable_gc, dwork.work);
 596         set = gc->set;
 597         h = set->data;
 598 
 599         spin_lock_bh(&set->lock);
 600         t = ipset_dereference_set(h->table, set);
 601         atomic_inc(&t->uref);
 602         numof_locks = ahash_numof_locks(t->htable_bits);
 603         r = gc->region++;
 604         if (r >= numof_locks) {
 605                 r = gc->region = 0;
 606         }
 607         next_run = (IPSET_GC_PERIOD(set->timeout) * HZ) / numof_locks;
 608         if (next_run < HZ/10)
 609                 next_run = HZ/10;
 610         spin_unlock_bh(&set->lock);
 611 
 612         mtype_gc_do(set, h, t, r);
 613 
 614         if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
 615                 pr_debug("Table destroy after resize by expire: %p\n", t);
 616                 mtype_ahash_destroy(set, t, false);
 617         }
 618 
 619         queue_delayed_work(system_power_efficient_wq, &gc->dwork, next_run);
 620 
 621 }
 622 
 623 static void
 624 mtype_gc_init(struct htable_gc *gc)
 625 {
 626         INIT_DEFERRABLE_WORK(&gc->dwork, mtype_gc);
 627         queue_delayed_work(system_power_efficient_wq, &gc->dwork, HZ);
 628 }
 629 
 630 static int
 631 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
 632           struct ip_set_ext *mext, u32 flags);
 633 static int
 634 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
 635           struct ip_set_ext *mext, u32 flags);
 636 
 637 /* Resize a hash: create a new hash table with doubling the hashsize
 638  * and inserting the elements to it. Repeat until we succeed or
 639  * fail due to memory pressures.
 640  */
 641 static int
 642 mtype_resize(struct ip_set *set, bool retried)
 643 {
 644         struct htype *h = set->data;
 645         struct htable *t, *orig;
 646         u8 htable_bits;
 647         size_t dsize = set->dsize;
 648 #ifdef IP_SET_HASH_WITH_NETS
 649         u8 flags;
 650         struct mtype_elem *tmp;
 651 #endif
 652         struct mtype_elem *data;
 653         struct mtype_elem *d;
 654         struct hbucket *n, *m;
 655         struct list_head *l, *lt;
 656         struct mtype_resize_ad *x;
 657         u32 i, j, r, nr, key;
 658         int ret;
 659 
 660 #ifdef IP_SET_HASH_WITH_NETS
 661         tmp = kmalloc(dsize, GFP_KERNEL);
 662         if (!tmp)
 663                 return -ENOMEM;
 664 #endif
 665         orig = ipset_dereference_bh_nfnl(h->table);
 666         htable_bits = orig->htable_bits;
 667 
 668 retry:
 669         ret = 0;
 670         htable_bits++;
 671         if (!htable_bits) {
 672                 /* In case we have plenty of memory :-) */
 673                 pr_warn("Cannot increase the hashsize of set %s further\n",
 674                         set->name);
 675                 ret = -IPSET_ERR_HASH_FULL;
 676                 goto out;
 677         }
 678         t = ip_set_alloc(htable_size(htable_bits));
 679         if (!t) {
 680                 ret = -ENOMEM;
 681                 goto out;
 682         }
 683         t->hregion = ip_set_alloc(ahash_sizeof_regions(htable_bits));
 684         if (!t->hregion) {
 685                 kfree(t);
 686                 ret = -ENOMEM;
 687                 goto out;
 688         }
 689         t->htable_bits = htable_bits;
 690         t->maxelem = h->maxelem / ahash_numof_locks(htable_bits);
 691         for (i = 0; i < ahash_numof_locks(htable_bits); i++)
 692                 spin_lock_init(&t->hregion[i].lock);
 693 
 694         /* There can't be another parallel resizing,
 695          * but dumping, gc, kernel side add/del are possible
 696          */
 697         orig = ipset_dereference_bh_nfnl(h->table);
 698         atomic_set(&orig->ref, 1);
 699         atomic_inc(&orig->uref);
 700         pr_debug("attempt to resize set %s from %u to %u, t %p\n",
 701                  set->name, orig->htable_bits, htable_bits, orig);
 702         for (r = 0; r < ahash_numof_locks(orig->htable_bits); r++) {
 703                 /* Expire may replace a hbucket with another one */
 704                 rcu_read_lock_bh();
 705                 for (i = ahash_bucket_start(r, orig->htable_bits);
 706                      i < ahash_bucket_end(r, orig->htable_bits); i++) {
 707                         n = __ipset_dereference(hbucket(orig, i));
 708                         if (!n)
 709                                 continue;
 710                         for (j = 0; j < n->pos; j++) {
 711                                 if (!test_bit(j, n->used))
 712                                         continue;
 713                                 data = ahash_data(n, j, dsize);
 714                                 if (SET_ELEM_EXPIRED(set, data))
 715                                         continue;
 716 #ifdef IP_SET_HASH_WITH_NETS
 717                                 /* We have readers running parallel with us,
 718                                  * so the live data cannot be modified.
 719                                  */
 720                                 flags = 0;
 721                                 memcpy(tmp, data, dsize);
 722                                 data = tmp;
 723                                 mtype_data_reset_flags(data, &flags);
 724 #endif
 725                                 key = HKEY(data, h->initval, htable_bits);
 726                                 m = __ipset_dereference(hbucket(t, key));
 727                                 nr = ahash_region(key, htable_bits);
 728                                 if (!m) {
 729                                         m = kzalloc(sizeof(*m) +
 730                                             AHASH_INIT_SIZE * dsize,
 731                                             GFP_ATOMIC);
 732                                         if (!m) {
 733                                                 ret = -ENOMEM;
 734                                                 goto cleanup;
 735                                         }
 736                                         m->size = AHASH_INIT_SIZE;
 737                                         t->hregion[nr].ext_size +=
 738                                                 ext_size(AHASH_INIT_SIZE,
 739                                                          dsize);
 740                                         RCU_INIT_POINTER(hbucket(t, key), m);
 741                                 } else if (m->pos >= m->size) {
 742                                         struct hbucket *ht;
 743 
 744                                         if (m->size >= AHASH_MAX(h)) {
 745                                                 ret = -EAGAIN;
 746                                         } else {
 747                                                 ht = kzalloc(sizeof(*ht) +
 748                                                 (m->size + AHASH_INIT_SIZE)
 749                                                 * dsize,
 750                                                 GFP_ATOMIC);
 751                                                 if (!ht)
 752                                                         ret = -ENOMEM;
 753                                         }
 754                                         if (ret < 0)
 755                                                 goto cleanup;
 756                                         memcpy(ht, m, sizeof(struct hbucket) +
 757                                                m->size * dsize);
 758                                         ht->size = m->size + AHASH_INIT_SIZE;
 759                                         t->hregion[nr].ext_size +=
 760                                                 ext_size(AHASH_INIT_SIZE,
 761                                                          dsize);
 762                                         kfree(m);
 763                                         m = ht;
 764                                         RCU_INIT_POINTER(hbucket(t, key), ht);
 765                                 }
 766                                 d = ahash_data(m, m->pos, dsize);
 767                                 memcpy(d, data, dsize);
 768                                 set_bit(m->pos++, m->used);
 769                                 t->hregion[nr].elements++;
 770 #ifdef IP_SET_HASH_WITH_NETS
 771                                 mtype_data_reset_flags(d, &flags);
 772 #endif
 773                         }
 774                 }
 775                 rcu_read_unlock_bh();
 776         }
 777 
 778         /* There can't be any other writer. */
 779         rcu_assign_pointer(h->table, t);
 780 
 781         /* Give time to other readers of the set */
 782         synchronize_rcu();
 783 
 784         pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
 785                  orig->htable_bits, orig, t->htable_bits, t);
 786         /* Add/delete elements processed by the SET target during resize.
 787          * Kernel-side add cannot trigger a resize and userspace actions
 788          * are serialized by the mutex.
 789          */
 790         list_for_each_safe(l, lt, &h->ad) {
 791                 x = list_entry(l, struct mtype_resize_ad, list);
 792                 if (x->ad == IPSET_ADD) {
 793                         mtype_add(set, &x->d, &x->ext, &x->mext, x->flags);
 794                 } else {
 795                         mtype_del(set, &x->d, NULL, NULL, 0);
 796                 }
 797                 list_del(l);
 798                 kfree(l);
 799         }
 800         /* If there's nobody else using the table, destroy it */
 801         if (atomic_dec_and_test(&orig->uref)) {
 802                 pr_debug("Table destroy by resize %p\n", orig);
 803                 mtype_ahash_destroy(set, orig, false);
 804         }
 805 
 806 out:
 807 #ifdef IP_SET_HASH_WITH_NETS
 808         kfree(tmp);
 809 #endif
 810         return ret;
 811 
 812 cleanup:
 813         rcu_read_unlock_bh();
 814         atomic_set(&orig->ref, 0);
 815         atomic_dec(&orig->uref);
 816         mtype_ahash_destroy(set, t, false);
 817         if (ret == -EAGAIN)
 818                 goto retry;
 819         goto out;
 820 }
 821 
 822 /* Get the current number of elements and ext_size in the set  */
 823 static void
 824 mtype_ext_size(struct ip_set *set, u32 *elements, size_t *ext_size)
 825 {
 826         struct htype *h = set->data;
 827         const struct htable *t;
 828         u32 i, j, r;
 829         struct hbucket *n;
 830         struct mtype_elem *data;
 831 
 832         t = rcu_dereference_bh(h->table);
 833         for (r = 0; r < ahash_numof_locks(t->htable_bits); r++) {
 834                 for (i = ahash_bucket_start(r, t->htable_bits);
 835                      i < ahash_bucket_end(r, t->htable_bits); i++) {
 836                         n = rcu_dereference_bh(hbucket(t, i));
 837                         if (!n)
 838                                 continue;
 839                         for (j = 0; j < n->pos; j++) {
 840                                 if (!test_bit(j, n->used))
 841                                         continue;
 842                                 data = ahash_data(n, j, set->dsize);
 843                                 if (!SET_ELEM_EXPIRED(set, data))
 844                                         (*elements)++;
 845                         }
 846                 }
 847                 *ext_size += t->hregion[r].ext_size;
 848         }
 849 }
 850 
 851 /* Add an element to a hash and update the internal counters when succeeded,
 852  * otherwise report the proper error code.
 853  */
 854 static int
 855 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
 856           struct ip_set_ext *mext, u32 flags)
 857 {
 858         struct htype *h = set->data;
 859         struct htable *t;
 860         const struct mtype_elem *d = value;
 861         struct mtype_elem *data;
 862         struct hbucket *n, *old = ERR_PTR(-ENOENT);
 863         int i, j = -1, ret;
 864         bool flag_exist = flags & IPSET_FLAG_EXIST;
 865         bool deleted = false, forceadd = false, reuse = false;
 866         u32 r, key, multi = 0, elements, maxelem;
 867 
 868         rcu_read_lock_bh();
 869         t = rcu_dereference_bh(h->table);
 870         key = HKEY(value, h->initval, t->htable_bits);
 871         r = ahash_region(key, t->htable_bits);
 872         atomic_inc(&t->uref);
 873         elements = t->hregion[r].elements;
 874         maxelem = t->maxelem;
 875         if (elements >= maxelem) {
 876                 u32 e;
 877                 if (SET_WITH_TIMEOUT(set)) {
 878                         rcu_read_unlock_bh();
 879                         mtype_gc_do(set, h, t, r);
 880                         rcu_read_lock_bh();
 881                 }
 882                 maxelem = h->maxelem;
 883                 elements = 0;
 884                 for (e = 0; e < ahash_numof_locks(t->htable_bits); e++)
 885                         elements += t->hregion[e].elements;
 886                 if (elements >= maxelem && SET_WITH_FORCEADD(set))
 887                         forceadd = true;
 888         }
 889         rcu_read_unlock_bh();
 890 
 891         spin_lock_bh(&t->hregion[r].lock);
 892         n = rcu_dereference_bh(hbucket(t, key));
 893         if (!n) {
 894                 if (forceadd || elements >= maxelem)
 895                         goto set_full;
 896                 old = NULL;
 897                 n = kzalloc(sizeof(*n) + AHASH_INIT_SIZE * set->dsize,
 898                             GFP_ATOMIC);
 899                 if (!n) {
 900                         ret = -ENOMEM;
 901                         goto unlock;
 902                 }
 903                 n->size = AHASH_INIT_SIZE;
 904                 t->hregion[r].ext_size +=
 905                         ext_size(AHASH_INIT_SIZE, set->dsize);
 906                 goto copy_elem;
 907         }
 908         for (i = 0; i < n->pos; i++) {
 909                 if (!test_bit(i, n->used)) {
 910                         /* Reuse first deleted entry */
 911                         if (j == -1) {
 912                                 deleted = reuse = true;
 913                                 j = i;
 914                         }
 915                         continue;
 916                 }
 917                 data = ahash_data(n, i, set->dsize);
 918                 if (mtype_data_equal(data, d, &multi)) {
 919                         if (flag_exist || SET_ELEM_EXPIRED(set, data)) {
 920                                 /* Just the extensions could be overwritten */
 921                                 j = i;
 922                                 goto overwrite_extensions;
 923                         }
 924                         ret = -IPSET_ERR_EXIST;
 925                         goto unlock;
 926                 }
 927                 /* Reuse first timed out entry */
 928                 if (SET_ELEM_EXPIRED(set, data) && j == -1) {
 929                         j = i;
 930                         reuse = true;
 931                 }
 932         }
 933         if (reuse || forceadd) {
 934                 if (j == -1)
 935                         j = 0;
 936                 data = ahash_data(n, j, set->dsize);
 937                 if (!deleted) {
 938 #ifdef IP_SET_HASH_WITH_NETS
 939                         for (i = 0; i < IPSET_NET_COUNT; i++)
 940                                 mtype_del_cidr(set, h,
 941                                         NCIDR_PUT(DCIDR_GET(data->cidr, i)),
 942                                         i);
 943 #endif
 944                         ip_set_ext_destroy(set, data);
 945                         t->hregion[r].elements--;
 946                 }
 947                 goto copy_data;
 948         }
 949         if (elements >= maxelem)
 950                 goto set_full;
 951         /* Create a new slot */
 952         if (n->pos >= n->size) {
 953                 TUNE_AHASH_MAX(h, multi);
 954                 if (n->size >= AHASH_MAX(h)) {
 955                         /* Trigger rehashing */
 956                         mtype_data_next(&h->next, d);
 957                         ret = -EAGAIN;
 958                         goto resize;
 959                 }
 960                 old = n;
 961                 n = kzalloc(sizeof(*n) +
 962                             (old->size + AHASH_INIT_SIZE) * set->dsize,
 963                             GFP_ATOMIC);
 964                 if (!n) {
 965                         ret = -ENOMEM;
 966                         goto unlock;
 967                 }
 968                 memcpy(n, old, sizeof(struct hbucket) +
 969                        old->size * set->dsize);
 970                 n->size = old->size + AHASH_INIT_SIZE;
 971                 t->hregion[r].ext_size +=
 972                         ext_size(AHASH_INIT_SIZE, set->dsize);
 973         }
 974 
 975 copy_elem:
 976         j = n->pos++;
 977         data = ahash_data(n, j, set->dsize);
 978 copy_data:
 979         t->hregion[r].elements++;
 980 #ifdef IP_SET_HASH_WITH_NETS
 981         for (i = 0; i < IPSET_NET_COUNT; i++)
 982                 mtype_add_cidr(set, h, NCIDR_PUT(DCIDR_GET(d->cidr, i)), i);
 983 #endif
 984         memcpy(data, d, sizeof(struct mtype_elem));
 985 overwrite_extensions:
 986 #ifdef IP_SET_HASH_WITH_NETS
 987         mtype_data_set_flags(data, flags);
 988 #endif
 989         if (SET_WITH_COUNTER(set))
 990                 ip_set_init_counter(ext_counter(data, set), ext);
 991         if (SET_WITH_COMMENT(set))
 992                 ip_set_init_comment(set, ext_comment(data, set), ext);
 993         if (SET_WITH_SKBINFO(set))
 994                 ip_set_init_skbinfo(ext_skbinfo(data, set), ext);
 995         /* Must come last for the case when timed out entry is reused */
 996         if (SET_WITH_TIMEOUT(set))
 997                 ip_set_timeout_set(ext_timeout(data, set), ext->timeout);
 998         smp_mb__before_atomic();
 999         set_bit(j, n->used);
1000         if (old != ERR_PTR(-ENOENT)) {
1001                 rcu_assign_pointer(hbucket(t, key), n);
1002                 if (old)
1003                         kfree_rcu(old, rcu);
1004         }
1005         ret = 0;
1006 resize:
1007         spin_unlock_bh(&t->hregion[r].lock);
1008         if (atomic_read(&t->ref) && ext->target) {
1009                 /* Resize is in process and kernel side add, save values */
1010                 struct mtype_resize_ad *x;
1011 
1012                 x = kzalloc(sizeof(struct mtype_resize_ad), GFP_ATOMIC);
1013                 if (!x)
1014                         /* Don't bother */
1015                         goto out;
1016                 x->ad = IPSET_ADD;
1017                 memcpy(&x->d, value, sizeof(struct mtype_elem));
1018                 memcpy(&x->ext, ext, sizeof(struct ip_set_ext));
1019                 memcpy(&x->mext, mext, sizeof(struct ip_set_ext));
1020                 x->flags = flags;
1021                 spin_lock_bh(&set->lock);
1022                 list_add_tail(&x->list, &h->ad);
1023                 spin_unlock_bh(&set->lock);
1024         }
1025         goto out;
1026 
1027 set_full:
1028         if (net_ratelimit())
1029                 pr_warn("Set %s is full, maxelem %u reached\n",
1030                         set->name, maxelem);
1031         ret = -IPSET_ERR_HASH_FULL;
1032 unlock:
1033         spin_unlock_bh(&t->hregion[r].lock);
1034 out:
1035         if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1036                 pr_debug("Table destroy after resize by add: %p\n", t);
1037                 mtype_ahash_destroy(set, t, false);
1038         }
1039         return ret;
1040 }
1041 
1042 /* Delete an element from the hash and free up space if possible.
1043  */
1044 static int
1045 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1046           struct ip_set_ext *mext, u32 flags)
1047 {
1048         struct htype *h = set->data;
1049         struct htable *t;
1050         const struct mtype_elem *d = value;
1051         struct mtype_elem *data;
1052         struct hbucket *n;
1053         struct mtype_resize_ad *x = NULL;
1054         int i, j, k, r, ret = -IPSET_ERR_EXIST;
1055         u32 key, multi = 0;
1056         size_t dsize = set->dsize;
1057 
1058         /* Userspace add and resize is excluded by the mutex.
1059          * Kernespace add does not trigger resize.
1060          */
1061         rcu_read_lock_bh();
1062         t = rcu_dereference_bh(h->table);
1063         key = HKEY(value, h->initval, t->htable_bits);
1064         r = ahash_region(key, t->htable_bits);
1065         atomic_inc(&t->uref);
1066         rcu_read_unlock_bh();
1067 
1068         spin_lock_bh(&t->hregion[r].lock);
1069         n = rcu_dereference_bh(hbucket(t, key));
1070         if (!n)
1071                 goto out;
1072         for (i = 0, k = 0; i < n->pos; i++) {
1073                 if (!test_bit(i, n->used)) {
1074                         k++;
1075                         continue;
1076                 }
1077                 data = ahash_data(n, i, dsize);
1078                 if (!mtype_data_equal(data, d, &multi))
1079                         continue;
1080                 if (SET_ELEM_EXPIRED(set, data))
1081                         goto out;
1082 
1083                 ret = 0;
1084                 clear_bit(i, n->used);
1085                 smp_mb__after_atomic();
1086                 if (i + 1 == n->pos)
1087                         n->pos--;
1088                 t->hregion[r].elements--;
1089 #ifdef IP_SET_HASH_WITH_NETS
1090                 for (j = 0; j < IPSET_NET_COUNT; j++)
1091                         mtype_del_cidr(set, h,
1092                                        NCIDR_PUT(DCIDR_GET(d->cidr, j)), j);
1093 #endif
1094                 ip_set_ext_destroy(set, data);
1095 
1096                 if (atomic_read(&t->ref) && ext->target) {
1097                         /* Resize is in process and kernel side del,
1098                          * save values
1099                          */
1100                         x = kzalloc(sizeof(struct mtype_resize_ad),
1101                                     GFP_ATOMIC);
1102                         if (x) {
1103                                 x->ad = IPSET_DEL;
1104                                 memcpy(&x->d, value,
1105                                        sizeof(struct mtype_elem));
1106                                 x->flags = flags;
1107                         }
1108                 }
1109                 for (; i < n->pos; i++) {
1110                         if (!test_bit(i, n->used))
1111                                 k++;
1112                 }
1113                 if (n->pos == 0 && k == 0) {
1114                         t->hregion[r].ext_size -= ext_size(n->size, dsize);
1115                         rcu_assign_pointer(hbucket(t, key), NULL);
1116                         kfree_rcu(n, rcu);
1117                 } else if (k >= AHASH_INIT_SIZE) {
1118                         struct hbucket *tmp = kzalloc(sizeof(*tmp) +
1119                                         (n->size - AHASH_INIT_SIZE) * dsize,
1120                                         GFP_ATOMIC);
1121                         if (!tmp)
1122                                 goto out;
1123                         tmp->size = n->size - AHASH_INIT_SIZE;
1124                         for (j = 0, k = 0; j < n->pos; j++) {
1125                                 if (!test_bit(j, n->used))
1126                                         continue;
1127                                 data = ahash_data(n, j, dsize);
1128                                 memcpy(tmp->value + k * dsize, data, dsize);
1129                                 set_bit(k, tmp->used);
1130                                 k++;
1131                         }
1132                         tmp->pos = k;
1133                         t->hregion[r].ext_size -=
1134                                 ext_size(AHASH_INIT_SIZE, dsize);
1135                         rcu_assign_pointer(hbucket(t, key), tmp);
1136                         kfree_rcu(n, rcu);
1137                 }
1138                 goto out;
1139         }
1140 
1141 out:
1142         spin_unlock_bh(&t->hregion[r].lock);
1143         if (x) {
1144                 spin_lock_bh(&set->lock);
1145                 list_add(&x->list, &h->ad);
1146                 spin_unlock_bh(&set->lock);
1147         }
1148         if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1149                 pr_debug("Table destroy after resize by del: %p\n", t);
1150                 mtype_ahash_destroy(set, t, false);
1151         }
1152         return ret;
1153 }
1154 
1155 static inline int
1156 mtype_data_match(struct mtype_elem *data, const struct ip_set_ext *ext,
1157                  struct ip_set_ext *mext, struct ip_set *set, u32 flags)
1158 {
1159         if (!ip_set_match_extensions(set, ext, mext, flags, data))
1160                 return 0;
1161         /* nomatch entries return -ENOTEMPTY */
1162         return mtype_do_data_match(data);
1163 }
1164 
1165 #ifdef IP_SET_HASH_WITH_NETS
1166 /* Special test function which takes into account the different network
1167  * sizes added to the set
1168  */
1169 static int
1170 mtype_test_cidrs(struct ip_set *set, struct mtype_elem *d,
1171                  const struct ip_set_ext *ext,
1172                  struct ip_set_ext *mext, u32 flags)
1173 {
1174         struct htype *h = set->data;
1175         struct htable *t = rcu_dereference_bh(h->table);
1176         struct hbucket *n;
1177         struct mtype_elem *data;
1178 #if IPSET_NET_COUNT == 2
1179         struct mtype_elem orig = *d;
1180         int ret, i, j = 0, k;
1181 #else
1182         int ret, i, j = 0;
1183 #endif
1184         u32 key, multi = 0;
1185 
1186         pr_debug("test by nets\n");
1187         for (; j < NLEN && h->nets[j].cidr[0] && !multi; j++) {
1188 #if IPSET_NET_COUNT == 2
1189                 mtype_data_reset_elem(d, &orig);
1190                 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]), false);
1191                 for (k = 0; k < NLEN && h->nets[k].cidr[1] && !multi;
1192                      k++) {
1193                         mtype_data_netmask(d, NCIDR_GET(h->nets[k].cidr[1]),
1194                                            true);
1195 #else
1196                 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]));
1197 #endif
1198                 key = HKEY(d, h->initval, t->htable_bits);
1199                 n = rcu_dereference_bh(hbucket(t, key));
1200                 if (!n)
1201                         continue;
1202                 for (i = 0; i < n->pos; i++) {
1203                         if (!test_bit(i, n->used))
1204                                 continue;
1205                         data = ahash_data(n, i, set->dsize);
1206                         if (!mtype_data_equal(data, d, &multi))
1207                                 continue;
1208                         ret = mtype_data_match(data, ext, mext, set, flags);
1209                         if (ret != 0)
1210                                 return ret;
1211 #ifdef IP_SET_HASH_WITH_MULTI
1212                         /* No match, reset multiple match flag */
1213                         multi = 0;
1214 #endif
1215                 }
1216 #if IPSET_NET_COUNT == 2
1217                 }
1218 #endif
1219         }
1220         return 0;
1221 }
1222 #endif
1223 
1224 /* Test whether the element is added to the set */
1225 static int
1226 mtype_test(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1227            struct ip_set_ext *mext, u32 flags)
1228 {
1229         struct htype *h = set->data;
1230         struct htable *t;
1231         struct mtype_elem *d = value;
1232         struct hbucket *n;
1233         struct mtype_elem *data;
1234         int i, ret = 0;
1235         u32 key, multi = 0;
1236 
1237         rcu_read_lock_bh();
1238         t = rcu_dereference_bh(h->table);
1239 #ifdef IP_SET_HASH_WITH_NETS
1240         /* If we test an IP address and not a network address,
1241          * try all possible network sizes
1242          */
1243         for (i = 0; i < IPSET_NET_COUNT; i++)
1244                 if (DCIDR_GET(d->cidr, i) != HOST_MASK)
1245                         break;
1246         if (i == IPSET_NET_COUNT) {
1247                 ret = mtype_test_cidrs(set, d, ext, mext, flags);
1248                 goto out;
1249         }
1250 #endif
1251 
1252         key = HKEY(d, h->initval, t->htable_bits);
1253         n = rcu_dereference_bh(hbucket(t, key));
1254         if (!n) {
1255                 ret = 0;
1256                 goto out;
1257         }
1258         for (i = 0; i < n->pos; i++) {
1259                 if (!test_bit(i, n->used))
1260                         continue;
1261                 data = ahash_data(n, i, set->dsize);
1262                 if (!mtype_data_equal(data, d, &multi))
1263                         continue;
1264                 ret = mtype_data_match(data, ext, mext, set, flags);
1265                 if (ret != 0)
1266                         goto out;
1267         }
1268 out:
1269         rcu_read_unlock_bh();
1270         return ret;
1271 }
1272 
1273 /* Reply a HEADER request: fill out the header part of the set */
1274 static int
1275 mtype_head(struct ip_set *set, struct sk_buff *skb)
1276 {
1277         struct htype *h = set->data;
1278         const struct htable *t;
1279         struct nlattr *nested;
1280         size_t memsize;
1281         u32 elements = 0;
1282         size_t ext_size = 0;
1283         u8 htable_bits;
1284 
1285         rcu_read_lock_bh();
1286         t = rcu_dereference_bh(h->table);
1287         mtype_ext_size(set, &elements, &ext_size);
1288         memsize = mtype_ahash_memsize(h, t) + ext_size + set->ext_size;
1289         htable_bits = t->htable_bits;
1290         rcu_read_unlock_bh();
1291 
1292         nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1293         if (!nested)
1294                 goto nla_put_failure;
1295         if (nla_put_net32(skb, IPSET_ATTR_HASHSIZE,
1296                           htonl(jhash_size(htable_bits))) ||
1297             nla_put_net32(skb, IPSET_ATTR_MAXELEM, htonl(h->maxelem)))
1298                 goto nla_put_failure;
1299 #ifdef IP_SET_HASH_WITH_NETMASK
1300         if (h->netmask != HOST_MASK &&
1301             nla_put_u8(skb, IPSET_ATTR_NETMASK, h->netmask))
1302                 goto nla_put_failure;
1303 #endif
1304 #ifdef IP_SET_HASH_WITH_MARKMASK
1305         if (nla_put_u32(skb, IPSET_ATTR_MARKMASK, h->markmask))
1306                 goto nla_put_failure;
1307 #endif
1308         if (nla_put_net32(skb, IPSET_ATTR_REFERENCES, htonl(set->ref)) ||
1309             nla_put_net32(skb, IPSET_ATTR_MEMSIZE, htonl(memsize)) ||
1310             nla_put_net32(skb, IPSET_ATTR_ELEMENTS, htonl(elements)))
1311                 goto nla_put_failure;
1312         if (unlikely(ip_set_put_flags(skb, set)))
1313                 goto nla_put_failure;
1314         nla_nest_end(skb, nested);
1315 
1316         return 0;
1317 nla_put_failure:
1318         return -EMSGSIZE;
1319 }
1320 
1321 /* Make possible to run dumping parallel with resizing */
1322 static void
1323 mtype_uref(struct ip_set *set, struct netlink_callback *cb, bool start)
1324 {
1325         struct htype *h = set->data;
1326         struct htable *t;
1327 
1328         if (start) {
1329                 rcu_read_lock_bh();
1330                 t = ipset_dereference_bh_nfnl(h->table);
1331                 atomic_inc(&t->uref);
1332                 cb->args[IPSET_CB_PRIVATE] = (unsigned long)t;
1333                 rcu_read_unlock_bh();
1334         } else if (cb->args[IPSET_CB_PRIVATE]) {
1335                 t = (struct htable *)cb->args[IPSET_CB_PRIVATE];
1336                 if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1337                         pr_debug("Table destroy after resize "
1338                                  " by dump: %p\n", t);
1339                         mtype_ahash_destroy(set, t, false);
1340                 }
1341                 cb->args[IPSET_CB_PRIVATE] = 0;
1342         }
1343 }
1344 
1345 /* Reply a LIST/SAVE request: dump the elements of the specified set */
1346 static int
1347 mtype_list(const struct ip_set *set,
1348            struct sk_buff *skb, struct netlink_callback *cb)
1349 {
1350         const struct htable *t;
1351         struct nlattr *atd, *nested;
1352         const struct hbucket *n;
1353         const struct mtype_elem *e;
1354         u32 first = cb->args[IPSET_CB_ARG0];
1355         /* We assume that one hash bucket fills into one page */
1356         void *incomplete;
1357         int i, ret = 0;
1358 
1359         atd = nla_nest_start(skb, IPSET_ATTR_ADT);
1360         if (!atd)
1361                 return -EMSGSIZE;
1362 
1363         pr_debug("list hash set %s\n", set->name);
1364         t = (const struct htable *)cb->args[IPSET_CB_PRIVATE];
1365         /* Expire may replace a hbucket with another one */
1366         rcu_read_lock();
1367         for (; cb->args[IPSET_CB_ARG0] < jhash_size(t->htable_bits);
1368              cb->args[IPSET_CB_ARG0]++) {
1369                 cond_resched_rcu();
1370                 incomplete = skb_tail_pointer(skb);
1371                 n = rcu_dereference(hbucket(t, cb->args[IPSET_CB_ARG0]));
1372                 pr_debug("cb->arg bucket: %lu, t %p n %p\n",
1373                          cb->args[IPSET_CB_ARG0], t, n);
1374                 if (!n)
1375                         continue;
1376                 for (i = 0; i < n->pos; i++) {
1377                         if (!test_bit(i, n->used))
1378                                 continue;
1379                         e = ahash_data(n, i, set->dsize);
1380                         if (SET_ELEM_EXPIRED(set, e))
1381                                 continue;
1382                         pr_debug("list hash %lu hbucket %p i %u, data %p\n",
1383                                  cb->args[IPSET_CB_ARG0], n, i, e);
1384                         nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1385                         if (!nested) {
1386                                 if (cb->args[IPSET_CB_ARG0] == first) {
1387                                         nla_nest_cancel(skb, atd);
1388                                         ret = -EMSGSIZE;
1389                                         goto out;
1390                                 }
1391                                 goto nla_put_failure;
1392                         }
1393                         if (mtype_data_list(skb, e))
1394                                 goto nla_put_failure;
1395                         if (ip_set_put_extensions(skb, set, e, true))
1396                                 goto nla_put_failure;
1397                         nla_nest_end(skb, nested);
1398                 }
1399         }
1400         nla_nest_end(skb, atd);
1401         /* Set listing finished */
1402         cb->args[IPSET_CB_ARG0] = 0;
1403 
1404         goto out;
1405 
1406 nla_put_failure:
1407         nlmsg_trim(skb, incomplete);
1408         if (unlikely(first == cb->args[IPSET_CB_ARG0])) {
1409                 pr_warn("Can't list set %s: one bucket does not fit into a message. Please report it!\n",
1410                         set->name);
1411                 cb->args[IPSET_CB_ARG0] = 0;
1412                 ret = -EMSGSIZE;
1413         } else {
1414                 nla_nest_end(skb, atd);
1415         }
1416 out:
1417         rcu_read_unlock();
1418         return ret;
1419 }
1420 
1421 static int
1422 IPSET_TOKEN(MTYPE, _kadt)(struct ip_set *set, const struct sk_buff *skb,
1423                           const struct xt_action_param *par,
1424                           enum ipset_adt adt, struct ip_set_adt_opt *opt);
1425 
1426 static int
1427 IPSET_TOKEN(MTYPE, _uadt)(struct ip_set *set, struct nlattr *tb[],
1428                           enum ipset_adt adt, u32 *lineno, u32 flags,
1429                           bool retried);
1430 
1431 static const struct ip_set_type_variant mtype_variant = {
1432         .kadt   = mtype_kadt,
1433         .uadt   = mtype_uadt,
1434         .adt    = {
1435                 [IPSET_ADD] = mtype_add,
1436                 [IPSET_DEL] = mtype_del,
1437                 [IPSET_TEST] = mtype_test,
1438         },
1439         .destroy = mtype_destroy,
1440         .flush  = mtype_flush,
1441         .head   = mtype_head,
1442         .list   = mtype_list,
1443         .uref   = mtype_uref,
1444         .resize = mtype_resize,
1445         .same_set = mtype_same_set,
1446         .region_lock = true,
1447 };
1448 
1449 #ifdef IP_SET_EMIT_CREATE
1450 static int
1451 IPSET_TOKEN(HTYPE, _create)(struct net *net, struct ip_set *set,
1452                             struct nlattr *tb[], u32 flags)
1453 {
1454         u32 hashsize = IPSET_DEFAULT_HASHSIZE, maxelem = IPSET_DEFAULT_MAXELEM;
1455 #ifdef IP_SET_HASH_WITH_MARKMASK
1456         u32 markmask;
1457 #endif
1458         u8 hbits;
1459 #ifdef IP_SET_HASH_WITH_NETMASK
1460         u8 netmask;
1461 #endif
1462         size_t hsize;
1463         struct htype *h;
1464         struct htable *t;
1465         u32 i;
1466 
1467         pr_debug("Create set %s with family %s\n",
1468                  set->name, set->family == NFPROTO_IPV4 ? "inet" : "inet6");
1469 
1470 #ifdef IP_SET_PROTO_UNDEF
1471         if (set->family != NFPROTO_UNSPEC)
1472                 return -IPSET_ERR_INVALID_FAMILY;
1473 #else
1474         if (!(set->family == NFPROTO_IPV4 || set->family == NFPROTO_IPV6))
1475                 return -IPSET_ERR_INVALID_FAMILY;
1476 #endif
1477 
1478         if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_HASHSIZE) ||
1479                      !ip_set_optattr_netorder(tb, IPSET_ATTR_MAXELEM) ||
1480                      !ip_set_optattr_netorder(tb, IPSET_ATTR_TIMEOUT) ||
1481                      !ip_set_optattr_netorder(tb, IPSET_ATTR_CADT_FLAGS)))
1482                 return -IPSET_ERR_PROTOCOL;
1483 
1484 #ifdef IP_SET_HASH_WITH_MARKMASK
1485         /* Separated condition in order to avoid directive in argument list */
1486         if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_MARKMASK)))
1487                 return -IPSET_ERR_PROTOCOL;
1488 
1489         markmask = 0xffffffff;
1490         if (tb[IPSET_ATTR_MARKMASK]) {
1491                 markmask = ntohl(nla_get_be32(tb[IPSET_ATTR_MARKMASK]));
1492                 if (markmask == 0)
1493                         return -IPSET_ERR_INVALID_MARKMASK;
1494         }
1495 #endif
1496 
1497 #ifdef IP_SET_HASH_WITH_NETMASK
1498         netmask = set->family == NFPROTO_IPV4 ? 32 : 128;
1499         if (tb[IPSET_ATTR_NETMASK]) {
1500                 netmask = nla_get_u8(tb[IPSET_ATTR_NETMASK]);
1501 
1502                 if ((set->family == NFPROTO_IPV4 && netmask > 32) ||
1503                     (set->family == NFPROTO_IPV6 && netmask > 128) ||
1504                     netmask == 0)
1505                         return -IPSET_ERR_INVALID_NETMASK;
1506         }
1507 #endif
1508 
1509         if (tb[IPSET_ATTR_HASHSIZE]) {
1510                 hashsize = ip_set_get_h32(tb[IPSET_ATTR_HASHSIZE]);
1511                 if (hashsize < IPSET_MIMINAL_HASHSIZE)
1512                         hashsize = IPSET_MIMINAL_HASHSIZE;
1513         }
1514 
1515         if (tb[IPSET_ATTR_MAXELEM])
1516                 maxelem = ip_set_get_h32(tb[IPSET_ATTR_MAXELEM]);
1517 
1518         hsize = sizeof(*h);
1519         h = kzalloc(hsize, GFP_KERNEL);
1520         if (!h)
1521                 return -ENOMEM;
1522 
1523         hbits = htable_bits(hashsize);
1524         hsize = htable_size(hbits);
1525         if (hsize == 0) {
1526                 kfree(h);
1527                 return -ENOMEM;
1528         }
1529         t = ip_set_alloc(hsize);
1530         if (!t) {
1531                 kfree(h);
1532                 return -ENOMEM;
1533         }
1534         t->hregion = ip_set_alloc(ahash_sizeof_regions(hbits));
1535         if (!t->hregion) {
1536                 kfree(t);
1537                 kfree(h);
1538                 return -ENOMEM;
1539         }
1540         h->gc.set = set;
1541         for (i = 0; i < ahash_numof_locks(hbits); i++)
1542                 spin_lock_init(&t->hregion[i].lock);
1543         h->maxelem = maxelem;
1544 #ifdef IP_SET_HASH_WITH_NETMASK
1545         h->netmask = netmask;
1546 #endif
1547 #ifdef IP_SET_HASH_WITH_MARKMASK
1548         h->markmask = markmask;
1549 #endif
1550         get_random_bytes(&h->initval, sizeof(h->initval));
1551 
1552         t->htable_bits = hbits;
1553         t->maxelem = h->maxelem / ahash_numof_locks(hbits);
1554         RCU_INIT_POINTER(h->table, t);
1555 
1556         INIT_LIST_HEAD(&h->ad);
1557         set->data = h;
1558 #ifndef IP_SET_PROTO_UNDEF
1559         if (set->family == NFPROTO_IPV4) {
1560 #endif
1561                 set->variant = &IPSET_TOKEN(HTYPE, 4_variant);
1562                 set->dsize = ip_set_elem_len(set, tb,
1563                         sizeof(struct IPSET_TOKEN(HTYPE, 4_elem)),
1564                         __alignof__(struct IPSET_TOKEN(HTYPE, 4_elem)));
1565 #ifndef IP_SET_PROTO_UNDEF
1566         } else {
1567                 set->variant = &IPSET_TOKEN(HTYPE, 6_variant);
1568                 set->dsize = ip_set_elem_len(set, tb,
1569                         sizeof(struct IPSET_TOKEN(HTYPE, 6_elem)),
1570                         __alignof__(struct IPSET_TOKEN(HTYPE, 6_elem)));
1571         }
1572 #endif
1573         set->timeout = IPSET_NO_TIMEOUT;
1574         if (tb[IPSET_ATTR_TIMEOUT]) {
1575                 set->timeout = ip_set_timeout_uget(tb[IPSET_ATTR_TIMEOUT]);
1576 #ifndef IP_SET_PROTO_UNDEF
1577                 if (set->family == NFPROTO_IPV4)
1578 #endif
1579                         IPSET_TOKEN(HTYPE, 4_gc_init)(&h->gc);
1580 #ifndef IP_SET_PROTO_UNDEF
1581                 else
1582                         IPSET_TOKEN(HTYPE, 6_gc_init)(&h->gc);
1583 #endif
1584         }
1585         pr_debug("create %s hashsize %u (%u) maxelem %u: %p(%p)\n",
1586                  set->name, jhash_size(t->htable_bits),
1587                  t->htable_bits, h->maxelem, set->data, t);
1588 
1589         return 0;
1590 }
1591 #endif /* IP_SET_EMIT_CREATE */
1592 
1593 #undef HKEY_DATALEN
/* [<][>][^][v][top][bottom][index][help] */
root/net/netfilter/ipset/ip_set_hash_gen.h

INCLUDED FROM

DEFINITIONS
