1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /* Copyright (C) 2013 Jozsef Kadlecsik <kadlec@netfilter.org> */
4 #ifndef _IP_SET_HASH_GEN_H
5 #define _IP_SET_HASH_GEN_H
7 #include <linux/rcupdate.h>
8 #include <linux/jhash.h>
9 #include <linux/types.h>
10 #include <linux/netfilter/ipset/ip_set_timeout.h>
12 #define __ipset_dereference_protected(p, c) rcu_dereference_protected(p, c)
13 #define ipset_dereference_protected(p, set) \
14 __ipset_dereference_protected(p, lockdep_is_held(&(set)->lock))
16 #define rcu_dereference_bh_nfnl(p) rcu_dereference_bh_check(p, 1)
18 /* Hashing which uses arrays to resolve clashing. The hash table is resized
19 * (doubled) when searching becomes too long.
20 * Internally jhash is used with the assumption that the size of the
21 * stored data is a multiple of sizeof(u32).
23 * Readers and resizing
25 * Resizing can be triggered by userspace command only, and those
26 * are serialized by the nfnl mutex. During resizing the set is
27 * read-locked, so the only possible concurrent operations are
28 * the kernel side readers. Those must be protected by proper RCU locking.
31 /* Number of elements to store in an initial array block */
32 #define AHASH_INIT_SIZE 4
33 /* Max number of elements to store in an array block */
34 #define AHASH_MAX_SIZE (3 * AHASH_INIT_SIZE)
35 /* Max muber of elements in the array block when tuned */
36 #define AHASH_MAX_TUNED 64
38 /* Max number of elements can be tuned */
39 #ifdef IP_SET_HASH_WITH_MULTI
40 #define AHASH_MAX(h) ((h)->ahash_max)
43 tune_ahash_max(u8 curr, u32 multi)
50 n = curr + AHASH_INIT_SIZE;
51 /* Currently, at listing one hash bucket must fit into a message.
52 * Therefore we have a hard limit here.
54 return n > curr && n <= AHASH_MAX_TUNED ? n : curr;
57 #define TUNE_AHASH_MAX(h, multi) \
58 ((h)->ahash_max = tune_ahash_max((h)->ahash_max, multi))
60 #define AHASH_MAX(h) AHASH_MAX_SIZE
61 #define TUNE_AHASH_MAX(h, multi)
66 struct rcu_head rcu; /* for call_rcu */
67 /* Which positions are used in the array */
68 DECLARE_BITMAP(used, AHASH_MAX_TUNED);
69 u8 size; /* size of the array */
70 u8 pos; /* position of the first free entry */
71 unsigned char value[0] /* the array of the values */
72 __aligned(__alignof__(u64));
75 /* The hash table: the table size stored here in order to make resizing easy */
77 atomic_t ref; /* References for resizing */
78 atomic_t uref; /* References for dumping */
79 u8 htable_bits; /* size of hash table == 2^htable_bits */
80 struct hbucket __rcu *bucket[0]; /* hashtable buckets */
83 #define hbucket(h, i) ((h)->bucket[i])
84 #define ext_size(n, dsize) \
85 (sizeof(struct hbucket) + (n) * (dsize))
87 #ifndef IPSET_NET_COUNT
88 #define IPSET_NET_COUNT 1
91 /* Book-keeping of the prefixes added to the set */
93 u32 nets[IPSET_NET_COUNT]; /* number of elements for this cidr */
94 u8 cidr[IPSET_NET_COUNT]; /* the cidr value */
97 /* Compute the hash table size */
103 /* We must fit both into u32 in jhash and size_t */
106 hsize = jhash_size(hbits);
107 if ((((size_t)-1) - sizeof(struct htable)) / sizeof(struct hbucket *)
111 return hsize * sizeof(struct hbucket *) + sizeof(struct htable);
114 /* Compute htable_bits from the user input parameter hashsize */
116 htable_bits(u32 hashsize)
118 /* Assume that hashsize == 2^htable_bits */
119 u8 bits = fls(hashsize - 1);
121 if (jhash_size(bits) != hashsize)
122 /* Round up to the first 2^n value */
123 bits = fls(hashsize);
128 #ifdef IP_SET_HASH_WITH_NETS
129 #if IPSET_NET_COUNT > 1
130 #define __CIDR(cidr, i) (cidr[i])
132 #define __CIDR(cidr, i) (cidr)
135 /* cidr + 1 is stored in net_prefixes to support /0 */
136 #define NCIDR_PUT(cidr) ((cidr) + 1)
137 #define NCIDR_GET(cidr) ((cidr) - 1)
139 #ifdef IP_SET_HASH_WITH_NETS_PACKED
140 /* When cidr is packed with nomatch, cidr - 1 is stored in the data entry */
141 #define DCIDR_PUT(cidr) ((cidr) - 1)
142 #define DCIDR_GET(cidr, i) (__CIDR(cidr, i) + 1)
144 #define DCIDR_PUT(cidr) (cidr)
145 #define DCIDR_GET(cidr, i) __CIDR(cidr, i)
148 #define INIT_CIDR(cidr, host_mask) \
149 DCIDR_PUT(((cidr) ? NCIDR_GET(cidr) : host_mask))
151 #ifdef IP_SET_HASH_WITH_NET0
152 /* cidr from 0 to HOST_MASK value and c = cidr + 1 */
153 #define NLEN (HOST_MASK + 1)
154 #define CIDR_POS(c) ((c) - 1)
156 /* cidr from 1 to HOST_MASK value and c = cidr + 1 */
157 #define NLEN HOST_MASK
158 #define CIDR_POS(c) ((c) - 2)
163 #endif /* IP_SET_HASH_WITH_NETS */
165 #endif /* _IP_SET_HASH_GEN_H */
168 #error "MTYPE is not defined!"
172 #error "HTYPE is not defined!"
176 #error "HOST_MASK is not defined!"
179 /* Family dependent templates */
182 #undef mtype_data_equal
183 #undef mtype_do_data_match
184 #undef mtype_data_set_flags
185 #undef mtype_data_reset_elem
186 #undef mtype_data_reset_flags
187 #undef mtype_data_netmask
188 #undef mtype_data_list
189 #undef mtype_data_next
192 #undef mtype_ahash_destroy
193 #undef mtype_ext_cleanup
194 #undef mtype_add_cidr
195 #undef mtype_del_cidr
196 #undef mtype_ahash_memsize
199 #undef mtype_same_set
205 #undef mtype_test_cidrs
215 #undef mtype_data_match
220 #define mtype_data_equal IPSET_TOKEN(MTYPE, _data_equal)
221 #ifdef IP_SET_HASH_WITH_NETS
222 #define mtype_do_data_match IPSET_TOKEN(MTYPE, _do_data_match)
224 #define mtype_do_data_match(d) 1
226 #define mtype_data_set_flags IPSET_TOKEN(MTYPE, _data_set_flags)
227 #define mtype_data_reset_elem IPSET_TOKEN(MTYPE, _data_reset_elem)
228 #define mtype_data_reset_flags IPSET_TOKEN(MTYPE, _data_reset_flags)
229 #define mtype_data_netmask IPSET_TOKEN(MTYPE, _data_netmask)
230 #define mtype_data_list IPSET_TOKEN(MTYPE, _data_list)
231 #define mtype_data_next IPSET_TOKEN(MTYPE, _data_next)
232 #define mtype_elem IPSET_TOKEN(MTYPE, _elem)
234 #define mtype_ahash_destroy IPSET_TOKEN(MTYPE, _ahash_destroy)
235 #define mtype_ext_cleanup IPSET_TOKEN(MTYPE, _ext_cleanup)
236 #define mtype_add_cidr IPSET_TOKEN(MTYPE, _add_cidr)
237 #define mtype_del_cidr IPSET_TOKEN(MTYPE, _del_cidr)
238 #define mtype_ahash_memsize IPSET_TOKEN(MTYPE, _ahash_memsize)
239 #define mtype_flush IPSET_TOKEN(MTYPE, _flush)
240 #define mtype_destroy IPSET_TOKEN(MTYPE, _destroy)
241 #define mtype_same_set IPSET_TOKEN(MTYPE, _same_set)
242 #define mtype_kadt IPSET_TOKEN(MTYPE, _kadt)
243 #define mtype_uadt IPSET_TOKEN(MTYPE, _uadt)
245 #define mtype_add IPSET_TOKEN(MTYPE, _add)
246 #define mtype_del IPSET_TOKEN(MTYPE, _del)
247 #define mtype_test_cidrs IPSET_TOKEN(MTYPE, _test_cidrs)
248 #define mtype_test IPSET_TOKEN(MTYPE, _test)
249 #define mtype_uref IPSET_TOKEN(MTYPE, _uref)
250 #define mtype_expire IPSET_TOKEN(MTYPE, _expire)
251 #define mtype_resize IPSET_TOKEN(MTYPE, _resize)
252 #define mtype_head IPSET_TOKEN(MTYPE, _head)
253 #define mtype_list IPSET_TOKEN(MTYPE, _list)
254 #define mtype_gc IPSET_TOKEN(MTYPE, _gc)
255 #define mtype_gc_init IPSET_TOKEN(MTYPE, _gc_init)
256 #define mtype_variant IPSET_TOKEN(MTYPE, _variant)
257 #define mtype_data_match IPSET_TOKEN(MTYPE, _data_match)
260 #define HKEY_DATALEN sizeof(struct mtype_elem)
265 #define HKEY(data, initval, htable_bits) \
267 const u32 *__k = (const u32 *)data; \
268 u32 __l = HKEY_DATALEN / sizeof(u32); \
270 BUILD_BUG_ON(HKEY_DATALEN % sizeof(u32) != 0); \
272 jhash2(__k, __l, initval) & jhash_mask(htable_bits); \
275 /* The generic hash structure */
277 struct htable __rcu *table; /* the hash table */
278 struct timer_list gc; /* garbage collection when timeout enabled */
279 struct ip_set *set; /* attached to this ip_set */
280 u32 maxelem; /* max elements in the hash */
281 u32 initval; /* random jhash init value */
282 #ifdef IP_SET_HASH_WITH_MARKMASK
283 u32 markmask; /* markmask value for mark mask to store */
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 struct mtype_elem next; /* temporary storage for uadd */
292 #ifdef IP_SET_HASH_WITH_NETS
293 struct net_prefixes nets[NLEN]; /* 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 n)
306 /* Add in increasing prefix order, so larger cidr first */
307 for (i = 0, j = -1; i < NLEN && 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 n)
328 u8 i, j, net_end = NLEN - 1;
330 for (i = 0; i < NLEN; 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)
348 return sizeof(*h) + sizeof(*t);
351 /* Get the ith element from the array block n */
352 #define ahash_data(n, i, dsize) \
353 ((struct mtype_elem *)((n)->value + ((i) * (dsize))))
356 mtype_ext_cleanup(struct ip_set *set, struct hbucket *n)
360 for (i = 0; i < n->pos; i++)
361 if (test_bit(i, n->used))
362 ip_set_ext_destroy(set, ahash_data(n, i, set->dsize));
365 /* Flush a hash type of set: destroy all elements */
367 mtype_flush(struct ip_set *set)
369 struct htype *h = set->data;
374 t = ipset_dereference_protected(h->table, set);
375 for (i = 0; i < jhash_size(t->htable_bits); i++) {
376 n = __ipset_dereference_protected(hbucket(t, i), 1);
379 if (set->extensions & IPSET_EXT_DESTROY)
380 mtype_ext_cleanup(set, n);
381 /* FIXME: use slab cache */
382 rcu_assign_pointer(hbucket(t, i), NULL);
385 #ifdef IP_SET_HASH_WITH_NETS
386 memset(h->nets, 0, sizeof(h->nets));
392 /* Destroy the hashtable part of the set */
394 mtype_ahash_destroy(struct ip_set *set, struct htable *t, bool ext_destroy)
399 for (i = 0; i < jhash_size(t->htable_bits); i++) {
400 n = __ipset_dereference_protected(hbucket(t, i), 1);
403 if (set->extensions & IPSET_EXT_DESTROY && ext_destroy)
404 mtype_ext_cleanup(set, n);
405 /* FIXME: use slab cache */
412 /* Destroy a hash type of set */
414 mtype_destroy(struct ip_set *set)
416 struct htype *h = set->data;
418 if (SET_WITH_TIMEOUT(set))
419 del_timer_sync(&h->gc);
421 mtype_ahash_destroy(set,
422 __ipset_dereference_protected(h->table, 1), true);
429 mtype_gc_init(struct ip_set *set, void (*gc)(struct timer_list *t))
431 struct htype *h = set->data;
433 timer_setup(&h->gc, gc, 0);
434 mod_timer(&h->gc, jiffies + IPSET_GC_PERIOD(set->timeout) * HZ);
435 pr_debug("gc initialized, run in every %u\n",
436 IPSET_GC_PERIOD(set->timeout));
440 mtype_same_set(const struct ip_set *a, const struct ip_set *b)
442 const struct htype *x = a->data;
443 const struct htype *y = b->data;
445 /* Resizing changes htable_bits, so we ignore it */
446 return x->maxelem == y->maxelem &&
447 a->timeout == b->timeout &&
448 #ifdef IP_SET_HASH_WITH_NETMASK
449 x->netmask == y->netmask &&
451 #ifdef IP_SET_HASH_WITH_MARKMASK
452 x->markmask == y->markmask &&
454 a->extensions == b->extensions;
457 /* Delete expired elements from the hashtable */
459 mtype_expire(struct ip_set *set, struct htype *h)
462 struct hbucket *n, *tmp;
463 struct mtype_elem *data;
465 size_t dsize = set->dsize;
466 #ifdef IP_SET_HASH_WITH_NETS
470 t = ipset_dereference_protected(h->table, set);
471 for (i = 0; i < jhash_size(t->htable_bits); i++) {
472 n = __ipset_dereference_protected(hbucket(t, i), 1);
475 for (j = 0, d = 0; j < n->pos; j++) {
476 if (!test_bit(j, n->used)) {
480 data = ahash_data(n, j, dsize);
481 if (!ip_set_timeout_expired(ext_timeout(data, set)))
483 pr_debug("expired %u/%u\n", i, j);
484 clear_bit(j, n->used);
485 smp_mb__after_atomic();
486 #ifdef IP_SET_HASH_WITH_NETS
487 for (k = 0; k < IPSET_NET_COUNT; k++)
489 NCIDR_PUT(DCIDR_GET(data->cidr, k)),
492 ip_set_ext_destroy(set, data);
496 if (d >= AHASH_INIT_SIZE) {
498 rcu_assign_pointer(hbucket(t, i), NULL);
502 tmp = kzalloc(sizeof(*tmp) +
503 (n->size - AHASH_INIT_SIZE) * dsize,
506 /* Still try to delete expired elements */
508 tmp->size = n->size - AHASH_INIT_SIZE;
509 for (j = 0, d = 0; j < n->pos; j++) {
510 if (!test_bit(j, n->used))
512 data = ahash_data(n, j, dsize);
513 memcpy(tmp->value + d * dsize, data, dsize);
514 set_bit(d, tmp->used);
518 set->ext_size -= ext_size(AHASH_INIT_SIZE, dsize);
519 rcu_assign_pointer(hbucket(t, i), tmp);
526 mtype_gc(struct timer_list *t)
528 struct htype *h = from_timer(h, t, gc);
529 struct ip_set *set = h->set;
531 pr_debug("called\n");
532 spin_lock_bh(&set->lock);
533 mtype_expire(set, h);
534 spin_unlock_bh(&set->lock);
536 h->gc.expires = jiffies + IPSET_GC_PERIOD(set->timeout) * HZ;
540 /* Resize a hash: create a new hash table with doubling the hashsize
541 * and inserting the elements to it. Repeat until we succeed or
542 * fail due to memory pressures.
545 mtype_resize(struct ip_set *set, bool retried)
547 struct htype *h = set->data;
548 struct htable *t, *orig;
550 size_t extsize, dsize = set->dsize;
551 #ifdef IP_SET_HASH_WITH_NETS
553 struct mtype_elem *tmp;
555 struct mtype_elem *data;
556 struct mtype_elem *d;
557 struct hbucket *n, *m;
561 #ifdef IP_SET_HASH_WITH_NETS
562 tmp = kmalloc(dsize, GFP_KERNEL);
567 orig = rcu_dereference_bh_nfnl(h->table);
568 htable_bits = orig->htable_bits;
569 rcu_read_unlock_bh();
575 /* In case we have plenty of memory :-) */
576 pr_warn("Cannot increase the hashsize of set %s further\n",
578 ret = -IPSET_ERR_HASH_FULL;
581 t = ip_set_alloc(htable_size(htable_bits));
586 t->htable_bits = htable_bits;
588 spin_lock_bh(&set->lock);
589 orig = __ipset_dereference_protected(h->table, 1);
590 /* There can't be another parallel resizing, but dumping is possible */
591 atomic_set(&orig->ref, 1);
592 atomic_inc(&orig->uref);
594 pr_debug("attempt to resize set %s from %u to %u, t %p\n",
595 set->name, orig->htable_bits, htable_bits, orig);
596 for (i = 0; i < jhash_size(orig->htable_bits); i++) {
597 n = __ipset_dereference_protected(hbucket(orig, i), 1);
600 for (j = 0; j < n->pos; j++) {
601 if (!test_bit(j, n->used))
603 data = ahash_data(n, j, dsize);
604 #ifdef IP_SET_HASH_WITH_NETS
605 /* We have readers running parallel with us,
606 * so the live data cannot be modified.
609 memcpy(tmp, data, dsize);
611 mtype_data_reset_flags(data, &flags);
613 key = HKEY(data, h->initval, htable_bits);
614 m = __ipset_dereference_protected(hbucket(t, key), 1);
616 m = kzalloc(sizeof(*m) +
617 AHASH_INIT_SIZE * dsize,
623 m->size = AHASH_INIT_SIZE;
624 extsize += ext_size(AHASH_INIT_SIZE, dsize);
625 RCU_INIT_POINTER(hbucket(t, key), m);
626 } else if (m->pos >= m->size) {
629 if (m->size >= AHASH_MAX(h)) {
632 ht = kzalloc(sizeof(*ht) +
633 (m->size + AHASH_INIT_SIZE)
641 memcpy(ht, m, sizeof(struct hbucket) +
643 ht->size = m->size + AHASH_INIT_SIZE;
644 extsize += ext_size(AHASH_INIT_SIZE, dsize);
647 RCU_INIT_POINTER(hbucket(t, key), ht);
649 d = ahash_data(m, m->pos, dsize);
650 memcpy(d, data, dsize);
651 set_bit(m->pos++, m->used);
652 #ifdef IP_SET_HASH_WITH_NETS
653 mtype_data_reset_flags(d, &flags);
657 rcu_assign_pointer(h->table, t);
658 set->ext_size = extsize;
660 spin_unlock_bh(&set->lock);
662 /* Give time to other readers of the set */
665 pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
666 orig->htable_bits, orig, t->htable_bits, t);
667 /* If there's nobody else dumping the table, destroy it */
668 if (atomic_dec_and_test(&orig->uref)) {
669 pr_debug("Table destroy by resize %p\n", orig);
670 mtype_ahash_destroy(set, orig, false);
674 #ifdef IP_SET_HASH_WITH_NETS
680 atomic_set(&orig->ref, 0);
681 atomic_dec(&orig->uref);
682 spin_unlock_bh(&set->lock);
683 mtype_ahash_destroy(set, t, false);
689 /* Add an element to a hash and update the internal counters when succeeded,
690 * otherwise report the proper error code.
693 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
694 struct ip_set_ext *mext, u32 flags)
696 struct htype *h = set->data;
698 const struct mtype_elem *d = value;
699 struct mtype_elem *data;
700 struct hbucket *n, *old = ERR_PTR(-ENOENT);
702 bool flag_exist = flags & IPSET_FLAG_EXIST;
703 bool deleted = false, forceadd = false, reuse = false;
706 if (set->elements >= h->maxelem) {
707 if (SET_WITH_TIMEOUT(set))
708 /* FIXME: when set is full, we slow down here */
709 mtype_expire(set, h);
710 if (set->elements >= h->maxelem && SET_WITH_FORCEADD(set))
714 t = ipset_dereference_protected(h->table, set);
715 key = HKEY(value, h->initval, t->htable_bits);
716 n = __ipset_dereference_protected(hbucket(t, key), 1);
718 if (forceadd || set->elements >= h->maxelem)
721 n = kzalloc(sizeof(*n) + AHASH_INIT_SIZE * set->dsize,
725 n->size = AHASH_INIT_SIZE;
726 set->ext_size += ext_size(AHASH_INIT_SIZE, set->dsize);
729 for (i = 0; i < n->pos; i++) {
730 if (!test_bit(i, n->used)) {
731 /* Reuse first deleted entry */
733 deleted = reuse = true;
738 data = ahash_data(n, i, set->dsize);
739 if (mtype_data_equal(data, d, &multi)) {
741 (SET_WITH_TIMEOUT(set) &&
742 ip_set_timeout_expired(ext_timeout(data, set)))) {
743 /* Just the extensions could be overwritten */
745 goto overwrite_extensions;
747 return -IPSET_ERR_EXIST;
749 /* Reuse first timed out entry */
750 if (SET_WITH_TIMEOUT(set) &&
751 ip_set_timeout_expired(ext_timeout(data, set)) &&
757 if (reuse || forceadd) {
758 data = ahash_data(n, j, set->dsize);
760 #ifdef IP_SET_HASH_WITH_NETS
761 for (i = 0; i < IPSET_NET_COUNT; i++)
763 NCIDR_PUT(DCIDR_GET(data->cidr, i)),
766 ip_set_ext_destroy(set, data);
771 if (set->elements >= h->maxelem)
773 /* Create a new slot */
774 if (n->pos >= n->size) {
775 TUNE_AHASH_MAX(h, multi);
776 if (n->size >= AHASH_MAX(h)) {
777 /* Trigger rehashing */
778 mtype_data_next(&h->next, d);
782 n = kzalloc(sizeof(*n) +
783 (old->size + AHASH_INIT_SIZE) * set->dsize,
787 memcpy(n, old, sizeof(struct hbucket) +
788 old->size * set->dsize);
789 n->size = old->size + AHASH_INIT_SIZE;
790 set->ext_size += ext_size(AHASH_INIT_SIZE, set->dsize);
795 data = ahash_data(n, j, set->dsize);
798 #ifdef IP_SET_HASH_WITH_NETS
799 for (i = 0; i < IPSET_NET_COUNT; i++)
800 mtype_add_cidr(h, NCIDR_PUT(DCIDR_GET(d->cidr, i)), i);
802 memcpy(data, d, sizeof(struct mtype_elem));
803 overwrite_extensions:
804 #ifdef IP_SET_HASH_WITH_NETS
805 mtype_data_set_flags(data, flags);
807 if (SET_WITH_COUNTER(set))
808 ip_set_init_counter(ext_counter(data, set), ext);
809 if (SET_WITH_COMMENT(set))
810 ip_set_init_comment(set, ext_comment(data, set), ext);
811 if (SET_WITH_SKBINFO(set))
812 ip_set_init_skbinfo(ext_skbinfo(data, set), ext);
813 /* Must come last for the case when timed out entry is reused */
814 if (SET_WITH_TIMEOUT(set))
815 ip_set_timeout_set(ext_timeout(data, set), ext->timeout);
816 smp_mb__before_atomic();
818 if (old != ERR_PTR(-ENOENT)) {
819 rcu_assign_pointer(hbucket(t, key), n);
827 pr_warn("Set %s is full, maxelem %u reached\n",
828 set->name, h->maxelem);
829 return -IPSET_ERR_HASH_FULL;
832 /* Delete an element from the hash and free up space if possible.
835 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
836 struct ip_set_ext *mext, u32 flags)
838 struct htype *h = set->data;
840 const struct mtype_elem *d = value;
841 struct mtype_elem *data;
843 int i, j, k, ret = -IPSET_ERR_EXIST;
845 size_t dsize = set->dsize;
847 t = ipset_dereference_protected(h->table, set);
848 key = HKEY(value, h->initval, t->htable_bits);
849 n = __ipset_dereference_protected(hbucket(t, key), 1);
852 for (i = 0, k = 0; i < n->pos; i++) {
853 if (!test_bit(i, n->used)) {
857 data = ahash_data(n, i, dsize);
858 if (!mtype_data_equal(data, d, &multi))
860 if (SET_WITH_TIMEOUT(set) &&
861 ip_set_timeout_expired(ext_timeout(data, set)))
865 clear_bit(i, n->used);
866 smp_mb__after_atomic();
870 #ifdef IP_SET_HASH_WITH_NETS
871 for (j = 0; j < IPSET_NET_COUNT; j++)
872 mtype_del_cidr(h, NCIDR_PUT(DCIDR_GET(d->cidr, j)),
875 ip_set_ext_destroy(set, data);
877 for (; i < n->pos; i++) {
878 if (!test_bit(i, n->used))
881 if (n->pos == 0 && k == 0) {
882 set->ext_size -= ext_size(n->size, dsize);
883 rcu_assign_pointer(hbucket(t, key), NULL);
885 } else if (k >= AHASH_INIT_SIZE) {
886 struct hbucket *tmp = kzalloc(sizeof(*tmp) +
887 (n->size - AHASH_INIT_SIZE) * dsize,
891 tmp->size = n->size - AHASH_INIT_SIZE;
892 for (j = 0, k = 0; j < n->pos; j++) {
893 if (!test_bit(j, n->used))
895 data = ahash_data(n, j, dsize);
896 memcpy(tmp->value + k * dsize, data, dsize);
897 set_bit(k, tmp->used);
901 set->ext_size -= ext_size(AHASH_INIT_SIZE, dsize);
902 rcu_assign_pointer(hbucket(t, key), tmp);
913 mtype_data_match(struct mtype_elem *data, const struct ip_set_ext *ext,
914 struct ip_set_ext *mext, struct ip_set *set, u32 flags)
916 if (!ip_set_match_extensions(set, ext, mext, flags, data))
918 /* nomatch entries return -ENOTEMPTY */
919 return mtype_do_data_match(data);
922 #ifdef IP_SET_HASH_WITH_NETS
923 /* Special test function which takes into account the different network
924 * sizes added to the set
927 mtype_test_cidrs(struct ip_set *set, struct mtype_elem *d,
928 const struct ip_set_ext *ext,
929 struct ip_set_ext *mext, u32 flags)
931 struct htype *h = set->data;
932 struct htable *t = rcu_dereference_bh(h->table);
934 struct mtype_elem *data;
935 #if IPSET_NET_COUNT == 2
936 struct mtype_elem orig = *d;
937 int ret, i, j = 0, k;
943 pr_debug("test by nets\n");
944 for (; j < NLEN && h->nets[j].cidr[0] && !multi; j++) {
945 #if IPSET_NET_COUNT == 2
946 mtype_data_reset_elem(d, &orig);
947 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]), false);
948 for (k = 0; k < NLEN && h->nets[k].cidr[1] && !multi;
950 mtype_data_netmask(d, NCIDR_GET(h->nets[k].cidr[1]),
953 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]));
955 key = HKEY(d, h->initval, t->htable_bits);
956 n = rcu_dereference_bh(hbucket(t, key));
959 for (i = 0; i < n->pos; i++) {
960 if (!test_bit(i, n->used))
962 data = ahash_data(n, i, set->dsize);
963 if (!mtype_data_equal(data, d, &multi))
965 ret = mtype_data_match(data, ext, mext, set, flags);
968 #ifdef IP_SET_HASH_WITH_MULTI
969 /* No match, reset multiple match flag */
973 #if IPSET_NET_COUNT == 2
981 /* Test whether the element is added to the set */
983 mtype_test(struct ip_set *set, void *value, const struct ip_set_ext *ext,
984 struct ip_set_ext *mext, u32 flags)
986 struct htype *h = set->data;
988 struct mtype_elem *d = value;
990 struct mtype_elem *data;
994 t = rcu_dereference_bh(h->table);
995 #ifdef IP_SET_HASH_WITH_NETS
996 /* If we test an IP address and not a network address,
997 * try all possible network sizes
999 for (i = 0; i < IPSET_NET_COUNT; i++)
1000 if (DCIDR_GET(d->cidr, i) != HOST_MASK)
1002 if (i == IPSET_NET_COUNT) {
1003 ret = mtype_test_cidrs(set, d, ext, mext, flags);
1008 key = HKEY(d, h->initval, t->htable_bits);
1009 n = rcu_dereference_bh(hbucket(t, key));
1014 for (i = 0; i < n->pos; i++) {
1015 if (!test_bit(i, n->used))
1017 data = ahash_data(n, i, set->dsize);
1018 if (!mtype_data_equal(data, d, &multi))
1020 ret = mtype_data_match(data, ext, mext, set, flags);
1028 /* Reply a HEADER request: fill out the header part of the set */
1030 mtype_head(struct ip_set *set, struct sk_buff *skb)
1032 struct htype *h = set->data;
1033 const struct htable *t;
1034 struct nlattr *nested;
1038 /* If any members have expired, set->elements will be wrong
1039 * mytype_expire function will update it with the right count.
1040 * we do not hold set->lock here, so grab it first.
1041 * set->elements can still be incorrect in the case of a huge set,
1042 * because elements might time out during the listing.
1044 if (SET_WITH_TIMEOUT(set)) {
1045 spin_lock_bh(&set->lock);
1046 mtype_expire(set, h);
1047 spin_unlock_bh(&set->lock);
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 = nla_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 nla_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 = nla_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]++) {
1134 incomplete = skb_tail_pointer(skb);
1135 n = rcu_dereference(hbucket(t, cb->args[IPSET_CB_ARG0]));
1136 pr_debug("cb->arg bucket: %lu, t %p n %p\n",
1137 cb->args[IPSET_CB_ARG0], t, n);
1140 for (i = 0; i < n->pos; i++) {
1141 if (!test_bit(i, n->used))
1143 e = ahash_data(n, i, set->dsize);
1144 if (SET_WITH_TIMEOUT(set) &&
1145 ip_set_timeout_expired(ext_timeout(e, set)))
1147 pr_debug("list hash %lu hbucket %p i %u, data %p\n",
1148 cb->args[IPSET_CB_ARG0], n, i, e);
1149 nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1151 if (cb->args[IPSET_CB_ARG0] == first) {
1152 nla_nest_cancel(skb, atd);
1156 goto nla_put_failure;
1158 if (mtype_data_list(skb, e))
1159 goto nla_put_failure;
1160 if (ip_set_put_extensions(skb, set, e, true))
1161 goto nla_put_failure;
1162 nla_nest_end(skb, nested);
1165 nla_nest_end(skb, atd);
1166 /* Set listing finished */
1167 cb->args[IPSET_CB_ARG0] = 0;
1172 nlmsg_trim(skb, incomplete);
1173 if (unlikely(first == cb->args[IPSET_CB_ARG0])) {
1174 pr_warn("Can't list set %s: one bucket does not fit into a message. Please report it!\n",
1176 cb->args[IPSET_CB_ARG0] = 0;
1179 nla_nest_end(skb, atd);
1187 IPSET_TOKEN(MTYPE, _kadt)(struct ip_set *set, const struct sk_buff *skb,
1188 const struct xt_action_param *par,
1189 enum ipset_adt adt, struct ip_set_adt_opt *opt);
1192 IPSET_TOKEN(MTYPE, _uadt)(struct ip_set *set, struct nlattr *tb[],
1193 enum ipset_adt adt, u32 *lineno, u32 flags,
1196 static const struct ip_set_type_variant mtype_variant = {
1200 [IPSET_ADD] = mtype_add,
1201 [IPSET_DEL] = mtype_del,
1202 [IPSET_TEST] = mtype_test,
1204 .destroy = mtype_destroy,
1205 .flush = mtype_flush,
1209 .resize = mtype_resize,
1210 .same_set = mtype_same_set,
1213 #ifdef IP_SET_EMIT_CREATE
1215 IPSET_TOKEN(HTYPE, _create)(struct net *net, struct ip_set *set,
1216 struct nlattr *tb[], u32 flags)
1218 u32 hashsize = IPSET_DEFAULT_HASHSIZE, maxelem = IPSET_DEFAULT_MAXELEM;
1219 #ifdef IP_SET_HASH_WITH_MARKMASK
1223 #ifdef IP_SET_HASH_WITH_NETMASK
1230 pr_debug("Create set %s with family %s\n",
1231 set->name, set->family == NFPROTO_IPV4 ? "inet" : "inet6");
1233 #ifdef IP_SET_PROTO_UNDEF
1234 if (set->family != NFPROTO_UNSPEC)
1235 return -IPSET_ERR_INVALID_FAMILY;
1237 if (!(set->family == NFPROTO_IPV4 || set->family == NFPROTO_IPV6))
1238 return -IPSET_ERR_INVALID_FAMILY;
1241 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_HASHSIZE) ||
1242 !ip_set_optattr_netorder(tb, IPSET_ATTR_MAXELEM) ||
1243 !ip_set_optattr_netorder(tb, IPSET_ATTR_TIMEOUT) ||
1244 !ip_set_optattr_netorder(tb, IPSET_ATTR_CADT_FLAGS)))
1245 return -IPSET_ERR_PROTOCOL;
1247 #ifdef IP_SET_HASH_WITH_MARKMASK
1248 /* Separated condition in order to avoid directive in argument list */
1249 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_MARKMASK)))
1250 return -IPSET_ERR_PROTOCOL;
1252 markmask = 0xffffffff;
1253 if (tb[IPSET_ATTR_MARKMASK]) {
1254 markmask = ntohl(nla_get_be32(tb[IPSET_ATTR_MARKMASK]));
1256 return -IPSET_ERR_INVALID_MARKMASK;
1260 #ifdef IP_SET_HASH_WITH_NETMASK
1261 netmask = set->family == NFPROTO_IPV4 ? 32 : 128;
1262 if (tb[IPSET_ATTR_NETMASK]) {
1263 netmask = nla_get_u8(tb[IPSET_ATTR_NETMASK]);
1265 if ((set->family == NFPROTO_IPV4 && netmask > 32) ||
1266 (set->family == NFPROTO_IPV6 && netmask > 128) ||
1268 return -IPSET_ERR_INVALID_NETMASK;
1272 if (tb[IPSET_ATTR_HASHSIZE]) {
1273 hashsize = ip_set_get_h32(tb[IPSET_ATTR_HASHSIZE]);
1274 if (hashsize < IPSET_MIMINAL_HASHSIZE)
1275 hashsize = IPSET_MIMINAL_HASHSIZE;
1278 if (tb[IPSET_ATTR_MAXELEM])
1279 maxelem = ip_set_get_h32(tb[IPSET_ATTR_MAXELEM]);
1282 h = kzalloc(hsize, GFP_KERNEL);
1286 hbits = htable_bits(hashsize);
1287 hsize = htable_size(hbits);
1292 t = ip_set_alloc(hsize);
1297 h->maxelem = maxelem;
1298 #ifdef IP_SET_HASH_WITH_NETMASK
1299 h->netmask = netmask;
1301 #ifdef IP_SET_HASH_WITH_MARKMASK
1302 h->markmask = markmask;
1304 get_random_bytes(&h->initval, sizeof(h->initval));
1306 t->htable_bits = hbits;
1307 RCU_INIT_POINTER(h->table, t);
1311 #ifndef IP_SET_PROTO_UNDEF
1312 if (set->family == NFPROTO_IPV4) {
1314 set->variant = &IPSET_TOKEN(HTYPE, 4_variant);
1315 set->dsize = ip_set_elem_len(set, tb,
1316 sizeof(struct IPSET_TOKEN(HTYPE, 4_elem)),
1317 __alignof__(struct IPSET_TOKEN(HTYPE, 4_elem)));
1318 #ifndef IP_SET_PROTO_UNDEF
1320 set->variant = &IPSET_TOKEN(HTYPE, 6_variant);
1321 set->dsize = ip_set_elem_len(set, tb,
1322 sizeof(struct IPSET_TOKEN(HTYPE, 6_elem)),
1323 __alignof__(struct IPSET_TOKEN(HTYPE, 6_elem)));
1326 set->timeout = IPSET_NO_TIMEOUT;
1327 if (tb[IPSET_ATTR_TIMEOUT]) {
1328 set->timeout = ip_set_timeout_uget(tb[IPSET_ATTR_TIMEOUT]);
1329 #ifndef IP_SET_PROTO_UNDEF
1330 if (set->family == NFPROTO_IPV4)
1332 IPSET_TOKEN(HTYPE, 4_gc_init)(set,
1333 IPSET_TOKEN(HTYPE, 4_gc));
1334 #ifndef IP_SET_PROTO_UNDEF
1336 IPSET_TOKEN(HTYPE, 6_gc_init)(set,
1337 IPSET_TOKEN(HTYPE, 6_gc));
1340 pr_debug("create %s hashsize %u (%u) maxelem %u: %p(%p)\n",
1341 set->name, jhash_size(t->htable_bits),
1342 t->htable_bits, h->maxelem, set->data, t);
1346 #endif /* IP_SET_EMIT_CREATE */