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 } __attribute__ ((aligned));
78 /* The hash table: the table size stored here in order to make resizing easy */
80 atomic_t ref; /* References for resizing */
81 atomic_t uref; /* References for dumping */
82 u8 htable_bits; /* size of hash table == 2^htable_bits */
83 struct hbucket __rcu *bucket[0]; /* hashtable buckets */
86 #define hbucket(h, i) ((h)->bucket[i])
88 #ifndef IPSET_NET_COUNT
89 #define IPSET_NET_COUNT 1
92 /* Book-keeping of the prefixes added to the set */
94 u32 nets[IPSET_NET_COUNT]; /* number of elements for this cidr */
95 u8 cidr[IPSET_NET_COUNT]; /* the cidr value */
98 /* Compute the hash table size */
100 htable_size(u8 hbits)
104 /* We must fit both into u32 in jhash and size_t */
107 hsize = jhash_size(hbits);
108 if ((((size_t)-1) - sizeof(struct htable)) / sizeof(struct hbucket *)
112 return hsize * sizeof(struct hbucket *) + sizeof(struct htable);
115 /* Compute htable_bits from the user input parameter hashsize */
117 htable_bits(u32 hashsize)
119 /* Assume that hashsize == 2^htable_bits */
120 u8 bits = fls(hashsize - 1);
122 if (jhash_size(bits) != hashsize)
123 /* Round up to the first 2^n value */
124 bits = fls(hashsize);
129 #ifdef IP_SET_HASH_WITH_NETS
130 #if IPSET_NET_COUNT > 1
131 #define __CIDR(cidr, i) (cidr[i])
133 #define __CIDR(cidr, i) (cidr)
136 /* cidr + 1 is stored in net_prefixes to support /0 */
137 #define NCIDR_PUT(cidr) ((cidr) + 1)
138 #define NCIDR_GET(cidr) ((cidr) - 1)
140 #ifdef IP_SET_HASH_WITH_NETS_PACKED
141 /* When cidr is packed with nomatch, cidr - 1 is stored in the data entry */
142 #define DCIDR_PUT(cidr) ((cidr) - 1)
143 #define DCIDR_GET(cidr, i) (__CIDR(cidr, i) + 1)
145 #define DCIDR_PUT(cidr) (cidr)
146 #define DCIDR_GET(cidr, i) __CIDR(cidr, i)
149 #define INIT_CIDR(cidr, host_mask) \
150 DCIDR_PUT(((cidr) ? NCIDR_GET(cidr) : host_mask))
152 #define SET_HOST_MASK(family) (family == AF_INET ? 32 : 128)
154 #ifdef IP_SET_HASH_WITH_NET0
155 #define NLEN(family) (SET_HOST_MASK(family) + 1)
157 #define NLEN(family) SET_HOST_MASK(family)
161 #define NLEN(family) 0
162 #endif /* IP_SET_HASH_WITH_NETS */
164 #endif /* _IP_SET_HASH_GEN_H */
166 /* Family dependent templates */
169 #undef mtype_data_equal
170 #undef mtype_do_data_match
171 #undef mtype_data_set_flags
172 #undef mtype_data_reset_elem
173 #undef mtype_data_reset_flags
174 #undef mtype_data_netmask
175 #undef mtype_data_list
176 #undef mtype_data_next
179 #undef mtype_ahash_destroy
180 #undef mtype_ext_cleanup
181 #undef mtype_add_cidr
182 #undef mtype_del_cidr
183 #undef mtype_ahash_memsize
186 #undef mtype_same_set
193 #undef mtype_test_cidrs
203 #undef mtype_data_match
207 #define mtype_data_equal IPSET_TOKEN(MTYPE, _data_equal)
208 #ifdef IP_SET_HASH_WITH_NETS
209 #define mtype_do_data_match IPSET_TOKEN(MTYPE, _do_data_match)
211 #define mtype_do_data_match(d) 1
213 #define mtype_data_set_flags IPSET_TOKEN(MTYPE, _data_set_flags)
214 #define mtype_data_reset_elem IPSET_TOKEN(MTYPE, _data_reset_elem)
215 #define mtype_data_reset_flags IPSET_TOKEN(MTYPE, _data_reset_flags)
216 #define mtype_data_netmask IPSET_TOKEN(MTYPE, _data_netmask)
217 #define mtype_data_list IPSET_TOKEN(MTYPE, _data_list)
218 #define mtype_data_next IPSET_TOKEN(MTYPE, _data_next)
219 #define mtype_elem IPSET_TOKEN(MTYPE, _elem)
221 #define mtype_ahash_destroy IPSET_TOKEN(MTYPE, _ahash_destroy)
222 #define mtype_ext_cleanup IPSET_TOKEN(MTYPE, _ext_cleanup)
223 #define mtype_add_cidr IPSET_TOKEN(MTYPE, _add_cidr)
224 #define mtype_del_cidr IPSET_TOKEN(MTYPE, _del_cidr)
225 #define mtype_ahash_memsize IPSET_TOKEN(MTYPE, _ahash_memsize)
226 #define mtype_flush IPSET_TOKEN(MTYPE, _flush)
227 #define mtype_destroy IPSET_TOKEN(MTYPE, _destroy)
228 #define mtype_same_set IPSET_TOKEN(MTYPE, _same_set)
229 #define mtype_kadt IPSET_TOKEN(MTYPE, _kadt)
230 #define mtype_uadt IPSET_TOKEN(MTYPE, _uadt)
233 #define mtype_add IPSET_TOKEN(MTYPE, _add)
234 #define mtype_del IPSET_TOKEN(MTYPE, _del)
235 #define mtype_test_cidrs IPSET_TOKEN(MTYPE, _test_cidrs)
236 #define mtype_test IPSET_TOKEN(MTYPE, _test)
237 #define mtype_uref IPSET_TOKEN(MTYPE, _uref)
238 #define mtype_expire IPSET_TOKEN(MTYPE, _expire)
239 #define mtype_resize IPSET_TOKEN(MTYPE, _resize)
240 #define mtype_head IPSET_TOKEN(MTYPE, _head)
241 #define mtype_list IPSET_TOKEN(MTYPE, _list)
242 #define mtype_gc IPSET_TOKEN(MTYPE, _gc)
243 #define mtype_gc_init IPSET_TOKEN(MTYPE, _gc_init)
244 #define mtype_variant IPSET_TOKEN(MTYPE, _variant)
245 #define mtype_data_match IPSET_TOKEN(MTYPE, _data_match)
248 #error "MTYPE is not defined!"
252 #error "HOST_MASK is not defined!"
256 #define HKEY_DATALEN sizeof(struct mtype_elem)
259 #define HKEY(data, initval, htable_bits) \
260 (jhash2((u32 *)(data), HKEY_DATALEN / sizeof(u32), initval) \
261 & jhash_mask(htable_bits))
265 #error "HTYPE is not defined!"
269 /* The generic hash structure */
271 struct htable __rcu *table; /* the hash table */
272 u32 maxelem; /* max elements in the hash */
273 u32 elements; /* current element (vs timeout) */
274 u32 initval; /* random jhash init value */
275 #ifdef IP_SET_HASH_WITH_MARKMASK
276 u32 markmask; /* markmask value for mark mask to store */
278 struct timer_list gc; /* garbage collection when timeout enabled */
279 struct mtype_elem next; /* temporary storage for uadd */
280 #ifdef IP_SET_HASH_WITH_MULTI
281 u8 ahash_max; /* max elements in an array block */
283 #ifdef IP_SET_HASH_WITH_NETMASK
284 u8 netmask; /* netmask value for subnets to store */
286 #ifdef IP_SET_HASH_WITH_NETS
287 struct net_prefixes nets[0]; /* book-keeping of prefixes */
292 #ifdef IP_SET_HASH_WITH_NETS
293 /* Network cidr size book keeping when the hash stores different
294 * sized networks. cidr == real cidr + 1 to support /0.
297 mtype_add_cidr(struct htype *h, u8 cidr, u8 nets_length, u8 n)
301 /* Add in increasing prefix order, so larger cidr first */
302 for (i = 0, j = -1; i < nets_length && h->nets[i].cidr[n]; i++) {
305 } else if (h->nets[i].cidr[n] < cidr) {
307 } else if (h->nets[i].cidr[n] == cidr) {
308 h->nets[cidr - 1].nets[n]++;
314 h->nets[i].cidr[n] = h->nets[i - 1].cidr[n];
316 h->nets[i].cidr[n] = cidr;
317 h->nets[cidr - 1].nets[n] = 1;
321 mtype_del_cidr(struct htype *h, u8 cidr, u8 nets_length, u8 n)
323 u8 i, j, net_end = nets_length - 1;
325 for (i = 0; i < nets_length; i++) {
326 if (h->nets[i].cidr[n] != cidr)
328 h->nets[cidr - 1].nets[n]--;
329 if (h->nets[cidr - 1].nets[n] > 0)
331 for (j = i; j < net_end && h->nets[j].cidr[n]; j++)
332 h->nets[j].cidr[n] = h->nets[j + 1].cidr[n];
333 h->nets[j].cidr[n] = 0;
339 /* Calculate the actual memory size of the set data */
341 mtype_ahash_memsize(const struct htype *h, const struct htable *t,
342 u8 nets_length, size_t dsize)
346 size_t memsize = sizeof(*h) + sizeof(*t);
348 #ifdef IP_SET_HASH_WITH_NETS
349 memsize += sizeof(struct net_prefixes) * nets_length;
351 for (i = 0; i < jhash_size(t->htable_bits); i++) {
352 n = rcu_dereference_bh(hbucket(t, i));
355 memsize += sizeof(struct hbucket) + n->size * dsize;
361 /* Get the ith element from the array block n */
362 #define ahash_data(n, i, dsize) \
363 ((struct mtype_elem *)((n)->value + ((i) * (dsize))))
366 mtype_ext_cleanup(struct ip_set *set, struct hbucket *n)
370 for (i = 0; i < n->pos; i++)
371 if (test_bit(i, n->used))
372 ip_set_ext_destroy(set, ahash_data(n, i, set->dsize));
375 /* Flush a hash type of set: destroy all elements */
377 mtype_flush(struct ip_set *set)
379 struct htype *h = set->data;
384 t = ipset_dereference_protected(h->table, set);
385 for (i = 0; i < jhash_size(t->htable_bits); i++) {
386 n = __ipset_dereference_protected(hbucket(t, i), 1);
389 if (set->extensions & IPSET_EXT_DESTROY)
390 mtype_ext_cleanup(set, n);
391 /* FIXME: use slab cache */
392 rcu_assign_pointer(hbucket(t, i), NULL);
395 #ifdef IP_SET_HASH_WITH_NETS
396 memset(h->nets, 0, sizeof(struct net_prefixes) * NLEN(set->family));
401 /* Destroy the hashtable part of the set */
403 mtype_ahash_destroy(struct ip_set *set, struct htable *t, bool ext_destroy)
408 for (i = 0; i < jhash_size(t->htable_bits); i++) {
409 n = __ipset_dereference_protected(hbucket(t, i), 1);
412 if (set->extensions & IPSET_EXT_DESTROY && ext_destroy)
413 mtype_ext_cleanup(set, n);
414 /* FIXME: use slab cache */
421 /* Destroy a hash type of set */
423 mtype_destroy(struct ip_set *set)
425 struct htype *h = set->data;
427 if (SET_WITH_TIMEOUT(set))
428 del_timer_sync(&h->gc);
430 mtype_ahash_destroy(set,
431 __ipset_dereference_protected(h->table, 1), true);
438 mtype_gc_init(struct ip_set *set, void (*gc)(unsigned long ul_set))
440 struct htype *h = set->data;
443 h->gc.data = (unsigned long)set;
445 h->gc.expires = jiffies + IPSET_GC_PERIOD(set->timeout) * HZ;
447 pr_debug("gc initialized, run in every %u\n",
448 IPSET_GC_PERIOD(set->timeout));
452 mtype_same_set(const struct ip_set *a, const struct ip_set *b)
454 const struct htype *x = a->data;
455 const struct htype *y = b->data;
457 /* Resizing changes htable_bits, so we ignore it */
458 return x->maxelem == y->maxelem &&
459 a->timeout == b->timeout &&
460 #ifdef IP_SET_HASH_WITH_NETMASK
461 x->netmask == y->netmask &&
463 #ifdef IP_SET_HASH_WITH_MARKMASK
464 x->markmask == y->markmask &&
466 a->extensions == b->extensions;
469 /* Delete expired elements from the hashtable */
471 mtype_expire(struct ip_set *set, struct htype *h, u8 nets_length, size_t dsize)
475 struct mtype_elem *data;
477 #ifdef IP_SET_HASH_WITH_NETS
481 t = ipset_dereference_protected(h->table, set);
482 for (i = 0; i < jhash_size(t->htable_bits); i++) {
483 n = __ipset_dereference_protected(hbucket(t, i), 1);
486 for (j = 0, d = 0; j < n->pos; j++) {
487 if (!test_bit(j, n->used)) {
491 data = ahash_data(n, j, dsize);
492 if (ip_set_timeout_expired(ext_timeout(data, set))) {
493 pr_debug("expired %u/%u\n", i, j);
494 clear_bit(j, n->used);
495 smp_mb__after_atomic();
496 #ifdef IP_SET_HASH_WITH_NETS
497 for (k = 0; k < IPSET_NET_COUNT; k++)
499 NCIDR_PUT(DCIDR_GET(data->cidr,
503 ip_set_ext_destroy(set, data);
508 if (d >= AHASH_INIT_SIZE) {
509 struct hbucket *tmp = kzalloc(sizeof(*tmp) +
510 (n->size - AHASH_INIT_SIZE) * dsize,
513 /* Still try to delete expired elements */
515 tmp->size = n->size - AHASH_INIT_SIZE;
516 for (j = 0, d = 0; j < n->pos; j++) {
517 if (!test_bit(j, n->used))
519 data = ahash_data(n, j, dsize);
520 memcpy(tmp->value + d * dsize, data, dsize);
521 set_bit(j, tmp->used);
525 rcu_assign_pointer(hbucket(t, i), tmp);
532 mtype_gc(unsigned long ul_set)
534 struct ip_set *set = (struct ip_set *)ul_set;
535 struct htype *h = set->data;
537 pr_debug("called\n");
538 spin_lock_bh(&set->lock);
539 mtype_expire(set, h, NLEN(set->family), set->dsize);
540 spin_unlock_bh(&set->lock);
542 h->gc.expires = jiffies + IPSET_GC_PERIOD(set->timeout) * HZ;
546 /* Resize a hash: create a new hash table with doubling the hashsize
547 * and inserting the elements to it. Repeat until we succeed or
548 * fail due to memory pressures.
551 mtype_resize(struct ip_set *set, bool retried)
553 struct htype *h = set->data;
554 struct htable *t, *orig;
556 size_t dsize = set->dsize;
557 #ifdef IP_SET_HASH_WITH_NETS
559 struct mtype_elem *tmp;
561 struct mtype_elem *data;
562 struct mtype_elem *d;
563 struct hbucket *n, *m;
567 #ifdef IP_SET_HASH_WITH_NETS
568 tmp = kmalloc(dsize, GFP_KERNEL);
573 orig = rcu_dereference_bh_nfnl(h->table);
574 htable_bits = orig->htable_bits;
575 rcu_read_unlock_bh();
581 /* In case we have plenty of memory :-) */
582 pr_warn("Cannot increase the hashsize of set %s further\n",
584 ret = -IPSET_ERR_HASH_FULL;
587 t = ip_set_alloc(htable_size(htable_bits));
592 t->htable_bits = htable_bits;
594 spin_lock_bh(&set->lock);
595 orig = __ipset_dereference_protected(h->table, 1);
596 /* There can't be another parallel resizing, but dumping is possible */
597 atomic_set(&orig->ref, 1);
598 atomic_inc(&orig->uref);
599 pr_debug("attempt to resize set %s from %u to %u, t %p\n",
600 set->name, orig->htable_bits, htable_bits, orig);
601 for (i = 0; i < jhash_size(orig->htable_bits); i++) {
602 n = __ipset_dereference_protected(hbucket(orig, i), 1);
605 for (j = 0; j < n->pos; j++) {
606 if (!test_bit(j, n->used))
608 data = ahash_data(n, j, dsize);
609 #ifdef IP_SET_HASH_WITH_NETS
610 /* We have readers running parallel with us,
611 * so the live data cannot be modified.
614 memcpy(tmp, data, dsize);
616 mtype_data_reset_flags(data, &flags);
618 key = HKEY(data, h->initval, htable_bits);
619 m = __ipset_dereference_protected(hbucket(t, key), 1);
621 m = kzalloc(sizeof(*m) +
622 AHASH_INIT_SIZE * dsize,
628 m->size = AHASH_INIT_SIZE;
629 RCU_INIT_POINTER(hbucket(t, key), m);
630 } else if (m->pos >= m->size) {
633 if (m->size >= AHASH_MAX(h)) {
636 ht = kzalloc(sizeof(*ht) +
637 (m->size + AHASH_INIT_SIZE)
645 memcpy(ht, m, sizeof(struct hbucket) +
647 ht->size = m->size + AHASH_INIT_SIZE;
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);
662 spin_unlock_bh(&set->lock);
664 /* Give time to other readers of the set */
665 synchronize_rcu_bh();
667 pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
668 orig->htable_bits, orig, t->htable_bits, t);
669 /* If there's nobody else dumping the table, destroy it */
670 if (atomic_dec_and_test(&orig->uref)) {
671 pr_debug("Table destroy by resize %p\n", orig);
672 mtype_ahash_destroy(set, orig, false);
676 #ifdef IP_SET_HASH_WITH_NETS
682 atomic_set(&orig->ref, 0);
683 atomic_dec(&orig->uref);
684 spin_unlock_bh(&set->lock);
685 mtype_ahash_destroy(set, t, false);
691 /* Add an element to a hash and update the internal counters when succeeded,
692 * otherwise report the proper error code.
695 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
696 struct ip_set_ext *mext, u32 flags)
698 struct htype *h = set->data;
700 const struct mtype_elem *d = value;
701 struct mtype_elem *data;
702 struct hbucket *n, *old = ERR_PTR(-ENOENT);
704 bool flag_exist = flags & IPSET_FLAG_EXIST;
705 bool deleted = false, forceadd = false, reuse = false;
708 if (h->elements >= h->maxelem) {
709 if (SET_WITH_TIMEOUT(set))
710 /* FIXME: when set is full, we slow down here */
711 mtype_expire(set, h, NLEN(set->family), set->dsize);
712 if (h->elements >= h->maxelem && SET_WITH_FORCEADD(set))
716 t = ipset_dereference_protected(h->table, set);
717 key = HKEY(value, h->initval, t->htable_bits);
718 n = __ipset_dereference_protected(hbucket(t, key), 1);
722 pr_warn("Set %s is full, maxelem %u reached\n",
723 set->name, h->maxelem);
724 return -IPSET_ERR_HASH_FULL;
725 } else if (h->elements >= h->maxelem) {
729 n = kzalloc(sizeof(*n) + AHASH_INIT_SIZE * set->dsize,
733 n->size = AHASH_INIT_SIZE;
736 for (i = 0; i < n->pos; i++) {
737 if (!test_bit(i, n->used)) {
738 /* Reuse first deleted entry */
740 deleted = reuse = true;
745 data = ahash_data(n, i, set->dsize);
746 if (mtype_data_equal(data, d, &multi)) {
748 (SET_WITH_TIMEOUT(set) &&
749 ip_set_timeout_expired(ext_timeout(data, set)))) {
750 /* Just the extensions could be overwritten */
752 goto overwrite_extensions;
754 return -IPSET_ERR_EXIST;
756 /* Reuse first timed out entry */
757 if (SET_WITH_TIMEOUT(set) &&
758 ip_set_timeout_expired(ext_timeout(data, set)) &&
764 if (reuse || forceadd) {
765 data = ahash_data(n, j, set->dsize);
767 #ifdef IP_SET_HASH_WITH_NETS
768 for (i = 0; i < IPSET_NET_COUNT; i++)
770 NCIDR_PUT(DCIDR_GET(data->cidr, i)),
771 NLEN(set->family), i);
773 ip_set_ext_destroy(set, data);
778 if (h->elements >= h->maxelem)
780 /* Create a new slot */
781 if (n->pos >= n->size) {
782 TUNE_AHASH_MAX(h, multi);
783 if (n->size >= AHASH_MAX(h)) {
784 /* Trigger rehashing */
785 mtype_data_next(&h->next, d);
789 n = kzalloc(sizeof(*n) +
790 (old->size + AHASH_INIT_SIZE) * set->dsize,
794 memcpy(n, old, sizeof(struct hbucket) +
795 old->size * set->dsize);
796 n->size = old->size + AHASH_INIT_SIZE;
801 data = ahash_data(n, j, set->dsize);
804 #ifdef IP_SET_HASH_WITH_NETS
805 for (i = 0; i < IPSET_NET_COUNT; i++)
806 mtype_add_cidr(h, NCIDR_PUT(DCIDR_GET(d->cidr, i)),
807 NLEN(set->family), i);
809 memcpy(data, d, sizeof(struct mtype_elem));
810 overwrite_extensions:
811 #ifdef IP_SET_HASH_WITH_NETS
812 mtype_data_set_flags(data, flags);
814 if (SET_WITH_COUNTER(set))
815 ip_set_init_counter(ext_counter(data, set), ext);
816 if (SET_WITH_COMMENT(set))
817 ip_set_init_comment(ext_comment(data, set), ext);
818 if (SET_WITH_SKBINFO(set))
819 ip_set_init_skbinfo(ext_skbinfo(data, set), ext);
820 /* Must come last for the case when timed out entry is reused */
821 if (SET_WITH_TIMEOUT(set))
822 ip_set_timeout_set(ext_timeout(data, set), ext->timeout);
823 smp_mb__before_atomic();
825 if (old != ERR_PTR(-ENOENT)) {
826 rcu_assign_pointer(hbucket(t, key), n);
834 pr_warn("Set %s is full, maxelem %u reached\n",
835 set->name, h->maxelem);
836 return -IPSET_ERR_HASH_FULL;
839 /* Delete an element from the hash and free up space if possible.
842 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
843 struct ip_set_ext *mext, u32 flags)
845 struct htype *h = set->data;
847 const struct mtype_elem *d = value;
848 struct mtype_elem *data;
850 int i, j, k, ret = -IPSET_ERR_EXIST;
852 size_t dsize = set->dsize;
854 t = ipset_dereference_protected(h->table, set);
855 key = HKEY(value, h->initval, t->htable_bits);
856 n = __ipset_dereference_protected(hbucket(t, key), 1);
859 for (i = 0, k = 0; i < n->pos; i++) {
860 if (!test_bit(i, n->used)) {
864 data = ahash_data(n, i, dsize);
865 if (!mtype_data_equal(data, d, &multi))
867 if (SET_WITH_TIMEOUT(set) &&
868 ip_set_timeout_expired(ext_timeout(data, set)))
872 clear_bit(i, n->used);
873 smp_mb__after_atomic();
877 #ifdef IP_SET_HASH_WITH_NETS
878 for (j = 0; j < IPSET_NET_COUNT; j++)
879 mtype_del_cidr(h, NCIDR_PUT(DCIDR_GET(d->cidr, j)),
880 NLEN(set->family), j);
882 ip_set_ext_destroy(set, data);
884 for (; i < n->pos; i++) {
885 if (!test_bit(i, n->used))
888 if (n->pos == 0 && k == 0) {
889 rcu_assign_pointer(hbucket(t, key), NULL);
891 } else if (k >= AHASH_INIT_SIZE) {
892 struct hbucket *tmp = kzalloc(sizeof(*tmp) +
893 (n->size - AHASH_INIT_SIZE) * dsize,
897 tmp->size = n->size - AHASH_INIT_SIZE;
898 for (j = 0, k = 0; j < n->pos; j++) {
899 if (!test_bit(j, n->used))
901 data = ahash_data(n, j, dsize);
902 memcpy(tmp->value + k * dsize, data, dsize);
903 set_bit(j, tmp->used);
907 rcu_assign_pointer(hbucket(t, key), tmp);
918 mtype_data_match(struct mtype_elem *data, const struct ip_set_ext *ext,
919 struct ip_set_ext *mext, struct ip_set *set, u32 flags)
921 if (SET_WITH_COUNTER(set))
922 ip_set_update_counter(ext_counter(data, set),
924 if (SET_WITH_SKBINFO(set))
925 ip_set_get_skbinfo(ext_skbinfo(data, set),
927 return mtype_do_data_match(data);
930 #ifdef IP_SET_HASH_WITH_NETS
931 /* Special test function which takes into account the different network
932 * sizes added to the set
935 mtype_test_cidrs(struct ip_set *set, struct mtype_elem *d,
936 const struct ip_set_ext *ext,
937 struct ip_set_ext *mext, u32 flags)
939 struct htype *h = set->data;
940 struct htable *t = rcu_dereference_bh(h->table);
942 struct mtype_elem *data;
943 #if IPSET_NET_COUNT == 2
944 struct mtype_elem orig = *d;
950 u8 nets_length = NLEN(set->family);
952 pr_debug("test by nets\n");
953 for (; j < nets_length && h->nets[j].cidr[0] && !multi; j++) {
954 #if IPSET_NET_COUNT == 2
955 mtype_data_reset_elem(d, &orig);
956 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]), false);
957 for (k = 0; k < nets_length && h->nets[k].cidr[1] && !multi;
959 mtype_data_netmask(d, NCIDR_GET(h->nets[k].cidr[1]),
962 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]));
964 key = HKEY(d, h->initval, t->htable_bits);
965 n = rcu_dereference_bh(hbucket(t, key));
968 for (i = 0; i < n->pos; i++) {
969 if (!test_bit(i, n->used))
971 data = ahash_data(n, i, set->dsize);
972 if (!mtype_data_equal(data, d, &multi))
974 if (SET_WITH_TIMEOUT(set)) {
975 if (!ip_set_timeout_expired(
976 ext_timeout(data, set)))
977 return mtype_data_match(data, ext,
980 #ifdef IP_SET_HASH_WITH_MULTI
984 return mtype_data_match(data, ext,
987 #if IPSET_NET_COUNT == 2
995 /* Test whether the element is added to the set */
997 mtype_test(struct ip_set *set, void *value, const struct ip_set_ext *ext,
998 struct ip_set_ext *mext, u32 flags)
1000 struct htype *h = set->data;
1002 struct mtype_elem *d = value;
1004 struct mtype_elem *data;
1008 t = rcu_dereference_bh(h->table);
1009 #ifdef IP_SET_HASH_WITH_NETS
1010 /* If we test an IP address and not a network address,
1011 * try all possible network sizes
1013 for (i = 0; i < IPSET_NET_COUNT; i++)
1014 if (DCIDR_GET(d->cidr, i) != SET_HOST_MASK(set->family))
1016 if (i == IPSET_NET_COUNT) {
1017 ret = mtype_test_cidrs(set, d, ext, mext, flags);
1022 key = HKEY(d, h->initval, t->htable_bits);
1023 n = rcu_dereference_bh(hbucket(t, key));
1028 for (i = 0; i < n->pos; i++) {
1029 if (!test_bit(i, n->used))
1031 data = ahash_data(n, i, set->dsize);
1032 if (mtype_data_equal(data, d, &multi) &&
1033 !(SET_WITH_TIMEOUT(set) &&
1034 ip_set_timeout_expired(ext_timeout(data, set)))) {
1035 ret = mtype_data_match(data, ext, mext, set, flags);
1043 /* Reply a HEADER request: fill out the header part of the set */
1045 mtype_head(struct ip_set *set, struct sk_buff *skb)
1047 const struct htype *h = set->data;
1048 const struct htable *t;
1049 struct nlattr *nested;
1054 t = rcu_dereference_bh_nfnl(h->table);
1055 memsize = mtype_ahash_memsize(h, t, NLEN(set->family), set->dsize);
1056 htable_bits = t->htable_bits;
1057 rcu_read_unlock_bh();
1059 nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
1061 goto nla_put_failure;
1062 if (nla_put_net32(skb, IPSET_ATTR_HASHSIZE,
1063 htonl(jhash_size(htable_bits))) ||
1064 nla_put_net32(skb, IPSET_ATTR_MAXELEM, htonl(h->maxelem)))
1065 goto nla_put_failure;
1066 #ifdef IP_SET_HASH_WITH_NETMASK
1067 if (h->netmask != HOST_MASK &&
1068 nla_put_u8(skb, IPSET_ATTR_NETMASK, h->netmask))
1069 goto nla_put_failure;
1071 #ifdef IP_SET_HASH_WITH_MARKMASK
1072 if (nla_put_u32(skb, IPSET_ATTR_MARKMASK, h->markmask))
1073 goto nla_put_failure;
1075 if (nla_put_net32(skb, IPSET_ATTR_REFERENCES, htonl(set->ref - 1)) ||
1076 nla_put_net32(skb, IPSET_ATTR_MEMSIZE, htonl(memsize)))
1077 goto nla_put_failure;
1078 if (unlikely(ip_set_put_flags(skb, set)))
1079 goto nla_put_failure;
1080 ipset_nest_end(skb, nested);
1087 /* Make possible to run dumping parallel with resizing */
1089 mtype_uref(struct ip_set *set, struct netlink_callback *cb, bool start)
1091 struct htype *h = set->data;
1096 t = rcu_dereference_bh_nfnl(h->table);
1097 atomic_inc(&t->uref);
1098 cb->args[IPSET_CB_PRIVATE] = (unsigned long)t;
1099 rcu_read_unlock_bh();
1100 } else if (cb->args[IPSET_CB_PRIVATE]) {
1101 t = (struct htable *)cb->args[IPSET_CB_PRIVATE];
1102 if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1103 /* Resizing didn't destroy the hash table */
1104 pr_debug("Table destroy by dump: %p\n", t);
1105 mtype_ahash_destroy(set, t, false);
1107 cb->args[IPSET_CB_PRIVATE] = 0;
1111 /* Reply a LIST/SAVE request: dump the elements of the specified set */
1113 mtype_list(const struct ip_set *set,
1114 struct sk_buff *skb, struct netlink_callback *cb)
1116 const struct htable *t;
1117 struct nlattr *atd, *nested;
1118 const struct hbucket *n;
1119 const struct mtype_elem *e;
1120 u32 first = cb->args[IPSET_CB_ARG0];
1121 /* We assume that one hash bucket fills into one page */
1125 atd = ipset_nest_start(skb, IPSET_ATTR_ADT);
1129 pr_debug("list hash set %s\n", set->name);
1130 t = (const struct htable *)cb->args[IPSET_CB_PRIVATE];
1131 /* Expire may replace a hbucket with another one */
1133 for (; cb->args[IPSET_CB_ARG0] < jhash_size(t->htable_bits);
1134 cb->args[IPSET_CB_ARG0]++) {
1135 incomplete = skb_tail_pointer(skb);
1136 n = rcu_dereference(hbucket(t, cb->args[IPSET_CB_ARG0]));
1137 pr_debug("cb->arg bucket: %lu, t %p n %p\n",
1138 cb->args[IPSET_CB_ARG0], t, n);
1141 for (i = 0; i < n->pos; i++) {
1142 if (!test_bit(i, n->used))
1144 e = ahash_data(n, i, set->dsize);
1145 if (SET_WITH_TIMEOUT(set) &&
1146 ip_set_timeout_expired(ext_timeout(e, set)))
1148 pr_debug("list hash %lu hbucket %p i %u, data %p\n",
1149 cb->args[IPSET_CB_ARG0], n, i, e);
1150 nested = ipset_nest_start(skb, IPSET_ATTR_DATA);
1152 if (cb->args[IPSET_CB_ARG0] == first) {
1153 nla_nest_cancel(skb, atd);
1157 goto nla_put_failure;
1159 if (mtype_data_list(skb, e))
1160 goto nla_put_failure;
1161 if (ip_set_put_extensions(skb, set, e, true))
1162 goto nla_put_failure;
1163 ipset_nest_end(skb, nested);
1166 ipset_nest_end(skb, atd);
1167 /* Set listing finished */
1168 cb->args[IPSET_CB_ARG0] = 0;
1173 nlmsg_trim(skb, incomplete);
1174 if (unlikely(first == cb->args[IPSET_CB_ARG0])) {
1175 pr_warn("Can't list set %s: one bucket does not fit into a message. Please report it!\n",
1177 cb->args[IPSET_CB_ARG0] = 0;
1180 ipset_nest_end(skb, atd);
1188 IPSET_TOKEN(MTYPE, _kadt)(struct ip_set *set, const struct sk_buff *skb,
1189 const struct xt_action_param *par,
1190 enum ipset_adt adt, struct ip_set_adt_opt *opt);
1193 IPSET_TOKEN(MTYPE, _uadt)(struct ip_set *set, struct nlattr *tb[],
1194 enum ipset_adt adt, u32 *lineno, u32 flags,
1197 static const struct ip_set_type_variant mtype_variant = {
1201 [IPSET_ADD] = mtype_add,
1202 [IPSET_DEL] = mtype_del,
1203 [IPSET_TEST] = mtype_test,
1205 .destroy = mtype_destroy,
1206 .flush = mtype_flush,
1210 .resize = mtype_resize,
1211 .same_set = mtype_same_set,
1214 #ifdef IP_SET_EMIT_CREATE
1216 IPSET_TOKEN(HTYPE, _create)(struct net *net, struct ip_set *set,
1217 struct nlattr *tb[], u32 flags)
1219 u32 hashsize = IPSET_DEFAULT_HASHSIZE, maxelem = IPSET_DEFAULT_MAXELEM;
1220 #ifdef IP_SET_HASH_WITH_MARKMASK
1224 #ifdef IP_SET_HASH_WITH_NETMASK
1231 #ifndef IP_SET_PROTO_UNDEF
1232 if (!(set->family == NFPROTO_IPV4 || set->family == NFPROTO_IPV6))
1233 return -IPSET_ERR_INVALID_FAMILY;
1236 #ifdef IP_SET_HASH_WITH_MARKMASK
1237 markmask = 0xffffffff;
1239 #ifdef IP_SET_HASH_WITH_NETMASK
1240 netmask = set->family == NFPROTO_IPV4 ? 32 : 128;
1241 pr_debug("Create set %s with family %s\n",
1242 set->name, set->family == NFPROTO_IPV4 ? "inet" : "inet6");
1245 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_HASHSIZE) ||
1246 !ip_set_optattr_netorder(tb, IPSET_ATTR_MAXELEM) ||
1247 !ip_set_optattr_netorder(tb, IPSET_ATTR_TIMEOUT) ||
1248 !ip_set_optattr_netorder(tb, IPSET_ATTR_CADT_FLAGS)))
1249 return -IPSET_ERR_PROTOCOL;
1250 #ifdef IP_SET_HASH_WITH_MARKMASK
1251 /* Separated condition in order to avoid directive in argument list */
1252 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_MARKMASK)))
1253 return -IPSET_ERR_PROTOCOL;
1256 if (tb[IPSET_ATTR_HASHSIZE]) {
1257 hashsize = ip_set_get_h32(tb[IPSET_ATTR_HASHSIZE]);
1258 if (hashsize < IPSET_MIMINAL_HASHSIZE)
1259 hashsize = IPSET_MIMINAL_HASHSIZE;
1262 if (tb[IPSET_ATTR_MAXELEM])
1263 maxelem = ip_set_get_h32(tb[IPSET_ATTR_MAXELEM]);
1265 #ifdef IP_SET_HASH_WITH_NETMASK
1266 if (tb[IPSET_ATTR_NETMASK]) {
1267 netmask = nla_get_u8(tb[IPSET_ATTR_NETMASK]);
1269 if ((set->family == NFPROTO_IPV4 && netmask > 32) ||
1270 (set->family == NFPROTO_IPV6 && netmask > 128) ||
1272 return -IPSET_ERR_INVALID_NETMASK;
1275 #ifdef IP_SET_HASH_WITH_MARKMASK
1276 if (tb[IPSET_ATTR_MARKMASK]) {
1277 markmask = ntohl(nla_get_be32(tb[IPSET_ATTR_MARKMASK]));
1280 return -IPSET_ERR_INVALID_MARKMASK;
1285 #ifdef IP_SET_HASH_WITH_NETS
1286 hsize += sizeof(struct net_prefixes) * NLEN(set->family);
1288 h = kzalloc(hsize, GFP_KERNEL);
1292 h->maxelem = maxelem;
1293 #ifdef IP_SET_HASH_WITH_NETMASK
1294 h->netmask = netmask;
1296 #ifdef IP_SET_HASH_WITH_MARKMASK
1297 h->markmask = markmask;
1299 get_random_bytes(&h->initval, sizeof(h->initval));
1300 set->timeout = IPSET_NO_TIMEOUT;
1302 hbits = htable_bits(hashsize);
1303 hsize = htable_size(hbits);
1308 t = ip_set_alloc(hsize);
1313 t->htable_bits = hbits;
1314 rcu_assign_pointer(h->table, t);
1317 #ifndef IP_SET_PROTO_UNDEF
1318 if (set->family == NFPROTO_IPV4) {
1320 set->variant = &IPSET_TOKEN(HTYPE, 4_variant);
1321 set->dsize = ip_set_elem_len(set, tb,
1322 sizeof(struct IPSET_TOKEN(HTYPE, 4_elem)));
1323 #ifndef IP_SET_PROTO_UNDEF
1325 set->variant = &IPSET_TOKEN(HTYPE, 6_variant);
1326 set->dsize = ip_set_elem_len(set, tb,
1327 sizeof(struct IPSET_TOKEN(HTYPE, 6_elem)));
1330 if (tb[IPSET_ATTR_TIMEOUT]) {
1331 set->timeout = ip_set_timeout_uget(tb[IPSET_ATTR_TIMEOUT]);
1332 #ifndef IP_SET_PROTO_UNDEF
1333 if (set->family == NFPROTO_IPV4)
1335 IPSET_TOKEN(HTYPE, 4_gc_init)(set,
1336 IPSET_TOKEN(HTYPE, 4_gc));
1337 #ifndef IP_SET_PROTO_UNDEF
1339 IPSET_TOKEN(HTYPE, 6_gc_init)(set,
1340 IPSET_TOKEN(HTYPE, 6_gc));
1343 pr_debug("create %s hashsize %u (%u) maxelem %u: %p(%p)\n",
1344 set->name, jhash_size(t->htable_bits),
1345 t->htable_bits, h->maxelem, set->data, t);
1349 #endif /* IP_SET_EMIT_CREATE */