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])
89 #ifndef IPSET_NET_COUNT
90 #define IPSET_NET_COUNT 1
93 /* Book-keeping of the prefixes added to the set */
95 u32 nets[IPSET_NET_COUNT]; /* number of elements for this cidr */
96 u8 cidr[IPSET_NET_COUNT]; /* the cidr value */
99 /* Compute the hash table size */
101 htable_size(u8 hbits)
105 /* We must fit both into u32 in jhash and size_t */
108 hsize = jhash_size(hbits);
109 if ((((size_t)-1) - sizeof(struct htable)) / sizeof(struct hbucket *)
113 return hsize * sizeof(struct hbucket *) + sizeof(struct htable);
116 /* Compute htable_bits from the user input parameter hashsize */
118 htable_bits(u32 hashsize)
120 /* Assume that hashsize == 2^htable_bits */
121 u8 bits = fls(hashsize - 1);
123 if (jhash_size(bits) != hashsize)
124 /* Round up to the first 2^n value */
125 bits = fls(hashsize);
130 #ifdef IP_SET_HASH_WITH_NETS
131 #if IPSET_NET_COUNT > 1
132 #define __CIDR(cidr, i) (cidr[i])
134 #define __CIDR(cidr, i) (cidr)
137 /* cidr + 1 is stored in net_prefixes to support /0 */
138 #define NCIDR_PUT(cidr) ((cidr) + 1)
139 #define NCIDR_GET(cidr) ((cidr) - 1)
141 #ifdef IP_SET_HASH_WITH_NETS_PACKED
142 /* When cidr is packed with nomatch, cidr - 1 is stored in the data entry */
143 #define DCIDR_PUT(cidr) ((cidr) - 1)
144 #define DCIDR_GET(cidr, i) (__CIDR(cidr, i) + 1)
146 #define DCIDR_PUT(cidr) (cidr)
147 #define DCIDR_GET(cidr, i) __CIDR(cidr, i)
150 #define INIT_CIDR(cidr, host_mask) \
151 DCIDR_PUT(((cidr) ? NCIDR_GET(cidr) : host_mask))
153 #define SET_HOST_MASK(family) (family == AF_INET ? 32 : 128)
155 #ifdef IP_SET_HASH_WITH_NET0
156 /* cidr from 0 to SET_HOST_MASK() value and c = cidr + 1 */
157 #define NLEN(family) (SET_HOST_MASK(family) + 1)
158 #define CIDR_POS(c) ((c) - 1)
160 /* cidr from 1 to SET_HOST_MASK() value and c = cidr + 1 */
161 #define NLEN(family) SET_HOST_MASK(family)
162 #define CIDR_POS(c) ((c) - 2)
166 #define NLEN(family) 0
167 #endif /* IP_SET_HASH_WITH_NETS */
169 #endif /* _IP_SET_HASH_GEN_H */
171 /* Family dependent templates */
174 #undef mtype_data_equal
175 #undef mtype_do_data_match
176 #undef mtype_data_set_flags
177 #undef mtype_data_reset_elem
178 #undef mtype_data_reset_flags
179 #undef mtype_data_netmask
180 #undef mtype_data_list
181 #undef mtype_data_next
184 #undef mtype_ahash_destroy
185 #undef mtype_ext_cleanup
186 #undef mtype_add_cidr
187 #undef mtype_del_cidr
188 #undef mtype_ahash_memsize
191 #undef mtype_same_set
198 #undef mtype_test_cidrs
208 #undef mtype_data_match
212 #define mtype_data_equal IPSET_TOKEN(MTYPE, _data_equal)
213 #ifdef IP_SET_HASH_WITH_NETS
214 #define mtype_do_data_match IPSET_TOKEN(MTYPE, _do_data_match)
216 #define mtype_do_data_match(d) 1
218 #define mtype_data_set_flags IPSET_TOKEN(MTYPE, _data_set_flags)
219 #define mtype_data_reset_elem IPSET_TOKEN(MTYPE, _data_reset_elem)
220 #define mtype_data_reset_flags IPSET_TOKEN(MTYPE, _data_reset_flags)
221 #define mtype_data_netmask IPSET_TOKEN(MTYPE, _data_netmask)
222 #define mtype_data_list IPSET_TOKEN(MTYPE, _data_list)
223 #define mtype_data_next IPSET_TOKEN(MTYPE, _data_next)
224 #define mtype_elem IPSET_TOKEN(MTYPE, _elem)
226 #define mtype_ahash_destroy IPSET_TOKEN(MTYPE, _ahash_destroy)
227 #define mtype_ext_cleanup IPSET_TOKEN(MTYPE, _ext_cleanup)
228 #define mtype_add_cidr IPSET_TOKEN(MTYPE, _add_cidr)
229 #define mtype_del_cidr IPSET_TOKEN(MTYPE, _del_cidr)
230 #define mtype_ahash_memsize IPSET_TOKEN(MTYPE, _ahash_memsize)
231 #define mtype_flush IPSET_TOKEN(MTYPE, _flush)
232 #define mtype_destroy IPSET_TOKEN(MTYPE, _destroy)
233 #define mtype_same_set IPSET_TOKEN(MTYPE, _same_set)
234 #define mtype_kadt IPSET_TOKEN(MTYPE, _kadt)
235 #define mtype_uadt IPSET_TOKEN(MTYPE, _uadt)
238 #define mtype_add IPSET_TOKEN(MTYPE, _add)
239 #define mtype_del IPSET_TOKEN(MTYPE, _del)
240 #define mtype_test_cidrs IPSET_TOKEN(MTYPE, _test_cidrs)
241 #define mtype_test IPSET_TOKEN(MTYPE, _test)
242 #define mtype_uref IPSET_TOKEN(MTYPE, _uref)
243 #define mtype_expire IPSET_TOKEN(MTYPE, _expire)
244 #define mtype_resize IPSET_TOKEN(MTYPE, _resize)
245 #define mtype_head IPSET_TOKEN(MTYPE, _head)
246 #define mtype_list IPSET_TOKEN(MTYPE, _list)
247 #define mtype_gc IPSET_TOKEN(MTYPE, _gc)
248 #define mtype_gc_init IPSET_TOKEN(MTYPE, _gc_init)
249 #define mtype_variant IPSET_TOKEN(MTYPE, _variant)
250 #define mtype_data_match IPSET_TOKEN(MTYPE, _data_match)
253 #error "MTYPE is not defined!"
257 #error "HOST_MASK is not defined!"
261 #define HKEY_DATALEN sizeof(struct mtype_elem)
264 #define HKEY(data, initval, htable_bits) \
265 (jhash2((u32 *)(data), HKEY_DATALEN / sizeof(u32), initval) \
266 & jhash_mask(htable_bits))
270 #error "HTYPE is not defined!"
274 /* The generic hash structure */
276 struct htable __rcu *table; /* the hash table */
277 u32 maxelem; /* max elements in the hash */
278 u32 elements; /* current element (vs timeout) */
279 u32 initval; /* random jhash init value */
280 #ifdef IP_SET_HASH_WITH_MARKMASK
281 u32 markmask; /* markmask value for mark mask to store */
283 struct timer_list gc; /* garbage collection when timeout enabled */
284 struct mtype_elem next; /* temporary storage for uadd */
285 #ifdef IP_SET_HASH_WITH_MULTI
286 u8 ahash_max; /* max elements in an array block */
288 #ifdef IP_SET_HASH_WITH_NETMASK
289 u8 netmask; /* netmask value for subnets to store */
291 #ifdef IP_SET_HASH_WITH_NETS
292 struct net_prefixes nets[0]; /* book-keeping of prefixes */
297 #ifdef IP_SET_HASH_WITH_NETS
298 /* Network cidr size book keeping when the hash stores different
299 * sized networks. cidr == real cidr + 1 to support /0.
302 mtype_add_cidr(struct htype *h, u8 cidr, u8 nets_length, u8 n)
306 /* Add in increasing prefix order, so larger cidr first */
307 for (i = 0, j = -1; i < nets_length && h->nets[i].cidr[n]; i++) {
310 } else if (h->nets[i].cidr[n] < cidr) {
312 } else if (h->nets[i].cidr[n] == cidr) {
313 h->nets[CIDR_POS(cidr)].nets[n]++;
319 h->nets[i].cidr[n] = h->nets[i - 1].cidr[n];
321 h->nets[i].cidr[n] = cidr;
322 h->nets[CIDR_POS(cidr)].nets[n] = 1;
326 mtype_del_cidr(struct htype *h, u8 cidr, u8 nets_length, u8 n)
328 u8 i, j, net_end = nets_length - 1;
330 for (i = 0; i < nets_length; i++) {
331 if (h->nets[i].cidr[n] != cidr)
333 h->nets[CIDR_POS(cidr)].nets[n]--;
334 if (h->nets[CIDR_POS(cidr)].nets[n] > 0)
336 for (j = i; j < net_end && h->nets[j].cidr[n]; j++)
337 h->nets[j].cidr[n] = h->nets[j + 1].cidr[n];
338 h->nets[j].cidr[n] = 0;
344 /* Calculate the actual memory size of the set data */
346 mtype_ahash_memsize(const struct htype *h, const struct htable *t,
347 u8 nets_length, size_t dsize)
351 size_t memsize = sizeof(*h) + sizeof(*t);
353 #ifdef IP_SET_HASH_WITH_NETS
354 memsize += sizeof(struct net_prefixes) * nets_length;
356 for (i = 0; i < jhash_size(t->htable_bits); i++) {
357 n = rcu_dereference_bh(hbucket(t, i));
360 memsize += sizeof(struct hbucket) + n->size * dsize;
366 /* Get the ith element from the array block n */
367 #define ahash_data(n, i, dsize) \
368 ((struct mtype_elem *)((n)->value + ((i) * (dsize))))
371 mtype_ext_cleanup(struct ip_set *set, struct hbucket *n)
375 for (i = 0; i < n->pos; i++)
376 if (test_bit(i, n->used))
377 ip_set_ext_destroy(set, ahash_data(n, i, set->dsize));
380 /* Flush a hash type of set: destroy all elements */
382 mtype_flush(struct ip_set *set)
384 struct htype *h = set->data;
389 t = ipset_dereference_protected(h->table, set);
390 for (i = 0; i < jhash_size(t->htable_bits); i++) {
391 n = __ipset_dereference_protected(hbucket(t, i), 1);
394 if (set->extensions & IPSET_EXT_DESTROY)
395 mtype_ext_cleanup(set, n);
396 /* FIXME: use slab cache */
397 rcu_assign_pointer(hbucket(t, i), NULL);
400 #ifdef IP_SET_HASH_WITH_NETS
401 memset(h->nets, 0, sizeof(struct net_prefixes) * NLEN(set->family));
406 /* Destroy the hashtable part of the set */
408 mtype_ahash_destroy(struct ip_set *set, struct htable *t, bool ext_destroy)
413 for (i = 0; i < jhash_size(t->htable_bits); i++) {
414 n = __ipset_dereference_protected(hbucket(t, i), 1);
417 if (set->extensions & IPSET_EXT_DESTROY && ext_destroy)
418 mtype_ext_cleanup(set, n);
419 /* FIXME: use slab cache */
426 /* Destroy a hash type of set */
428 mtype_destroy(struct ip_set *set)
430 struct htype *h = set->data;
432 if (SET_WITH_TIMEOUT(set))
433 del_timer_sync(&h->gc);
435 mtype_ahash_destroy(set,
436 __ipset_dereference_protected(h->table, 1), true);
443 mtype_gc_init(struct ip_set *set, void (*gc)(unsigned long ul_set))
445 struct htype *h = set->data;
448 h->gc.data = (unsigned long)set;
450 h->gc.expires = jiffies + IPSET_GC_PERIOD(set->timeout) * HZ;
452 pr_debug("gc initialized, run in every %u\n",
453 IPSET_GC_PERIOD(set->timeout));
457 mtype_same_set(const struct ip_set *a, const struct ip_set *b)
459 const struct htype *x = a->data;
460 const struct htype *y = b->data;
462 /* Resizing changes htable_bits, so we ignore it */
463 return x->maxelem == y->maxelem &&
464 a->timeout == b->timeout &&
465 #ifdef IP_SET_HASH_WITH_NETMASK
466 x->netmask == y->netmask &&
468 #ifdef IP_SET_HASH_WITH_MARKMASK
469 x->markmask == y->markmask &&
471 a->extensions == b->extensions;
474 /* Delete expired elements from the hashtable */
476 mtype_expire(struct ip_set *set, struct htype *h, u8 nets_length, size_t dsize)
479 struct hbucket *n, *tmp;
480 struct mtype_elem *data;
482 #ifdef IP_SET_HASH_WITH_NETS
486 t = ipset_dereference_protected(h->table, set);
487 for (i = 0; i < jhash_size(t->htable_bits); i++) {
488 n = __ipset_dereference_protected(hbucket(t, i), 1);
491 for (j = 0, d = 0; j < n->pos; j++) {
492 if (!test_bit(j, n->used)) {
496 data = ahash_data(n, j, dsize);
497 if (ip_set_timeout_expired(ext_timeout(data, set))) {
498 pr_debug("expired %u/%u\n", i, j);
499 clear_bit(j, n->used);
500 smp_mb__after_atomic();
501 #ifdef IP_SET_HASH_WITH_NETS
502 for (k = 0; k < IPSET_NET_COUNT; k++)
504 NCIDR_PUT(DCIDR_GET(data->cidr,
508 ip_set_ext_destroy(set, data);
513 if (d >= AHASH_INIT_SIZE) {
515 rcu_assign_pointer(hbucket(t, i), NULL);
519 tmp = kzalloc(sizeof(*tmp) +
520 (n->size - AHASH_INIT_SIZE) * dsize,
523 /* Still try to delete expired elements */
525 tmp->size = n->size - AHASH_INIT_SIZE;
526 for (j = 0, d = 0; j < n->pos; j++) {
527 if (!test_bit(j, n->used))
529 data = ahash_data(n, j, dsize);
530 memcpy(tmp->value + d * dsize, data, dsize);
531 set_bit(d, tmp->used);
535 rcu_assign_pointer(hbucket(t, i), tmp);
542 mtype_gc(unsigned long ul_set)
544 struct ip_set *set = (struct ip_set *)ul_set;
545 struct htype *h = set->data;
547 pr_debug("called\n");
548 spin_lock_bh(&set->lock);
549 mtype_expire(set, h, NLEN(set->family), set->dsize);
550 spin_unlock_bh(&set->lock);
552 h->gc.expires = jiffies + IPSET_GC_PERIOD(set->timeout) * HZ;
556 /* Resize a hash: create a new hash table with doubling the hashsize
557 * and inserting the elements to it. Repeat until we succeed or
558 * fail due to memory pressures.
561 mtype_resize(struct ip_set *set, bool retried)
563 struct htype *h = set->data;
564 struct htable *t, *orig;
566 size_t dsize = set->dsize;
567 #ifdef IP_SET_HASH_WITH_NETS
569 struct mtype_elem *tmp;
571 struct mtype_elem *data;
572 struct mtype_elem *d;
573 struct hbucket *n, *m;
577 #ifdef IP_SET_HASH_WITH_NETS
578 tmp = kmalloc(dsize, GFP_KERNEL);
583 orig = rcu_dereference_bh_nfnl(h->table);
584 htable_bits = orig->htable_bits;
585 rcu_read_unlock_bh();
591 /* In case we have plenty of memory :-) */
592 pr_warn("Cannot increase the hashsize of set %s further\n",
594 ret = -IPSET_ERR_HASH_FULL;
597 t = ip_set_alloc(htable_size(htable_bits));
602 t->htable_bits = htable_bits;
604 spin_lock_bh(&set->lock);
605 orig = __ipset_dereference_protected(h->table, 1);
606 /* There can't be another parallel resizing, but dumping is possible */
607 atomic_set(&orig->ref, 1);
608 atomic_inc(&orig->uref);
609 pr_debug("attempt to resize set %s from %u to %u, t %p\n",
610 set->name, orig->htable_bits, htable_bits, orig);
611 for (i = 0; i < jhash_size(orig->htable_bits); i++) {
612 n = __ipset_dereference_protected(hbucket(orig, i), 1);
615 for (j = 0; j < n->pos; j++) {
616 if (!test_bit(j, n->used))
618 data = ahash_data(n, j, dsize);
619 #ifdef IP_SET_HASH_WITH_NETS
620 /* We have readers running parallel with us,
621 * so the live data cannot be modified.
624 memcpy(tmp, data, dsize);
626 mtype_data_reset_flags(data, &flags);
628 key = HKEY(data, h->initval, htable_bits);
629 m = __ipset_dereference_protected(hbucket(t, key), 1);
631 m = kzalloc(sizeof(*m) +
632 AHASH_INIT_SIZE * dsize,
638 m->size = AHASH_INIT_SIZE;
639 RCU_INIT_POINTER(hbucket(t, key), m);
640 } else if (m->pos >= m->size) {
643 if (m->size >= AHASH_MAX(h)) {
646 ht = kzalloc(sizeof(*ht) +
647 (m->size + AHASH_INIT_SIZE)
655 memcpy(ht, m, sizeof(struct hbucket) +
657 ht->size = m->size + AHASH_INIT_SIZE;
660 RCU_INIT_POINTER(hbucket(t, key), ht);
662 d = ahash_data(m, m->pos, dsize);
663 memcpy(d, data, dsize);
664 set_bit(m->pos++, m->used);
665 #ifdef IP_SET_HASH_WITH_NETS
666 mtype_data_reset_flags(d, &flags);
670 rcu_assign_pointer(h->table, t);
672 spin_unlock_bh(&set->lock);
674 /* Give time to other readers of the set */
675 synchronize_rcu_bh();
677 pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
678 orig->htable_bits, orig, t->htable_bits, t);
679 /* If there's nobody else dumping the table, destroy it */
680 if (atomic_dec_and_test(&orig->uref)) {
681 pr_debug("Table destroy by resize %p\n", orig);
682 mtype_ahash_destroy(set, orig, false);
686 #ifdef IP_SET_HASH_WITH_NETS
692 atomic_set(&orig->ref, 0);
693 atomic_dec(&orig->uref);
694 spin_unlock_bh(&set->lock);
695 mtype_ahash_destroy(set, t, false);
701 /* Add an element to a hash and update the internal counters when succeeded,
702 * otherwise report the proper error code.
705 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
706 struct ip_set_ext *mext, u32 flags)
708 struct htype *h = set->data;
710 const struct mtype_elem *d = value;
711 struct mtype_elem *data;
712 struct hbucket *n, *old = ERR_PTR(-ENOENT);
714 bool flag_exist = flags & IPSET_FLAG_EXIST;
715 bool deleted = false, forceadd = false, reuse = false;
718 if (h->elements >= h->maxelem) {
719 if (SET_WITH_TIMEOUT(set))
720 /* FIXME: when set is full, we slow down here */
721 mtype_expire(set, h, NLEN(set->family), set->dsize);
722 if (h->elements >= h->maxelem && SET_WITH_FORCEADD(set))
726 t = ipset_dereference_protected(h->table, set);
727 key = HKEY(value, h->initval, t->htable_bits);
728 n = __ipset_dereference_protected(hbucket(t, key), 1);
732 pr_warn("Set %s is full, maxelem %u reached\n",
733 set->name, h->maxelem);
734 return -IPSET_ERR_HASH_FULL;
735 } else if (h->elements >= h->maxelem) {
739 n = kzalloc(sizeof(*n) + AHASH_INIT_SIZE * set->dsize,
743 n->size = AHASH_INIT_SIZE;
746 for (i = 0; i < n->pos; i++) {
747 if (!test_bit(i, n->used)) {
748 /* Reuse first deleted entry */
750 deleted = reuse = true;
755 data = ahash_data(n, i, set->dsize);
756 if (mtype_data_equal(data, d, &multi)) {
758 (SET_WITH_TIMEOUT(set) &&
759 ip_set_timeout_expired(ext_timeout(data, set)))) {
760 /* Just the extensions could be overwritten */
762 goto overwrite_extensions;
764 return -IPSET_ERR_EXIST;
766 /* Reuse first timed out entry */
767 if (SET_WITH_TIMEOUT(set) &&
768 ip_set_timeout_expired(ext_timeout(data, set)) &&
774 if (reuse || forceadd) {
775 data = ahash_data(n, j, set->dsize);
777 #ifdef IP_SET_HASH_WITH_NETS
778 for (i = 0; i < IPSET_NET_COUNT; i++)
780 NCIDR_PUT(DCIDR_GET(data->cidr, i)),
781 NLEN(set->family), i);
783 ip_set_ext_destroy(set, data);
788 if (h->elements >= h->maxelem)
790 /* Create a new slot */
791 if (n->pos >= n->size) {
792 TUNE_AHASH_MAX(h, multi);
793 if (n->size >= AHASH_MAX(h)) {
794 /* Trigger rehashing */
795 mtype_data_next(&h->next, d);
799 n = kzalloc(sizeof(*n) +
800 (old->size + AHASH_INIT_SIZE) * set->dsize,
804 memcpy(n, old, sizeof(struct hbucket) +
805 old->size * set->dsize);
806 n->size = old->size + AHASH_INIT_SIZE;
811 data = ahash_data(n, j, set->dsize);
814 #ifdef IP_SET_HASH_WITH_NETS
815 for (i = 0; i < IPSET_NET_COUNT; i++)
816 mtype_add_cidr(h, NCIDR_PUT(DCIDR_GET(d->cidr, i)),
817 NLEN(set->family), i);
819 memcpy(data, d, sizeof(struct mtype_elem));
820 overwrite_extensions:
821 #ifdef IP_SET_HASH_WITH_NETS
822 mtype_data_set_flags(data, flags);
824 if (SET_WITH_COUNTER(set))
825 ip_set_init_counter(ext_counter(data, set), ext);
826 if (SET_WITH_COMMENT(set))
827 ip_set_init_comment(ext_comment(data, set), ext);
828 if (SET_WITH_SKBINFO(set))
829 ip_set_init_skbinfo(ext_skbinfo(data, set), ext);
830 /* Must come last for the case when timed out entry is reused */
831 if (SET_WITH_TIMEOUT(set))
832 ip_set_timeout_set(ext_timeout(data, set), ext->timeout);
833 smp_mb__before_atomic();
835 if (old != ERR_PTR(-ENOENT)) {
836 rcu_assign_pointer(hbucket(t, key), n);
844 pr_warn("Set %s is full, maxelem %u reached\n",
845 set->name, h->maxelem);
846 return -IPSET_ERR_HASH_FULL;
849 /* Delete an element from the hash and free up space if possible.
852 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
853 struct ip_set_ext *mext, u32 flags)
855 struct htype *h = set->data;
857 const struct mtype_elem *d = value;
858 struct mtype_elem *data;
860 int i, j, k, ret = -IPSET_ERR_EXIST;
862 size_t dsize = set->dsize;
864 t = ipset_dereference_protected(h->table, set);
865 key = HKEY(value, h->initval, t->htable_bits);
866 n = __ipset_dereference_protected(hbucket(t, key), 1);
869 for (i = 0, k = 0; i < n->pos; i++) {
870 if (!test_bit(i, n->used)) {
874 data = ahash_data(n, i, dsize);
875 if (!mtype_data_equal(data, d, &multi))
877 if (SET_WITH_TIMEOUT(set) &&
878 ip_set_timeout_expired(ext_timeout(data, set)))
882 clear_bit(i, n->used);
883 smp_mb__after_atomic();
887 #ifdef IP_SET_HASH_WITH_NETS
888 for (j = 0; j < IPSET_NET_COUNT; j++)
889 mtype_del_cidr(h, NCIDR_PUT(DCIDR_GET(d->cidr, j)),
890 NLEN(set->family), j);
892 ip_set_ext_destroy(set, data);
894 for (; i < n->pos; i++) {
895 if (!test_bit(i, n->used))
898 if (n->pos == 0 && k == 0) {
899 rcu_assign_pointer(hbucket(t, key), NULL);
901 } else if (k >= AHASH_INIT_SIZE) {
902 struct hbucket *tmp = kzalloc(sizeof(*tmp) +
903 (n->size - AHASH_INIT_SIZE) * dsize,
907 tmp->size = n->size - AHASH_INIT_SIZE;
908 for (j = 0, k = 0; j < n->pos; j++) {
909 if (!test_bit(j, n->used))
911 data = ahash_data(n, j, dsize);
912 memcpy(tmp->value + k * dsize, data, dsize);
913 set_bit(j, tmp->used);
917 rcu_assign_pointer(hbucket(t, key), tmp);
928 mtype_data_match(struct mtype_elem *data, const struct ip_set_ext *ext,
929 struct ip_set_ext *mext, struct ip_set *set, u32 flags)
931 if (SET_WITH_COUNTER(set))
932 ip_set_update_counter(ext_counter(data, set),
934 if (SET_WITH_SKBINFO(set))
935 ip_set_get_skbinfo(ext_skbinfo(data, set),
937 return mtype_do_data_match(data);
940 #ifdef IP_SET_HASH_WITH_NETS
941 /* Special test function which takes into account the different network
942 * sizes added to the set
945 mtype_test_cidrs(struct ip_set *set, struct mtype_elem *d,
946 const struct ip_set_ext *ext,
947 struct ip_set_ext *mext, u32 flags)
949 struct htype *h = set->data;
950 struct htable *t = rcu_dereference_bh(h->table);
952 struct mtype_elem *data;
953 #if IPSET_NET_COUNT == 2
954 struct mtype_elem orig = *d;
960 u8 nets_length = NLEN(set->family);
962 pr_debug("test by nets\n");
963 for (; j < nets_length && h->nets[j].cidr[0] && !multi; j++) {
964 #if IPSET_NET_COUNT == 2
965 mtype_data_reset_elem(d, &orig);
966 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]), false);
967 for (k = 0; k < nets_length && h->nets[k].cidr[1] && !multi;
969 mtype_data_netmask(d, NCIDR_GET(h->nets[k].cidr[1]),
972 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]));
974 key = HKEY(d, h->initval, t->htable_bits);
975 n = rcu_dereference_bh(hbucket(t, key));
978 for (i = 0; i < n->pos; i++) {
979 if (!test_bit(i, n->used))
981 data = ahash_data(n, i, set->dsize);
982 if (!mtype_data_equal(data, d, &multi))
984 if (SET_WITH_TIMEOUT(set)) {
985 if (!ip_set_timeout_expired(
986 ext_timeout(data, set)))
987 return mtype_data_match(data, ext,
990 #ifdef IP_SET_HASH_WITH_MULTI
994 return mtype_data_match(data, ext,
997 #if IPSET_NET_COUNT == 2
1005 /* Test whether the element is added to the set */
1007 mtype_test(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1008 struct ip_set_ext *mext, u32 flags)
1010 struct htype *h = set->data;
1012 struct mtype_elem *d = value;
1014 struct mtype_elem *data;
1018 t = rcu_dereference_bh(h->table);
1019 #ifdef IP_SET_HASH_WITH_NETS
1020 /* If we test an IP address and not a network address,
1021 * try all possible network sizes
1023 for (i = 0; i < IPSET_NET_COUNT; i++)
1024 if (DCIDR_GET(d->cidr, i) != SET_HOST_MASK(set->family))
1026 if (i == IPSET_NET_COUNT) {
1027 ret = mtype_test_cidrs(set, d, ext, mext, flags);
1032 key = HKEY(d, h->initval, t->htable_bits);
1033 n = rcu_dereference_bh(hbucket(t, key));
1038 for (i = 0; i < n->pos; i++) {
1039 if (!test_bit(i, n->used))
1041 data = ahash_data(n, i, set->dsize);
1042 if (mtype_data_equal(data, d, &multi) &&
1043 !(SET_WITH_TIMEOUT(set) &&
1044 ip_set_timeout_expired(ext_timeout(data, set)))) {
1045 ret = mtype_data_match(data, ext, mext, set, flags);
1053 /* Reply a HEADER request: fill out the header part of the set */
1055 mtype_head(struct ip_set *set, struct sk_buff *skb)
1057 const struct htype *h = set->data;
1058 const struct htable *t;
1059 struct nlattr *nested;
1064 t = rcu_dereference_bh_nfnl(h->table);
1065 memsize = mtype_ahash_memsize(h, t, NLEN(set->family), set->dsize);
1066 htable_bits = t->htable_bits;
1067 rcu_read_unlock_bh();
1069 nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
1071 goto nla_put_failure;
1072 if (nla_put_net32(skb, IPSET_ATTR_HASHSIZE,
1073 htonl(jhash_size(htable_bits))) ||
1074 nla_put_net32(skb, IPSET_ATTR_MAXELEM, htonl(h->maxelem)))
1075 goto nla_put_failure;
1076 #ifdef IP_SET_HASH_WITH_NETMASK
1077 if (h->netmask != HOST_MASK &&
1078 nla_put_u8(skb, IPSET_ATTR_NETMASK, h->netmask))
1079 goto nla_put_failure;
1081 #ifdef IP_SET_HASH_WITH_MARKMASK
1082 if (nla_put_u32(skb, IPSET_ATTR_MARKMASK, h->markmask))
1083 goto nla_put_failure;
1085 if (nla_put_net32(skb, IPSET_ATTR_REFERENCES, htonl(set->ref - 1)) ||
1086 nla_put_net32(skb, IPSET_ATTR_MEMSIZE, htonl(memsize)))
1087 goto nla_put_failure;
1088 if (unlikely(ip_set_put_flags(skb, set)))
1089 goto nla_put_failure;
1090 ipset_nest_end(skb, nested);
1097 /* Make possible to run dumping parallel with resizing */
1099 mtype_uref(struct ip_set *set, struct netlink_callback *cb, bool start)
1101 struct htype *h = set->data;
1106 t = rcu_dereference_bh_nfnl(h->table);
1107 atomic_inc(&t->uref);
1108 cb->args[IPSET_CB_PRIVATE] = (unsigned long)t;
1109 rcu_read_unlock_bh();
1110 } else if (cb->args[IPSET_CB_PRIVATE]) {
1111 t = (struct htable *)cb->args[IPSET_CB_PRIVATE];
1112 if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1113 /* Resizing didn't destroy the hash table */
1114 pr_debug("Table destroy by dump: %p\n", t);
1115 mtype_ahash_destroy(set, t, false);
1117 cb->args[IPSET_CB_PRIVATE] = 0;
1121 /* Reply a LIST/SAVE request: dump the elements of the specified set */
1123 mtype_list(const struct ip_set *set,
1124 struct sk_buff *skb, struct netlink_callback *cb)
1126 const struct htable *t;
1127 struct nlattr *atd, *nested;
1128 const struct hbucket *n;
1129 const struct mtype_elem *e;
1130 u32 first = cb->args[IPSET_CB_ARG0];
1131 /* We assume that one hash bucket fills into one page */
1135 atd = ipset_nest_start(skb, IPSET_ATTR_ADT);
1139 pr_debug("list hash set %s\n", set->name);
1140 t = (const struct htable *)cb->args[IPSET_CB_PRIVATE];
1141 /* Expire may replace a hbucket with another one */
1143 for (; cb->args[IPSET_CB_ARG0] < jhash_size(t->htable_bits);
1144 cb->args[IPSET_CB_ARG0]++) {
1145 incomplete = skb_tail_pointer(skb);
1146 n = rcu_dereference(hbucket(t, cb->args[IPSET_CB_ARG0]));
1147 pr_debug("cb->arg bucket: %lu, t %p n %p\n",
1148 cb->args[IPSET_CB_ARG0], t, n);
1151 for (i = 0; i < n->pos; i++) {
1152 if (!test_bit(i, n->used))
1154 e = ahash_data(n, i, set->dsize);
1155 if (SET_WITH_TIMEOUT(set) &&
1156 ip_set_timeout_expired(ext_timeout(e, set)))
1158 pr_debug("list hash %lu hbucket %p i %u, data %p\n",
1159 cb->args[IPSET_CB_ARG0], n, i, e);
1160 nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
1162 if (cb->args[IPSET_CB_ARG0] == first) {
1163 nla_nest_cancel(skb, atd);
1167 goto nla_put_failure;
1169 if (mtype_data_list(skb, e))
1170 goto nla_put_failure;
1171 if (ip_set_put_extensions(skb, set, e, true))
1172 goto nla_put_failure;
1173 ipset_nest_end(skb, nested);
1176 ipset_nest_end(skb, atd);
1177 /* Set listing finished */
1178 cb->args[IPSET_CB_ARG0] = 0;
1183 nlmsg_trim(skb, incomplete);
1184 if (unlikely(first == cb->args[IPSET_CB_ARG0])) {
1185 pr_warn("Can't list set %s: one bucket does not fit into a message. Please report it!\n",
1187 cb->args[IPSET_CB_ARG0] = 0;
1190 ipset_nest_end(skb, atd);
1198 IPSET_TOKEN(MTYPE, _kadt)(struct ip_set *set, const struct sk_buff *skb,
1199 const struct xt_action_param *par,
1200 enum ipset_adt adt, struct ip_set_adt_opt *opt);
1203 IPSET_TOKEN(MTYPE, _uadt)(struct ip_set *set, struct nlattr *tb[],
1204 enum ipset_adt adt, u32 *lineno, u32 flags,
1207 static const struct ip_set_type_variant mtype_variant = {
1211 [IPSET_ADD] = mtype_add,
1212 [IPSET_DEL] = mtype_del,
1213 [IPSET_TEST] = mtype_test,
1215 .destroy = mtype_destroy,
1216 .flush = mtype_flush,
1220 .resize = mtype_resize,
1221 .same_set = mtype_same_set,
1224 #ifdef IP_SET_EMIT_CREATE
1226 IPSET_TOKEN(HTYPE, _create)(struct net *net, struct ip_set *set,
1227 struct nlattr *tb[], u32 flags)
1229 u32 hashsize = IPSET_DEFAULT_HASHSIZE, maxelem = IPSET_DEFAULT_MAXELEM;
1230 #ifdef IP_SET_HASH_WITH_MARKMASK
1234 #ifdef IP_SET_HASH_WITH_NETMASK
1241 #ifndef IP_SET_PROTO_UNDEF
1242 if (!(set->family == NFPROTO_IPV4 || set->family == NFPROTO_IPV6))
1243 return -IPSET_ERR_INVALID_FAMILY;
1246 #ifdef IP_SET_HASH_WITH_MARKMASK
1247 markmask = 0xffffffff;
1249 #ifdef IP_SET_HASH_WITH_NETMASK
1250 netmask = set->family == NFPROTO_IPV4 ? 32 : 128;
1251 pr_debug("Create set %s with family %s\n",
1252 set->name, set->family == NFPROTO_IPV4 ? "inet" : "inet6");
1255 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_HASHSIZE) ||
1256 !ip_set_optattr_netorder(tb, IPSET_ATTR_MAXELEM) ||
1257 !ip_set_optattr_netorder(tb, IPSET_ATTR_TIMEOUT) ||
1258 !ip_set_optattr_netorder(tb, IPSET_ATTR_CADT_FLAGS)))
1259 return -IPSET_ERR_PROTOCOL;
1260 #ifdef IP_SET_HASH_WITH_MARKMASK
1261 /* Separated condition in order to avoid directive in argument list */
1262 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_MARKMASK)))
1263 return -IPSET_ERR_PROTOCOL;
1266 if (tb[IPSET_ATTR_HASHSIZE]) {
1267 hashsize = ip_set_get_h32(tb[IPSET_ATTR_HASHSIZE]);
1268 if (hashsize < IPSET_MIMINAL_HASHSIZE)
1269 hashsize = IPSET_MIMINAL_HASHSIZE;
1272 if (tb[IPSET_ATTR_MAXELEM])
1273 maxelem = ip_set_get_h32(tb[IPSET_ATTR_MAXELEM]);
1275 #ifdef IP_SET_HASH_WITH_NETMASK
1276 if (tb[IPSET_ATTR_NETMASK]) {
1277 netmask = nla_get_u8(tb[IPSET_ATTR_NETMASK]);
1279 if ((set->family == NFPROTO_IPV4 && netmask > 32) ||
1280 (set->family == NFPROTO_IPV6 && netmask > 128) ||
1282 return -IPSET_ERR_INVALID_NETMASK;
1285 #ifdef IP_SET_HASH_WITH_MARKMASK
1286 if (tb[IPSET_ATTR_MARKMASK]) {
1287 markmask = ntohl(nla_get_be32(tb[IPSET_ATTR_MARKMASK]));
1290 return -IPSET_ERR_INVALID_MARKMASK;
1295 #ifdef IP_SET_HASH_WITH_NETS
1296 hsize += sizeof(struct net_prefixes) * NLEN(set->family);
1298 h = kzalloc(hsize, GFP_KERNEL);
1302 h->maxelem = maxelem;
1303 #ifdef IP_SET_HASH_WITH_NETMASK
1304 h->netmask = netmask;
1306 #ifdef IP_SET_HASH_WITH_MARKMASK
1307 h->markmask = markmask;
1309 get_random_bytes(&h->initval, sizeof(h->initval));
1310 set->timeout = IPSET_NO_TIMEOUT;
1312 hbits = htable_bits(hashsize);
1313 hsize = htable_size(hbits);
1318 t = ip_set_alloc(hsize);
1323 t->htable_bits = hbits;
1324 rcu_assign_pointer(h->table, t);
1327 #ifndef IP_SET_PROTO_UNDEF
1328 if (set->family == NFPROTO_IPV4) {
1330 set->variant = &IPSET_TOKEN(HTYPE, 4_variant);
1331 set->dsize = ip_set_elem_len(set, tb,
1332 sizeof(struct IPSET_TOKEN(HTYPE, 4_elem)),
1333 __alignof__(struct IPSET_TOKEN(HTYPE, 4_elem)));
1334 #ifndef IP_SET_PROTO_UNDEF
1336 set->variant = &IPSET_TOKEN(HTYPE, 6_variant);
1337 set->dsize = ip_set_elem_len(set, tb,
1338 sizeof(struct IPSET_TOKEN(HTYPE, 6_elem)),
1339 __alignof__(struct IPSET_TOKEN(HTYPE, 6_elem)));
1342 if (tb[IPSET_ATTR_TIMEOUT]) {
1343 set->timeout = ip_set_timeout_uget(tb[IPSET_ATTR_TIMEOUT]);
1344 #ifndef IP_SET_PROTO_UNDEF
1345 if (set->family == NFPROTO_IPV4)
1347 IPSET_TOKEN(HTYPE, 4_gc_init)(set,
1348 IPSET_TOKEN(HTYPE, 4_gc));
1349 #ifndef IP_SET_PROTO_UNDEF
1351 IPSET_TOKEN(HTYPE, 6_gc_init)(set,
1352 IPSET_TOKEN(HTYPE, 6_gc));
1355 pr_debug("create %s hashsize %u (%u) maxelem %u: %p(%p)\n",
1356 set->name, jhash_size(t->htable_bits),
1357 t->htable_bits, h->maxelem, set->data, t);
1361 #endif /* IP_SET_EMIT_CREATE */