1 /* Copyright (C) 2013 Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
3 * This program is free software; you can redistribute it and/or modify
4 * it under the terms of the GNU General Public License version 2 as
5 * published by the Free Software Foundation.
8 #ifndef _IP_SET_HASH_GEN_H
9 #define _IP_SET_HASH_GEN_H
11 #include <linux/rcupdate.h>
12 #include <linux/jhash.h>
13 #include <linux/types.h>
14 #include <linux/netfilter/ipset/ip_set_timeout.h>
16 #define __ipset_dereference_protected(p, c) rcu_dereference_protected(p, c)
17 #define ipset_dereference_protected(p, set) \
18 __ipset_dereference_protected(p, spin_is_locked(&(set)->lock))
20 #define rcu_dereference_bh_nfnl(p) rcu_dereference_bh_check(p, 1)
22 /* Hashing which uses arrays to resolve clashing. The hash table is resized
23 * (doubled) when searching becomes too long.
24 * Internally jhash is used with the assumption that the size of the
25 * stored data is a multiple of sizeof(u32).
27 * Readers and resizing
29 * Resizing can be triggered by userspace command only, and those
30 * are serialized by the nfnl mutex. During resizing the set is
31 * read-locked, so the only possible concurrent operations are
32 * the kernel side readers. Those must be protected by proper RCU locking.
35 /* Number of elements to store in an initial array block */
36 #define AHASH_INIT_SIZE 4
37 /* Max number of elements to store in an array block */
38 #define AHASH_MAX_SIZE (3 * AHASH_INIT_SIZE)
39 /* Max muber of elements in the array block when tuned */
40 #define AHASH_MAX_TUNED 64
42 /* Max number of elements can be tuned */
43 #ifdef IP_SET_HASH_WITH_MULTI
44 #define AHASH_MAX(h) ((h)->ahash_max)
47 tune_ahash_max(u8 curr, u32 multi)
54 n = curr + AHASH_INIT_SIZE;
55 /* Currently, at listing one hash bucket must fit into a message.
56 * Therefore we have a hard limit here.
58 return n > curr && n <= AHASH_MAX_TUNED ? n : curr;
61 #define TUNE_AHASH_MAX(h, multi) \
62 ((h)->ahash_max = tune_ahash_max((h)->ahash_max, multi))
64 #define AHASH_MAX(h) AHASH_MAX_SIZE
65 #define TUNE_AHASH_MAX(h, multi)
70 struct rcu_head rcu; /* for call_rcu_bh */
71 /* Which positions are used in the array */
72 DECLARE_BITMAP(used, AHASH_MAX_TUNED);
73 u8 size; /* size of the array */
74 u8 pos; /* position of the first free entry */
75 unsigned char value[0] /* the array of the values */
76 __aligned(__alignof__(u64));
79 /* The hash table: the table size stored here in order to make resizing easy */
81 atomic_t ref; /* References for resizing */
82 atomic_t uref; /* References for dumping */
83 u8 htable_bits; /* size of hash table == 2^htable_bits */
84 struct hbucket __rcu *bucket[0]; /* hashtable buckets */
87 #define hbucket(h, i) ((h)->bucket[i])
88 #define ext_size(n, dsize) \
89 (sizeof(struct hbucket) + (n) * (dsize))
91 #ifndef IPSET_NET_COUNT
92 #define IPSET_NET_COUNT 1
95 /* Book-keeping of the prefixes added to the set */
97 u32 nets[IPSET_NET_COUNT]; /* number of elements for this cidr */
98 u8 cidr[IPSET_NET_COUNT]; /* the cidr value */
101 /* Compute the hash table size */
103 htable_size(u8 hbits)
107 /* We must fit both into u32 in jhash and size_t */
110 hsize = jhash_size(hbits);
111 if ((((size_t)-1) - sizeof(struct htable)) / sizeof(struct hbucket *)
115 return hsize * sizeof(struct hbucket *) + sizeof(struct htable);
118 /* Compute htable_bits from the user input parameter hashsize */
120 htable_bits(u32 hashsize)
122 /* Assume that hashsize == 2^htable_bits */
123 u8 bits = fls(hashsize - 1);
125 if (jhash_size(bits) != hashsize)
126 /* Round up to the first 2^n value */
127 bits = fls(hashsize);
132 #ifdef IP_SET_HASH_WITH_NETS
133 #if IPSET_NET_COUNT > 1
134 #define __CIDR(cidr, i) (cidr[i])
136 #define __CIDR(cidr, i) (cidr)
139 /* cidr + 1 is stored in net_prefixes to support /0 */
140 #define NCIDR_PUT(cidr) ((cidr) + 1)
141 #define NCIDR_GET(cidr) ((cidr) - 1)
143 #ifdef IP_SET_HASH_WITH_NETS_PACKED
144 /* When cidr is packed with nomatch, cidr - 1 is stored in the data entry */
145 #define DCIDR_PUT(cidr) ((cidr) - 1)
146 #define DCIDR_GET(cidr, i) (__CIDR(cidr, i) + 1)
148 #define DCIDR_PUT(cidr) (cidr)
149 #define DCIDR_GET(cidr, i) __CIDR(cidr, i)
152 #define INIT_CIDR(cidr, host_mask) \
153 DCIDR_PUT(((cidr) ? NCIDR_GET(cidr) : host_mask))
155 #ifdef IP_SET_HASH_WITH_NET0
156 /* cidr from 0 to HOST_MASK value and c = cidr + 1 */
157 #define NLEN (HOST_MASK + 1)
158 #define CIDR_POS(c) ((c) - 1)
160 /* cidr from 1 to HOST_MASK value and c = cidr + 1 */
161 #define NLEN HOST_MASK
162 #define CIDR_POS(c) ((c) - 2)
167 #endif /* IP_SET_HASH_WITH_NETS */
169 #endif /* _IP_SET_HASH_GEN_H */
172 #error "MTYPE is not defined!"
176 #error "HTYPE is not defined!"
180 #error "HOST_MASK is not defined!"
183 /* Family dependent templates */
186 #undef mtype_data_equal
187 #undef mtype_do_data_match
188 #undef mtype_data_set_flags
189 #undef mtype_data_reset_elem
190 #undef mtype_data_reset_flags
191 #undef mtype_data_netmask
192 #undef mtype_data_list
193 #undef mtype_data_next
196 #undef mtype_ahash_destroy
197 #undef mtype_ext_cleanup
198 #undef mtype_add_cidr
199 #undef mtype_del_cidr
200 #undef mtype_ahash_memsize
203 #undef mtype_same_set
209 #undef mtype_test_cidrs
219 #undef mtype_data_match
224 #define mtype_data_equal IPSET_TOKEN(MTYPE, _data_equal)
225 #ifdef IP_SET_HASH_WITH_NETS
226 #define mtype_do_data_match IPSET_TOKEN(MTYPE, _do_data_match)
228 #define mtype_do_data_match(d) 1
230 #define mtype_data_set_flags IPSET_TOKEN(MTYPE, _data_set_flags)
231 #define mtype_data_reset_elem IPSET_TOKEN(MTYPE, _data_reset_elem)
232 #define mtype_data_reset_flags IPSET_TOKEN(MTYPE, _data_reset_flags)
233 #define mtype_data_netmask IPSET_TOKEN(MTYPE, _data_netmask)
234 #define mtype_data_list IPSET_TOKEN(MTYPE, _data_list)
235 #define mtype_data_next IPSET_TOKEN(MTYPE, _data_next)
236 #define mtype_elem IPSET_TOKEN(MTYPE, _elem)
238 #define mtype_ahash_destroy IPSET_TOKEN(MTYPE, _ahash_destroy)
239 #define mtype_ext_cleanup IPSET_TOKEN(MTYPE, _ext_cleanup)
240 #define mtype_add_cidr IPSET_TOKEN(MTYPE, _add_cidr)
241 #define mtype_del_cidr IPSET_TOKEN(MTYPE, _del_cidr)
242 #define mtype_ahash_memsize IPSET_TOKEN(MTYPE, _ahash_memsize)
243 #define mtype_flush IPSET_TOKEN(MTYPE, _flush)
244 #define mtype_destroy IPSET_TOKEN(MTYPE, _destroy)
245 #define mtype_same_set IPSET_TOKEN(MTYPE, _same_set)
246 #define mtype_kadt IPSET_TOKEN(MTYPE, _kadt)
247 #define mtype_uadt IPSET_TOKEN(MTYPE, _uadt)
249 #define mtype_add IPSET_TOKEN(MTYPE, _add)
250 #define mtype_del IPSET_TOKEN(MTYPE, _del)
251 #define mtype_test_cidrs IPSET_TOKEN(MTYPE, _test_cidrs)
252 #define mtype_test IPSET_TOKEN(MTYPE, _test)
253 #define mtype_uref IPSET_TOKEN(MTYPE, _uref)
254 #define mtype_expire IPSET_TOKEN(MTYPE, _expire)
255 #define mtype_resize IPSET_TOKEN(MTYPE, _resize)
256 #define mtype_head IPSET_TOKEN(MTYPE, _head)
257 #define mtype_list IPSET_TOKEN(MTYPE, _list)
258 #define mtype_gc IPSET_TOKEN(MTYPE, _gc)
259 #define mtype_gc_init IPSET_TOKEN(MTYPE, _gc_init)
260 #define mtype_variant IPSET_TOKEN(MTYPE, _variant)
261 #define mtype_data_match IPSET_TOKEN(MTYPE, _data_match)
264 #define HKEY_DATALEN sizeof(struct mtype_elem)
269 #define HKEY(data, initval, htable_bits) \
271 const u32 *__k = (const u32 *)data; \
272 u32 __l = HKEY_DATALEN / sizeof(u32); \
274 BUILD_BUG_ON(HKEY_DATALEN % sizeof(u32) != 0); \
276 jhash2(__k, __l, initval) & jhash_mask(htable_bits); \
279 /* The generic hash structure */
281 struct htable __rcu *table; /* the hash table */
282 struct timer_list gc; /* garbage collection when timeout enabled */
283 u32 maxelem; /* max elements in the hash */
284 u32 initval; /* random jhash init value */
285 #ifdef IP_SET_HASH_WITH_MARKMASK
286 u32 markmask; /* markmask value for mark mask to store */
288 #ifdef IP_SET_HASH_WITH_MULTI
289 u8 ahash_max; /* max elements in an array block */
291 #ifdef IP_SET_HASH_WITH_NETMASK
292 u8 netmask; /* netmask value for subnets to store */
294 struct mtype_elem next; /* temporary storage for uadd */
295 #ifdef IP_SET_HASH_WITH_NETS
296 struct net_prefixes nets[NLEN]; /* book-keeping of prefixes */
300 #ifdef IP_SET_HASH_WITH_NETS
301 /* Network cidr size book keeping when the hash stores different
302 * sized networks. cidr == real cidr + 1 to support /0.
305 mtype_add_cidr(struct htype *h, u8 cidr, u8 n)
309 /* Add in increasing prefix order, so larger cidr first */
310 for (i = 0, j = -1; i < NLEN && h->nets[i].cidr[n]; i++) {
313 } else if (h->nets[i].cidr[n] < cidr) {
315 } else if (h->nets[i].cidr[n] == cidr) {
316 h->nets[CIDR_POS(cidr)].nets[n]++;
322 h->nets[i].cidr[n] = h->nets[i - 1].cidr[n];
324 h->nets[i].cidr[n] = cidr;
325 h->nets[CIDR_POS(cidr)].nets[n] = 1;
329 mtype_del_cidr(struct htype *h, u8 cidr, u8 n)
331 u8 i, j, net_end = NLEN - 1;
333 for (i = 0; i < NLEN; i++) {
334 if (h->nets[i].cidr[n] != cidr)
336 h->nets[CIDR_POS(cidr)].nets[n]--;
337 if (h->nets[CIDR_POS(cidr)].nets[n] > 0)
339 for (j = i; j < net_end && h->nets[j].cidr[n]; j++)
340 h->nets[j].cidr[n] = h->nets[j + 1].cidr[n];
341 h->nets[j].cidr[n] = 0;
347 /* Calculate the actual memory size of the set data */
349 mtype_ahash_memsize(const struct htype *h, const struct htable *t)
351 return sizeof(*h) + sizeof(*t);
354 /* Get the ith element from the array block n */
355 #define ahash_data(n, i, dsize) \
356 ((struct mtype_elem *)((n)->value + ((i) * (dsize))))
359 mtype_ext_cleanup(struct ip_set *set, struct hbucket *n)
363 for (i = 0; i < n->pos; i++)
364 if (test_bit(i, n->used))
365 ip_set_ext_destroy(set, ahash_data(n, i, set->dsize));
368 /* Flush a hash type of set: destroy all elements */
370 mtype_flush(struct ip_set *set)
372 struct htype *h = set->data;
377 t = ipset_dereference_protected(h->table, set);
378 for (i = 0; i < jhash_size(t->htable_bits); i++) {
379 n = __ipset_dereference_protected(hbucket(t, i), 1);
382 if (set->extensions & IPSET_EXT_DESTROY)
383 mtype_ext_cleanup(set, n);
384 /* FIXME: use slab cache */
385 rcu_assign_pointer(hbucket(t, i), NULL);
388 #ifdef IP_SET_HASH_WITH_NETS
389 memset(h->nets, 0, sizeof(h->nets));
395 /* Destroy the hashtable part of the set */
397 mtype_ahash_destroy(struct ip_set *set, struct htable *t, bool ext_destroy)
402 for (i = 0; i < jhash_size(t->htable_bits); i++) {
403 n = __ipset_dereference_protected(hbucket(t, i), 1);
406 if (set->extensions & IPSET_EXT_DESTROY && ext_destroy)
407 mtype_ext_cleanup(set, n);
408 /* FIXME: use slab cache */
415 /* Destroy a hash type of set */
417 mtype_destroy(struct ip_set *set)
419 struct htype *h = set->data;
421 if (SET_WITH_TIMEOUT(set))
422 del_timer_sync(&h->gc);
424 mtype_ahash_destroy(set,
425 __ipset_dereference_protected(h->table, 1), true);
432 mtype_gc_init(struct ip_set *set, void (*gc)(unsigned long ul_set))
434 struct htype *h = set->data;
436 setup_timer(&h->gc, gc, (unsigned long)set);
437 mod_timer(&h->gc, jiffies + IPSET_GC_PERIOD(set->timeout) * HZ);
438 pr_debug("gc initialized, run in every %u\n",
439 IPSET_GC_PERIOD(set->timeout));
443 mtype_same_set(const struct ip_set *a, const struct ip_set *b)
445 const struct htype *x = a->data;
446 const struct htype *y = b->data;
448 /* Resizing changes htable_bits, so we ignore it */
449 return x->maxelem == y->maxelem &&
450 a->timeout == b->timeout &&
451 #ifdef IP_SET_HASH_WITH_NETMASK
452 x->netmask == y->netmask &&
454 #ifdef IP_SET_HASH_WITH_MARKMASK
455 x->markmask == y->markmask &&
457 a->extensions == b->extensions;
460 /* Delete expired elements from the hashtable */
462 mtype_expire(struct ip_set *set, struct htype *h)
465 struct hbucket *n, *tmp;
466 struct mtype_elem *data;
468 size_t dsize = set->dsize;
469 #ifdef IP_SET_HASH_WITH_NETS
473 t = ipset_dereference_protected(h->table, set);
474 for (i = 0; i < jhash_size(t->htable_bits); i++) {
475 n = __ipset_dereference_protected(hbucket(t, i), 1);
478 for (j = 0, d = 0; j < n->pos; j++) {
479 if (!test_bit(j, n->used)) {
483 data = ahash_data(n, j, dsize);
484 if (!ip_set_timeout_expired(ext_timeout(data, set)))
486 pr_debug("expired %u/%u\n", i, j);
487 clear_bit(j, n->used);
488 smp_mb__after_atomic();
489 #ifdef IP_SET_HASH_WITH_NETS
490 for (k = 0; k < IPSET_NET_COUNT; k++)
492 NCIDR_PUT(DCIDR_GET(data->cidr, k)),
495 ip_set_ext_destroy(set, data);
499 if (d >= AHASH_INIT_SIZE) {
501 rcu_assign_pointer(hbucket(t, i), NULL);
505 tmp = kzalloc(sizeof(*tmp) +
506 (n->size - AHASH_INIT_SIZE) * dsize,
509 /* Still try to delete expired elements */
511 tmp->size = n->size - AHASH_INIT_SIZE;
512 for (j = 0, d = 0; j < n->pos; j++) {
513 if (!test_bit(j, n->used))
515 data = ahash_data(n, j, dsize);
516 memcpy(tmp->value + d * dsize, data, dsize);
517 set_bit(d, tmp->used);
521 set->ext_size -= ext_size(AHASH_INIT_SIZE, dsize);
522 rcu_assign_pointer(hbucket(t, i), tmp);
529 mtype_gc(unsigned long ul_set)
531 struct ip_set *set = (struct ip_set *)ul_set;
532 struct htype *h = set->data;
534 pr_debug("called\n");
535 spin_lock_bh(&set->lock);
536 mtype_expire(set, h);
537 spin_unlock_bh(&set->lock);
539 h->gc.expires = jiffies + IPSET_GC_PERIOD(set->timeout) * HZ;
543 /* Resize a hash: create a new hash table with doubling the hashsize
544 * and inserting the elements to it. Repeat until we succeed or
545 * fail due to memory pressures.
548 mtype_resize(struct ip_set *set, bool retried)
550 struct htype *h = set->data;
551 struct htable *t, *orig;
553 size_t extsize, dsize = set->dsize;
554 #ifdef IP_SET_HASH_WITH_NETS
556 struct mtype_elem *tmp;
558 struct mtype_elem *data;
559 struct mtype_elem *d;
560 struct hbucket *n, *m;
564 #ifdef IP_SET_HASH_WITH_NETS
565 tmp = kmalloc(dsize, GFP_KERNEL);
570 orig = rcu_dereference_bh_nfnl(h->table);
571 htable_bits = orig->htable_bits;
572 rcu_read_unlock_bh();
578 /* In case we have plenty of memory :-) */
579 pr_warn("Cannot increase the hashsize of set %s further\n",
581 ret = -IPSET_ERR_HASH_FULL;
584 t = ip_set_alloc(htable_size(htable_bits));
589 t->htable_bits = htable_bits;
591 spin_lock_bh(&set->lock);
592 orig = __ipset_dereference_protected(h->table, 1);
593 /* There can't be another parallel resizing, but dumping is possible */
594 atomic_set(&orig->ref, 1);
595 atomic_inc(&orig->uref);
597 pr_debug("attempt to resize set %s from %u to %u, t %p\n",
598 set->name, orig->htable_bits, htable_bits, orig);
599 for (i = 0; i < jhash_size(orig->htable_bits); i++) {
600 n = __ipset_dereference_protected(hbucket(orig, i), 1);
603 for (j = 0; j < n->pos; j++) {
604 if (!test_bit(j, n->used))
606 data = ahash_data(n, j, dsize);
607 #ifdef IP_SET_HASH_WITH_NETS
608 /* We have readers running parallel with us,
609 * so the live data cannot be modified.
612 memcpy(tmp, data, dsize);
614 mtype_data_reset_flags(data, &flags);
616 key = HKEY(data, h->initval, htable_bits);
617 m = __ipset_dereference_protected(hbucket(t, key), 1);
619 m = kzalloc(sizeof(*m) +
620 AHASH_INIT_SIZE * dsize,
626 m->size = AHASH_INIT_SIZE;
627 extsize = ext_size(AHASH_INIT_SIZE, dsize);
628 RCU_INIT_POINTER(hbucket(t, key), m);
629 } else if (m->pos >= m->size) {
632 if (m->size >= AHASH_MAX(h)) {
635 ht = kzalloc(sizeof(*ht) +
636 (m->size + AHASH_INIT_SIZE)
644 memcpy(ht, m, sizeof(struct hbucket) +
646 ht->size = m->size + AHASH_INIT_SIZE;
647 extsize += ext_size(AHASH_INIT_SIZE, dsize);
650 RCU_INIT_POINTER(hbucket(t, key), ht);
652 d = ahash_data(m, m->pos, dsize);
653 memcpy(d, data, dsize);
654 set_bit(m->pos++, m->used);
655 #ifdef IP_SET_HASH_WITH_NETS
656 mtype_data_reset_flags(d, &flags);
660 rcu_assign_pointer(h->table, t);
661 set->ext_size = extsize;
663 spin_unlock_bh(&set->lock);
665 /* Give time to other readers of the set */
666 synchronize_rcu_bh();
668 pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
669 orig->htable_bits, orig, t->htable_bits, t);
670 /* If there's nobody else dumping the table, destroy it */
671 if (atomic_dec_and_test(&orig->uref)) {
672 pr_debug("Table destroy by resize %p\n", orig);
673 mtype_ahash_destroy(set, orig, false);
677 #ifdef IP_SET_HASH_WITH_NETS
683 atomic_set(&orig->ref, 0);
684 atomic_dec(&orig->uref);
685 spin_unlock_bh(&set->lock);
686 mtype_ahash_destroy(set, t, false);
692 /* Add an element to a hash and update the internal counters when succeeded,
693 * otherwise report the proper error code.
696 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
697 struct ip_set_ext *mext, u32 flags)
699 struct htype *h = set->data;
701 const struct mtype_elem *d = value;
702 struct mtype_elem *data;
703 struct hbucket *n, *old = ERR_PTR(-ENOENT);
705 bool flag_exist = flags & IPSET_FLAG_EXIST;
706 bool deleted = false, forceadd = false, reuse = false;
709 if (set->elements >= h->maxelem) {
710 if (SET_WITH_TIMEOUT(set))
711 /* FIXME: when set is full, we slow down here */
712 mtype_expire(set, h);
713 if (set->elements >= h->maxelem && SET_WITH_FORCEADD(set))
717 t = ipset_dereference_protected(h->table, set);
718 key = HKEY(value, h->initval, t->htable_bits);
719 n = __ipset_dereference_protected(hbucket(t, key), 1);
721 if (forceadd || set->elements >= h->maxelem)
724 n = kzalloc(sizeof(*n) + AHASH_INIT_SIZE * set->dsize,
728 n->size = AHASH_INIT_SIZE;
729 set->ext_size += ext_size(AHASH_INIT_SIZE, set->dsize);
732 for (i = 0; i < n->pos; i++) {
733 if (!test_bit(i, n->used)) {
734 /* Reuse first deleted entry */
736 deleted = reuse = true;
741 data = ahash_data(n, i, set->dsize);
742 if (mtype_data_equal(data, d, &multi)) {
744 (SET_WITH_TIMEOUT(set) &&
745 ip_set_timeout_expired(ext_timeout(data, set)))) {
746 /* Just the extensions could be overwritten */
748 goto overwrite_extensions;
750 return -IPSET_ERR_EXIST;
752 /* Reuse first timed out entry */
753 if (SET_WITH_TIMEOUT(set) &&
754 ip_set_timeout_expired(ext_timeout(data, set)) &&
760 if (reuse || forceadd) {
761 data = ahash_data(n, j, set->dsize);
763 #ifdef IP_SET_HASH_WITH_NETS
764 for (i = 0; i < IPSET_NET_COUNT; i++)
766 NCIDR_PUT(DCIDR_GET(data->cidr, i)),
769 ip_set_ext_destroy(set, data);
774 if (set->elements >= h->maxelem)
776 /* Create a new slot */
777 if (n->pos >= n->size) {
778 TUNE_AHASH_MAX(h, multi);
779 if (n->size >= AHASH_MAX(h)) {
780 /* Trigger rehashing */
781 mtype_data_next(&h->next, d);
785 n = kzalloc(sizeof(*n) +
786 (old->size + AHASH_INIT_SIZE) * set->dsize,
790 memcpy(n, old, sizeof(struct hbucket) +
791 old->size * set->dsize);
792 n->size = old->size + AHASH_INIT_SIZE;
793 set->ext_size += ext_size(AHASH_INIT_SIZE, set->dsize);
798 data = ahash_data(n, j, set->dsize);
801 #ifdef IP_SET_HASH_WITH_NETS
802 for (i = 0; i < IPSET_NET_COUNT; i++)
803 mtype_add_cidr(h, NCIDR_PUT(DCIDR_GET(d->cidr, i)), i);
805 memcpy(data, d, sizeof(struct mtype_elem));
806 overwrite_extensions:
807 #ifdef IP_SET_HASH_WITH_NETS
808 mtype_data_set_flags(data, flags);
810 if (SET_WITH_COUNTER(set))
811 ip_set_init_counter(ext_counter(data, set), ext);
812 if (SET_WITH_COMMENT(set))
813 ip_set_init_comment(set, ext_comment(data, set), ext);
814 if (SET_WITH_SKBINFO(set))
815 ip_set_init_skbinfo(ext_skbinfo(data, set), ext);
816 /* Must come last for the case when timed out entry is reused */
817 if (SET_WITH_TIMEOUT(set))
818 ip_set_timeout_set(ext_timeout(data, set), ext->timeout);
819 smp_mb__before_atomic();
821 if (old != ERR_PTR(-ENOENT)) {
822 rcu_assign_pointer(hbucket(t, key), n);
830 pr_warn("Set %s is full, maxelem %u reached\n",
831 set->name, h->maxelem);
832 return -IPSET_ERR_HASH_FULL;
835 /* Delete an element from the hash and free up space if possible.
838 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
839 struct ip_set_ext *mext, u32 flags)
841 struct htype *h = set->data;
843 const struct mtype_elem *d = value;
844 struct mtype_elem *data;
846 int i, j, k, ret = -IPSET_ERR_EXIST;
848 size_t dsize = set->dsize;
850 t = ipset_dereference_protected(h->table, set);
851 key = HKEY(value, h->initval, t->htable_bits);
852 n = __ipset_dereference_protected(hbucket(t, key), 1);
855 for (i = 0, k = 0; i < n->pos; i++) {
856 if (!test_bit(i, n->used)) {
860 data = ahash_data(n, i, dsize);
861 if (!mtype_data_equal(data, d, &multi))
863 if (SET_WITH_TIMEOUT(set) &&
864 ip_set_timeout_expired(ext_timeout(data, set)))
868 clear_bit(i, n->used);
869 smp_mb__after_atomic();
873 #ifdef IP_SET_HASH_WITH_NETS
874 for (j = 0; j < IPSET_NET_COUNT; j++)
875 mtype_del_cidr(h, NCIDR_PUT(DCIDR_GET(d->cidr, j)),
878 ip_set_ext_destroy(set, data);
880 for (; i < n->pos; i++) {
881 if (!test_bit(i, n->used))
884 if (n->pos == 0 && k == 0) {
885 set->ext_size -= ext_size(n->size, dsize);
886 rcu_assign_pointer(hbucket(t, key), NULL);
888 } else if (k >= AHASH_INIT_SIZE) {
889 struct hbucket *tmp = kzalloc(sizeof(*tmp) +
890 (n->size - AHASH_INIT_SIZE) * dsize,
894 tmp->size = n->size - AHASH_INIT_SIZE;
895 for (j = 0, k = 0; j < n->pos; j++) {
896 if (!test_bit(j, n->used))
898 data = ahash_data(n, j, dsize);
899 memcpy(tmp->value + k * dsize, data, dsize);
900 set_bit(k, tmp->used);
904 set->ext_size -= ext_size(AHASH_INIT_SIZE, dsize);
905 rcu_assign_pointer(hbucket(t, key), tmp);
916 mtype_data_match(struct mtype_elem *data, const struct ip_set_ext *ext,
917 struct ip_set_ext *mext, struct ip_set *set, u32 flags)
919 if (SET_WITH_COUNTER(set))
920 ip_set_update_counter(ext_counter(data, set),
922 if (SET_WITH_SKBINFO(set))
923 ip_set_get_skbinfo(ext_skbinfo(data, set),
925 return mtype_do_data_match(data);
928 #ifdef IP_SET_HASH_WITH_NETS
929 /* Special test function which takes into account the different network
930 * sizes added to the set
933 mtype_test_cidrs(struct ip_set *set, struct mtype_elem *d,
934 const struct ip_set_ext *ext,
935 struct ip_set_ext *mext, u32 flags)
937 struct htype *h = set->data;
938 struct htable *t = rcu_dereference_bh(h->table);
940 struct mtype_elem *data;
941 #if IPSET_NET_COUNT == 2
942 struct mtype_elem orig = *d;
949 pr_debug("test by nets\n");
950 for (; j < NLEN && h->nets[j].cidr[0] && !multi; j++) {
951 #if IPSET_NET_COUNT == 2
952 mtype_data_reset_elem(d, &orig);
953 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]), false);
954 for (k = 0; k < NLEN && h->nets[k].cidr[1] && !multi;
956 mtype_data_netmask(d, NCIDR_GET(h->nets[k].cidr[1]),
959 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]));
961 key = HKEY(d, h->initval, t->htable_bits);
962 n = rcu_dereference_bh(hbucket(t, key));
965 for (i = 0; i < n->pos; i++) {
966 if (!test_bit(i, n->used))
968 data = ahash_data(n, i, set->dsize);
969 if (!mtype_data_equal(data, d, &multi))
971 if (SET_WITH_TIMEOUT(set)) {
972 if (!ip_set_timeout_expired(
973 ext_timeout(data, set)))
974 return mtype_data_match(data, ext,
977 #ifdef IP_SET_HASH_WITH_MULTI
981 return mtype_data_match(data, ext,
984 #if IPSET_NET_COUNT == 2
992 /* Test whether the element is added to the set */
994 mtype_test(struct ip_set *set, void *value, const struct ip_set_ext *ext,
995 struct ip_set_ext *mext, u32 flags)
997 struct htype *h = set->data;
999 struct mtype_elem *d = value;
1001 struct mtype_elem *data;
1005 t = rcu_dereference_bh(h->table);
1006 #ifdef IP_SET_HASH_WITH_NETS
1007 /* If we test an IP address and not a network address,
1008 * try all possible network sizes
1010 for (i = 0; i < IPSET_NET_COUNT; i++)
1011 if (DCIDR_GET(d->cidr, i) != HOST_MASK)
1013 if (i == IPSET_NET_COUNT) {
1014 ret = mtype_test_cidrs(set, d, ext, mext, flags);
1019 key = HKEY(d, h->initval, t->htable_bits);
1020 n = rcu_dereference_bh(hbucket(t, key));
1025 for (i = 0; i < n->pos; i++) {
1026 if (!test_bit(i, n->used))
1028 data = ahash_data(n, i, set->dsize);
1029 if (mtype_data_equal(data, d, &multi) &&
1030 !(SET_WITH_TIMEOUT(set) &&
1031 ip_set_timeout_expired(ext_timeout(data, set)))) {
1032 ret = mtype_data_match(data, ext, mext, set, flags);
1040 /* Reply a HEADER request: fill out the header part of the set */
1042 mtype_head(struct ip_set *set, struct sk_buff *skb)
1044 const struct htype *h = set->data;
1045 const struct htable *t;
1046 struct nlattr *nested;
1051 t = rcu_dereference_bh_nfnl(h->table);
1052 memsize = mtype_ahash_memsize(h, t) + set->ext_size;
1053 htable_bits = t->htable_bits;
1054 rcu_read_unlock_bh();
1056 nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
1058 goto nla_put_failure;
1059 if (nla_put_net32(skb, IPSET_ATTR_HASHSIZE,
1060 htonl(jhash_size(htable_bits))) ||
1061 nla_put_net32(skb, IPSET_ATTR_MAXELEM, htonl(h->maxelem)))
1062 goto nla_put_failure;
1063 #ifdef IP_SET_HASH_WITH_NETMASK
1064 if (h->netmask != HOST_MASK &&
1065 nla_put_u8(skb, IPSET_ATTR_NETMASK, h->netmask))
1066 goto nla_put_failure;
1068 #ifdef IP_SET_HASH_WITH_MARKMASK
1069 if (nla_put_u32(skb, IPSET_ATTR_MARKMASK, h->markmask))
1070 goto nla_put_failure;
1072 if (nla_put_net32(skb, IPSET_ATTR_REFERENCES, htonl(set->ref)) ||
1073 nla_put_net32(skb, IPSET_ATTR_MEMSIZE, htonl(memsize)) ||
1074 nla_put_net32(skb, IPSET_ATTR_ELEMENTS, htonl(set->elements)))
1075 goto nla_put_failure;
1076 if (unlikely(ip_set_put_flags(skb, set)))
1077 goto nla_put_failure;
1078 ipset_nest_end(skb, nested);
1085 /* Make possible to run dumping parallel with resizing */
1087 mtype_uref(struct ip_set *set, struct netlink_callback *cb, bool start)
1089 struct htype *h = set->data;
1094 t = rcu_dereference_bh_nfnl(h->table);
1095 atomic_inc(&t->uref);
1096 cb->args[IPSET_CB_PRIVATE] = (unsigned long)t;
1097 rcu_read_unlock_bh();
1098 } else if (cb->args[IPSET_CB_PRIVATE]) {
1099 t = (struct htable *)cb->args[IPSET_CB_PRIVATE];
1100 if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1101 /* Resizing didn't destroy the hash table */
1102 pr_debug("Table destroy by dump: %p\n", t);
1103 mtype_ahash_destroy(set, t, false);
1105 cb->args[IPSET_CB_PRIVATE] = 0;
1109 /* Reply a LIST/SAVE request: dump the elements of the specified set */
1111 mtype_list(const struct ip_set *set,
1112 struct sk_buff *skb, struct netlink_callback *cb)
1114 const struct htable *t;
1115 struct nlattr *atd, *nested;
1116 const struct hbucket *n;
1117 const struct mtype_elem *e;
1118 u32 first = cb->args[IPSET_CB_ARG0];
1119 /* We assume that one hash bucket fills into one page */
1123 atd = ipset_nest_start(skb, IPSET_ATTR_ADT);
1127 pr_debug("list hash set %s\n", set->name);
1128 t = (const struct htable *)cb->args[IPSET_CB_PRIVATE];
1129 /* Expire may replace a hbucket with another one */
1131 for (; cb->args[IPSET_CB_ARG0] < jhash_size(t->htable_bits);
1132 cb->args[IPSET_CB_ARG0]++) {
1133 incomplete = skb_tail_pointer(skb);
1134 n = rcu_dereference(hbucket(t, cb->args[IPSET_CB_ARG0]));
1135 pr_debug("cb->arg bucket: %lu, t %p n %p\n",
1136 cb->args[IPSET_CB_ARG0], t, n);
1139 for (i = 0; i < n->pos; i++) {
1140 if (!test_bit(i, n->used))
1142 e = ahash_data(n, i, set->dsize);
1143 if (SET_WITH_TIMEOUT(set) &&
1144 ip_set_timeout_expired(ext_timeout(e, set)))
1146 pr_debug("list hash %lu hbucket %p i %u, data %p\n",
1147 cb->args[IPSET_CB_ARG0], n, i, e);
1148 nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
1150 if (cb->args[IPSET_CB_ARG0] == first) {
1151 nla_nest_cancel(skb, atd);
1155 goto nla_put_failure;
1157 if (mtype_data_list(skb, e))
1158 goto nla_put_failure;
1159 if (ip_set_put_extensions(skb, set, e, true))
1160 goto nla_put_failure;
1161 ipset_nest_end(skb, nested);
1164 ipset_nest_end(skb, atd);
1165 /* Set listing finished */
1166 cb->args[IPSET_CB_ARG0] = 0;
1171 nlmsg_trim(skb, incomplete);
1172 if (unlikely(first == cb->args[IPSET_CB_ARG0])) {
1173 pr_warn("Can't list set %s: one bucket does not fit into a message. Please report it!\n",
1175 cb->args[IPSET_CB_ARG0] = 0;
1178 ipset_nest_end(skb, atd);
1186 IPSET_TOKEN(MTYPE, _kadt)(struct ip_set *set, const struct sk_buff *skb,
1187 const struct xt_action_param *par,
1188 enum ipset_adt adt, struct ip_set_adt_opt *opt);
1191 IPSET_TOKEN(MTYPE, _uadt)(struct ip_set *set, struct nlattr *tb[],
1192 enum ipset_adt adt, u32 *lineno, u32 flags,
1195 static const struct ip_set_type_variant mtype_variant = {
1199 [IPSET_ADD] = mtype_add,
1200 [IPSET_DEL] = mtype_del,
1201 [IPSET_TEST] = mtype_test,
1203 .destroy = mtype_destroy,
1204 .flush = mtype_flush,
1208 .resize = mtype_resize,
1209 .same_set = mtype_same_set,
1212 #ifdef IP_SET_EMIT_CREATE
1214 IPSET_TOKEN(HTYPE, _create)(struct net *net, struct ip_set *set,
1215 struct nlattr *tb[], u32 flags)
1217 u32 hashsize = IPSET_DEFAULT_HASHSIZE, maxelem = IPSET_DEFAULT_MAXELEM;
1218 #ifdef IP_SET_HASH_WITH_MARKMASK
1222 #ifdef IP_SET_HASH_WITH_NETMASK
1229 pr_debug("Create set %s with family %s\n",
1230 set->name, set->family == NFPROTO_IPV4 ? "inet" : "inet6");
1232 #ifndef IP_SET_PROTO_UNDEF
1233 if (!(set->family == NFPROTO_IPV4 || set->family == NFPROTO_IPV6))
1234 return -IPSET_ERR_INVALID_FAMILY;
1237 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_HASHSIZE) ||
1238 !ip_set_optattr_netorder(tb, IPSET_ATTR_MAXELEM) ||
1239 !ip_set_optattr_netorder(tb, IPSET_ATTR_TIMEOUT) ||
1240 !ip_set_optattr_netorder(tb, IPSET_ATTR_CADT_FLAGS)))
1241 return -IPSET_ERR_PROTOCOL;
1243 #ifdef IP_SET_HASH_WITH_MARKMASK
1244 /* Separated condition in order to avoid directive in argument list */
1245 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_MARKMASK)))
1246 return -IPSET_ERR_PROTOCOL;
1248 markmask = 0xffffffff;
1249 if (tb[IPSET_ATTR_MARKMASK]) {
1250 markmask = ntohl(nla_get_be32(tb[IPSET_ATTR_MARKMASK]));
1252 return -IPSET_ERR_INVALID_MARKMASK;
1256 #ifdef IP_SET_HASH_WITH_NETMASK
1257 netmask = set->family == NFPROTO_IPV4 ? 32 : 128;
1258 if (tb[IPSET_ATTR_NETMASK]) {
1259 netmask = nla_get_u8(tb[IPSET_ATTR_NETMASK]);
1261 if ((set->family == NFPROTO_IPV4 && netmask > 32) ||
1262 (set->family == NFPROTO_IPV6 && netmask > 128) ||
1264 return -IPSET_ERR_INVALID_NETMASK;
1268 if (tb[IPSET_ATTR_HASHSIZE]) {
1269 hashsize = ip_set_get_h32(tb[IPSET_ATTR_HASHSIZE]);
1270 if (hashsize < IPSET_MIMINAL_HASHSIZE)
1271 hashsize = IPSET_MIMINAL_HASHSIZE;
1274 if (tb[IPSET_ATTR_MAXELEM])
1275 maxelem = ip_set_get_h32(tb[IPSET_ATTR_MAXELEM]);
1278 h = kzalloc(hsize, GFP_KERNEL);
1282 hbits = htable_bits(hashsize);
1283 hsize = htable_size(hbits);
1288 t = ip_set_alloc(hsize);
1293 h->maxelem = maxelem;
1294 #ifdef IP_SET_HASH_WITH_NETMASK
1295 h->netmask = netmask;
1297 #ifdef IP_SET_HASH_WITH_MARKMASK
1298 h->markmask = markmask;
1300 get_random_bytes(&h->initval, sizeof(h->initval));
1302 t->htable_bits = hbits;
1303 RCU_INIT_POINTER(h->table, t);
1306 #ifndef IP_SET_PROTO_UNDEF
1307 if (set->family == NFPROTO_IPV4) {
1309 set->variant = &IPSET_TOKEN(HTYPE, 4_variant);
1310 set->dsize = ip_set_elem_len(set, tb,
1311 sizeof(struct IPSET_TOKEN(HTYPE, 4_elem)),
1312 __alignof__(struct IPSET_TOKEN(HTYPE, 4_elem)));
1313 #ifndef IP_SET_PROTO_UNDEF
1315 set->variant = &IPSET_TOKEN(HTYPE, 6_variant);
1316 set->dsize = ip_set_elem_len(set, tb,
1317 sizeof(struct IPSET_TOKEN(HTYPE, 6_elem)),
1318 __alignof__(struct IPSET_TOKEN(HTYPE, 6_elem)));
1321 set->timeout = IPSET_NO_TIMEOUT;
1322 if (tb[IPSET_ATTR_TIMEOUT]) {
1323 set->timeout = ip_set_timeout_uget(tb[IPSET_ATTR_TIMEOUT]);
1324 #ifndef IP_SET_PROTO_UNDEF
1325 if (set->family == NFPROTO_IPV4)
1327 IPSET_TOKEN(HTYPE, 4_gc_init)(set,
1328 IPSET_TOKEN(HTYPE, 4_gc));
1329 #ifndef IP_SET_PROTO_UNDEF
1331 IPSET_TOKEN(HTYPE, 6_gc_init)(set,
1332 IPSET_TOKEN(HTYPE, 6_gc));
1335 pr_debug("create %s hashsize %u (%u) maxelem %u: %p(%p)\n",
1336 set->name, jhash_size(t->htable_bits),
1337 t->htable_bits, h->maxelem, set->data, t);
1341 #endif /* IP_SET_EMIT_CREATE */