xref: /linux/net/netfilter/ipset/ip_set_hash_gen.h (revision b2128290c29902315e632ea59e0504d6bc9e9b42)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /* Copyright (C) 2013 Jozsef Kadlecsik <kadlec@netfilter.org> */
3 
4 #ifndef _IP_SET_HASH_GEN_H
5 #define _IP_SET_HASH_GEN_H
6 
7 #include <linux/rcupdate.h>
8 #include <linux/rcupdate_wait.h>
9 #include <linux/jhash.h>
10 #include <linux/types.h>
11 #include <linux/netfilter/nfnetlink.h>
12 #include <linux/netfilter/ipset/ip_set.h>
13 
14 #define __ipset_dereference(p)		\
15 	rcu_dereference_protected(p, 1)
16 #define ipset_dereference_nfnl(p)	\
17 	rcu_dereference_protected(p,	\
18 		lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET))
19 #define ipset_dereference_set(p, set) 	\
20 	rcu_dereference_protected(p,	\
21 		lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET) || \
22 		lockdep_is_held(&(set)->lock))
23 #define ipset_dereference_bh_nfnl(p)	\
24 	rcu_dereference_bh_check(p, 	\
25 		lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET))
26 
27 /* Hashing which uses arrays to resolve clashing. The hash table is resized
28  * (doubled) when searching becomes too long.
29  * Internally jhash is used with the assumption that the size of the
30  * stored data is a multiple of sizeof(u32).
31  *
32  * Readers and resizing
33  *
34  * Resizing can be triggered by userspace command only, and those
35  * are serialized by the nfnl mutex. During resizing the set is
36  * read-locked, so the only possible concurrent operations are
37  * the kernel side readers. Those must be protected by proper RCU locking.
38  */
39 
40 /* Number of elements to store in an initial array block */
41 #define AHASH_INIT_SIZE			2
42 /* Max number of elements to store in an array block */
43 #define AHASH_MAX_SIZE			(6 * AHASH_INIT_SIZE)
44 /* Max muber of elements in the array block when tuned */
45 #define AHASH_MAX_TUNED			64
46 #define AHASH_MAX(h)			((h)->bucketsize)
47 
48 /* A hash bucket */
49 struct hbucket {
50 	struct rcu_head rcu;	/* for call_rcu */
51 	/* Which positions are used in the array */
52 	DECLARE_BITMAP(used, AHASH_MAX_TUNED);
53 	u8 size;		/* size of the array */
54 	u8 pos;			/* position of the first free entry */
55 	unsigned char value[]	/* the array of the values */
56 		__aligned(__alignof__(u64));
57 };
58 
59 /* Region size for locking == 2^HTABLE_REGION_BITS */
60 #define HTABLE_REGION_BITS	10
61 #define ahash_numof_locks(htable_bits)		\
62 	((htable_bits) < HTABLE_REGION_BITS ? 1	\
63 		: jhash_size((htable_bits) - HTABLE_REGION_BITS))
64 #define ahash_sizeof_regions(htable_bits)		\
65 	(ahash_numof_locks(htable_bits) * sizeof(struct ip_set_region))
66 #define ahash_region(n)		\
67 	((n) / jhash_size(HTABLE_REGION_BITS))
68 #define ahash_bucket_start(h,  htable_bits)	\
69 	((htable_bits) < HTABLE_REGION_BITS ? 0	\
70 		: (h) * jhash_size(HTABLE_REGION_BITS))
71 #define ahash_bucket_end(h,  htable_bits)	\
72 	((htable_bits) < HTABLE_REGION_BITS ? jhash_size(htable_bits)	\
73 		: ((h) + 1) * jhash_size(HTABLE_REGION_BITS))
74 
75 struct htable_gc {
76 	struct delayed_work dwork;
77 	struct ip_set *set;	/* Set the gc belongs to */
78 	u32 region;		/* Last gc run position */
79 };
80 
81 /* The hash table: the table size stored here in order to make resizing easy */
82 struct htable {
83 	atomic_t ref;		/* References for resizing */
84 	atomic_t uref;		/* References for dumping and gc */
85 	u8 htable_bits;		/* size of hash table == 2^htable_bits */
86 	u32 maxelem;		/* Maxelem per region */
87 	struct ip_set_region *hregion;	/* Region locks and ext sizes */
88 	struct hbucket __rcu *bucket[]; /* hashtable buckets */
89 };
90 
91 #define hbucket(h, i)		((h)->bucket[i])
92 #define ext_size(n, dsize)	\
93 	(sizeof(struct hbucket) + (n) * (dsize))
94 
95 #ifndef IPSET_NET_COUNT
96 #define IPSET_NET_COUNT		1
97 #endif
98 
99 /* Book-keeping of the prefixes added to the set */
100 struct net_prefixes {
101 	u32 nets[IPSET_NET_COUNT]; /* number of elements for this cidr */
102 	u8 cidr[IPSET_NET_COUNT];  /* the cidr value */
103 };
104 
105 /* Compute the hash table size */
106 static size_t
107 htable_size(u8 hbits)
108 {
109 	size_t hsize;
110 
111 	/* We must fit both into u32 in jhash and INT_MAX in kvmalloc_node() */
112 	if (hbits > 31)
113 		return 0;
114 	hsize = jhash_size(hbits);
115 	if ((INT_MAX - sizeof(struct htable)) / sizeof(struct hbucket *)
116 	    < hsize)
117 		return 0;
118 
119 	return hsize * sizeof(struct hbucket *) + sizeof(struct htable);
120 }
121 
122 #ifdef IP_SET_HASH_WITH_NETS
123 #if IPSET_NET_COUNT > 1
124 #define __CIDR(cidr, i)		(cidr[i])
125 #else
126 #define __CIDR(cidr, i)		(cidr)
127 #endif
128 
129 /* cidr + 1 is stored in net_prefixes to support /0 */
130 #define NCIDR_PUT(cidr)		((cidr) + 1)
131 #define NCIDR_GET(cidr)		((cidr) - 1)
132 
133 #ifdef IP_SET_HASH_WITH_NETS_PACKED
134 /* When cidr is packed with nomatch, cidr - 1 is stored in the data entry */
135 #define DCIDR_PUT(cidr)		((cidr) - 1)
136 #define DCIDR_GET(cidr, i)	(__CIDR(cidr, i) + 1)
137 #else
138 #define DCIDR_PUT(cidr)		(cidr)
139 #define DCIDR_GET(cidr, i)	__CIDR(cidr, i)
140 #endif
141 
142 #define INIT_CIDR(cidr, host_mask)	\
143 	DCIDR_PUT(((cidr) ? NCIDR_GET(cidr) : host_mask))
144 
145 #ifdef IP_SET_HASH_WITH_NET0
146 /* cidr from 0 to HOST_MASK value and c = cidr + 1 */
147 #define NLEN			(HOST_MASK + 1)
148 #define CIDR_POS(c)		((c) - 1)
149 #else
150 /* cidr from 1 to HOST_MASK value and c = cidr + 1 */
151 #define NLEN			HOST_MASK
152 #define CIDR_POS(c)		((c) - 2)
153 #endif
154 
155 #else
156 #define NLEN			0
157 #endif /* IP_SET_HASH_WITH_NETS */
158 
159 #define SET_ELEM_EXPIRED(set, d)	\
160 	(SET_WITH_TIMEOUT(set) &&	\
161 	 ip_set_timeout_expired(ext_timeout(d, set)))
162 
163 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
164 static const union nf_inet_addr onesmask = {
165 	.all[0] = 0xffffffff,
166 	.all[1] = 0xffffffff,
167 	.all[2] = 0xffffffff,
168 	.all[3] = 0xffffffff
169 };
170 
171 static const union nf_inet_addr zeromask = {};
172 #endif
173 
174 #endif /* _IP_SET_HASH_GEN_H */
175 
176 #ifndef MTYPE
177 #error "MTYPE is not defined!"
178 #endif
179 
180 #ifndef HTYPE
181 #error "HTYPE is not defined!"
182 #endif
183 
184 #ifndef HOST_MASK
185 #error "HOST_MASK is not defined!"
186 #endif
187 
188 /* Family dependent templates */
189 
190 #undef ahash_data
191 #undef mtype_data_equal
192 #undef mtype_do_data_match
193 #undef mtype_data_set_flags
194 #undef mtype_data_reset_elem
195 #undef mtype_data_reset_flags
196 #undef mtype_data_netmask
197 #undef mtype_data_list
198 #undef mtype_data_next
199 #undef mtype_elem
200 
201 #undef mtype_ahash_destroy
202 #undef mtype_ext_cleanup
203 #undef mtype_add_cidr
204 #undef mtype_del_cidr
205 #undef mtype_ahash_memsize
206 #undef mtype_flush
207 #undef mtype_destroy
208 #undef mtype_same_set
209 #undef mtype_kadt
210 #undef mtype_uadt
211 
212 #undef mtype_add
213 #undef mtype_del
214 #undef mtype_test_cidrs
215 #undef mtype_test
216 #undef mtype_uref
217 #undef mtype_resize
218 #undef mtype_ext_size
219 #undef mtype_resize_ad
220 #undef mtype_head
221 #undef mtype_list
222 #undef mtype_gc_do
223 #undef mtype_gc
224 #undef mtype_gc_init
225 #undef mtype_cancel_gc
226 #undef mtype_variant
227 #undef mtype_data_match
228 
229 #undef htype
230 #undef HKEY
231 
232 #define mtype_data_equal	IPSET_TOKEN(MTYPE, _data_equal)
233 #ifdef IP_SET_HASH_WITH_NETS
234 #define mtype_do_data_match	IPSET_TOKEN(MTYPE, _do_data_match)
235 #else
236 #define mtype_do_data_match(d)	1
237 #endif
238 #define mtype_data_set_flags	IPSET_TOKEN(MTYPE, _data_set_flags)
239 #define mtype_data_reset_elem	IPSET_TOKEN(MTYPE, _data_reset_elem)
240 #define mtype_data_reset_flags	IPSET_TOKEN(MTYPE, _data_reset_flags)
241 #define mtype_data_netmask	IPSET_TOKEN(MTYPE, _data_netmask)
242 #define mtype_data_list		IPSET_TOKEN(MTYPE, _data_list)
243 #define mtype_data_next		IPSET_TOKEN(MTYPE, _data_next)
244 #define mtype_elem		IPSET_TOKEN(MTYPE, _elem)
245 
246 #define mtype_ahash_destroy	IPSET_TOKEN(MTYPE, _ahash_destroy)
247 #define mtype_ext_cleanup	IPSET_TOKEN(MTYPE, _ext_cleanup)
248 #define mtype_add_cidr		IPSET_TOKEN(MTYPE, _add_cidr)
249 #define mtype_del_cidr		IPSET_TOKEN(MTYPE, _del_cidr)
250 #define mtype_ahash_memsize	IPSET_TOKEN(MTYPE, _ahash_memsize)
251 #define mtype_flush		IPSET_TOKEN(MTYPE, _flush)
252 #define mtype_destroy		IPSET_TOKEN(MTYPE, _destroy)
253 #define mtype_same_set		IPSET_TOKEN(MTYPE, _same_set)
254 #define mtype_kadt		IPSET_TOKEN(MTYPE, _kadt)
255 #define mtype_uadt		IPSET_TOKEN(MTYPE, _uadt)
256 
257 #define mtype_add		IPSET_TOKEN(MTYPE, _add)
258 #define mtype_del		IPSET_TOKEN(MTYPE, _del)
259 #define mtype_test_cidrs	IPSET_TOKEN(MTYPE, _test_cidrs)
260 #define mtype_test		IPSET_TOKEN(MTYPE, _test)
261 #define mtype_uref		IPSET_TOKEN(MTYPE, _uref)
262 #define mtype_resize		IPSET_TOKEN(MTYPE, _resize)
263 #define mtype_ext_size		IPSET_TOKEN(MTYPE, _ext_size)
264 #define mtype_resize_ad		IPSET_TOKEN(MTYPE, _resize_ad)
265 #define mtype_head		IPSET_TOKEN(MTYPE, _head)
266 #define mtype_list		IPSET_TOKEN(MTYPE, _list)
267 #define mtype_gc_do		IPSET_TOKEN(MTYPE, _gc_do)
268 #define mtype_gc		IPSET_TOKEN(MTYPE, _gc)
269 #define mtype_gc_init		IPSET_TOKEN(MTYPE, _gc_init)
270 #define mtype_cancel_gc		IPSET_TOKEN(MTYPE, _cancel_gc)
271 #define mtype_variant		IPSET_TOKEN(MTYPE, _variant)
272 #define mtype_data_match	IPSET_TOKEN(MTYPE, _data_match)
273 
274 #ifndef HKEY_DATALEN
275 #define HKEY_DATALEN		sizeof(struct mtype_elem)
276 #endif
277 
278 #define htype			MTYPE
279 
280 #define HKEY(data, initval, htable_bits)			\
281 ({								\
282 	const u32 *__k = (const u32 *)data;			\
283 	u32 __l = HKEY_DATALEN / sizeof(u32);			\
284 								\
285 	BUILD_BUG_ON(HKEY_DATALEN % sizeof(u32) != 0);		\
286 								\
287 	jhash2(__k, __l, initval) & jhash_mask(htable_bits);	\
288 })
289 
290 /* The generic hash structure */
291 struct htype {
292 	struct htable __rcu *table; /* the hash table */
293 	struct htable_gc gc;	/* gc workqueue */
294 	u32 maxelem;		/* max elements in the hash */
295 	u32 initval;		/* random jhash init value */
296 #ifdef IP_SET_HASH_WITH_MARKMASK
297 	u32 markmask;		/* markmask value for mark mask to store */
298 #endif
299 	u8 bucketsize;		/* max elements in an array block */
300 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
301 	u8 netmask;		/* netmask value for subnets to store */
302 	union nf_inet_addr bitmask;	/* stores bitmask */
303 #endif
304 	struct list_head ad;	/* Resize add|del backlist */
305 	struct mtype_elem next; /* temporary storage for uadd */
306 #ifdef IP_SET_HASH_WITH_NETS
307 	struct net_prefixes nets[NLEN]; /* book-keeping of prefixes */
308 #endif
309 };
310 
311 /* ADD|DEL entries saved during resize */
312 struct mtype_resize_ad {
313 	struct list_head list;
314 	enum ipset_adt ad;	/* ADD|DEL element */
315 	struct mtype_elem d;	/* Element value */
316 	struct ip_set_ext ext;	/* Extensions for ADD */
317 	struct ip_set_ext mext;	/* Target extensions for ADD */
318 	u32 flags;		/* Flags for ADD */
319 };
320 
321 #ifdef IP_SET_HASH_WITH_NETS
322 /* Network cidr size book keeping when the hash stores different
323  * sized networks. cidr == real cidr + 1 to support /0.
324  */
325 static void
326 mtype_add_cidr(struct ip_set *set, struct htype *h, u8 cidr, u8 n)
327 {
328 	int i, j;
329 
330 	spin_lock_bh(&set->lock);
331 	/* Add in increasing prefix order, so larger cidr first */
332 	for (i = 0, j = -1; i < NLEN && h->nets[i].cidr[n]; i++) {
333 		if (j != -1) {
334 			continue;
335 		} else if (h->nets[i].cidr[n] < cidr) {
336 			j = i;
337 		} else if (h->nets[i].cidr[n] == cidr) {
338 			h->nets[CIDR_POS(cidr)].nets[n]++;
339 			goto unlock;
340 		}
341 	}
342 	if (j != -1) {
343 		for (; i > j; i--)
344 			h->nets[i].cidr[n] = h->nets[i - 1].cidr[n];
345 	}
346 	h->nets[i].cidr[n] = cidr;
347 	h->nets[CIDR_POS(cidr)].nets[n] = 1;
348 unlock:
349 	spin_unlock_bh(&set->lock);
350 }
351 
352 static void
353 mtype_del_cidr(struct ip_set *set, struct htype *h, u8 cidr, u8 n)
354 {
355 	u8 i, j, net_end = NLEN - 1;
356 
357 	spin_lock_bh(&set->lock);
358 	for (i = 0; i < NLEN; i++) {
359 		if (h->nets[i].cidr[n] != cidr)
360 			continue;
361 		h->nets[CIDR_POS(cidr)].nets[n]--;
362 		if (h->nets[CIDR_POS(cidr)].nets[n] > 0)
363 			goto unlock;
364 		for (j = i; j < net_end && h->nets[j].cidr[n]; j++)
365 			h->nets[j].cidr[n] = h->nets[j + 1].cidr[n];
366 		h->nets[j].cidr[n] = 0;
367 		goto unlock;
368 	}
369 unlock:
370 	spin_unlock_bh(&set->lock);
371 }
372 #endif
373 
374 /* Calculate the actual memory size of the set data */
375 static size_t
376 mtype_ahash_memsize(const struct htype *h, const struct htable *t)
377 {
378 	return sizeof(*h) + sizeof(*t) + ahash_sizeof_regions(t->htable_bits);
379 }
380 
381 /* Get the ith element from the array block n */
382 #define ahash_data(n, i, dsize)	\
383 	((struct mtype_elem *)((n)->value + ((i) * (dsize))))
384 
385 static void
386 mtype_ext_cleanup(struct ip_set *set, struct hbucket *n)
387 {
388 	int i;
389 	u8 pos = smp_load_acquire(&n->pos);
390 
391 	for (i = 0; i < pos; i++)
392 		if (test_bit(i, n->used))
393 			ip_set_ext_destroy(set, ahash_data(n, i, set->dsize));
394 }
395 
396 /* Flush a hash type of set: destroy all elements */
397 static void
398 mtype_flush(struct ip_set *set)
399 {
400 	struct htype *h = set->data;
401 	struct htable *t;
402 	struct hbucket *n;
403 	u32 r, i;
404 
405 	t = ipset_dereference_nfnl(h->table);
406 	for (r = 0; r < ahash_numof_locks(t->htable_bits); r++) {
407 		spin_lock_bh(&t->hregion[r].lock);
408 		for (i = ahash_bucket_start(r, t->htable_bits);
409 		     i < ahash_bucket_end(r, t->htable_bits); i++) {
410 			n = __ipset_dereference(hbucket(t, i));
411 			if (!n)
412 				continue;
413 			if (set->extensions & IPSET_EXT_DESTROY)
414 				mtype_ext_cleanup(set, n);
415 			/* FIXME: use slab cache */
416 			rcu_assign_pointer(hbucket(t, i), NULL);
417 			kfree_rcu(n, rcu);
418 		}
419 		t->hregion[r].ext_size = 0;
420 		t->hregion[r].elements = 0;
421 		spin_unlock_bh(&t->hregion[r].lock);
422 	}
423 #ifdef IP_SET_HASH_WITH_NETS
424 	memset(h->nets, 0, sizeof(h->nets));
425 #endif
426 }
427 
428 /* Destroy the hashtable part of the set */
429 static void
430 mtype_ahash_destroy(struct ip_set *set, struct htable *t, bool ext_destroy)
431 {
432 	struct hbucket *n;
433 	u32 i;
434 
435 	for (i = 0; i < jhash_size(t->htable_bits); i++) {
436 		n = (__force struct hbucket *)hbucket(t, i);
437 		if (!n)
438 			continue;
439 		if (set->extensions & IPSET_EXT_DESTROY && ext_destroy)
440 			mtype_ext_cleanup(set, n);
441 		/* FIXME: use slab cache */
442 		kfree(n);
443 	}
444 
445 	ip_set_free(t->hregion);
446 	ip_set_free(t);
447 }
448 
449 /* Destroy a hash type of set */
450 static void
451 mtype_destroy(struct ip_set *set)
452 {
453 	struct htype *h = set->data;
454 	struct list_head *l, *lt;
455 
456 	mtype_ahash_destroy(set, (__force struct htable *)h->table, true);
457 	list_for_each_safe(l, lt, &h->ad) {
458 		list_del(l);
459 		kfree(l);
460 	}
461 	kfree(h);
462 
463 	set->data = NULL;
464 }
465 
466 static bool
467 mtype_same_set(const struct ip_set *a, const struct ip_set *b)
468 {
469 	const struct htype *x = a->data;
470 	const struct htype *y = b->data;
471 
472 	/* Resizing changes htable_bits, so we ignore it */
473 	return x->maxelem == y->maxelem &&
474 	       a->timeout == b->timeout &&
475 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
476 	       nf_inet_addr_cmp(&x->bitmask, &y->bitmask) &&
477 #endif
478 #ifdef IP_SET_HASH_WITH_MARKMASK
479 	       x->markmask == y->markmask &&
480 #endif
481 	       a->extensions == b->extensions;
482 }
483 
484 static void
485 mtype_gc_do(struct ip_set *set, struct htype *h, struct htable *t, u32 r)
486 {
487 	struct hbucket *n, *tmp;
488 	struct mtype_elem *data;
489 	u32 i, j, d;
490 	size_t dsize = set->dsize;
491 #ifdef IP_SET_HASH_WITH_NETS
492 	u8 k;
493 #endif
494 	u8 pos, htable_bits = t->htable_bits;
495 
496 	spin_lock_bh(&t->hregion[r].lock);
497 	for (i = ahash_bucket_start(r, htable_bits);
498 	     i < ahash_bucket_end(r, htable_bits); i++) {
499 		n = __ipset_dereference(hbucket(t, i));
500 		if (!n)
501 			continue;
502 		pos = smp_load_acquire(&n->pos);
503 		for (j = 0, d = 0; j < pos; j++) {
504 			if (!test_bit(j, n->used)) {
505 				d++;
506 				continue;
507 			}
508 			data = ahash_data(n, j, dsize);
509 			if (!ip_set_timeout_expired(ext_timeout(data, set)))
510 				continue;
511 			pr_debug("expired %u/%u\n", i, j);
512 			clear_bit(j, n->used);
513 			smp_mb__after_atomic();
514 #ifdef IP_SET_HASH_WITH_NETS
515 			for (k = 0; k < IPSET_NET_COUNT; k++)
516 				mtype_del_cidr(set, h,
517 					NCIDR_PUT(DCIDR_GET(data->cidr, k)),
518 					k);
519 #endif
520 			t->hregion[r].elements--;
521 			ip_set_ext_destroy(set, data);
522 			d++;
523 		}
524 		if (d >= AHASH_INIT_SIZE) {
525 			if (d >= n->size) {
526 				t->hregion[r].ext_size -=
527 					ext_size(n->size, dsize);
528 				rcu_assign_pointer(hbucket(t, i), NULL);
529 				kfree_rcu(n, rcu);
530 				continue;
531 			}
532 			tmp = kzalloc(sizeof(*tmp) +
533 				(n->size - AHASH_INIT_SIZE) * dsize,
534 				GFP_ATOMIC);
535 			if (!tmp)
536 				/* Still try to delete expired elements. */
537 				continue;
538 			tmp->size = n->size - AHASH_INIT_SIZE;
539 			for (j = 0, d = 0; j < pos; j++) {
540 				if (!test_bit(j, n->used))
541 					continue;
542 				data = ahash_data(n, j, dsize);
543 				memcpy(tmp->value + d * dsize,
544 				       data, dsize);
545 				set_bit(d, tmp->used);
546 				d++;
547 			}
548 			tmp->pos = d;
549 			t->hregion[r].ext_size -=
550 				ext_size(AHASH_INIT_SIZE, dsize);
551 			rcu_assign_pointer(hbucket(t, i), tmp);
552 			kfree_rcu(n, rcu);
553 		}
554 	}
555 	spin_unlock_bh(&t->hregion[r].lock);
556 }
557 
558 static void
559 mtype_gc(struct work_struct *work)
560 {
561 	struct htable_gc *gc;
562 	struct ip_set *set;
563 	struct htype *h;
564 	struct htable *t;
565 	u32 r, numof_locks;
566 	unsigned int next_run;
567 
568 	gc = container_of(work, struct htable_gc, dwork.work);
569 	set = gc->set;
570 	h = set->data;
571 
572 	spin_lock_bh(&set->lock);
573 	t = ipset_dereference_set(h->table, set);
574 	atomic_inc(&t->uref);
575 	numof_locks = ahash_numof_locks(t->htable_bits);
576 	r = gc->region++;
577 	if (r >= numof_locks) {
578 		r = gc->region = 0;
579 	}
580 	next_run = (IPSET_GC_PERIOD(set->timeout) * HZ) / numof_locks;
581 	if (next_run < HZ/10)
582 		next_run = HZ/10;
583 	spin_unlock_bh(&set->lock);
584 
585 	mtype_gc_do(set, h, t, r);
586 
587 	if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
588 		pr_debug("Table destroy after resize by expire: %p\n", t);
589 		mtype_ahash_destroy(set, t, false);
590 	}
591 
592 	queue_delayed_work(system_power_efficient_wq, &gc->dwork, next_run);
593 
594 }
595 
596 static void
597 mtype_gc_init(struct htable_gc *gc)
598 {
599 	INIT_DEFERRABLE_WORK(&gc->dwork, mtype_gc);
600 	queue_delayed_work(system_power_efficient_wq, &gc->dwork, HZ);
601 }
602 
603 static void
604 mtype_cancel_gc(struct ip_set *set)
605 {
606 	struct htype *h = set->data;
607 
608 	if (SET_WITH_TIMEOUT(set))
609 		disable_delayed_work_sync(&h->gc.dwork);
610 }
611 
612 static int
613 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
614 	  struct ip_set_ext *mext, u32 flags);
615 static int
616 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
617 	  struct ip_set_ext *mext, u32 flags);
618 
619 /* Resize a hash: create a new hash table with doubling the hashsize
620  * and inserting the elements to it. Repeat until we succeed or
621  * fail due to memory pressures.
622  */
623 static int
624 mtype_resize(struct ip_set *set, bool retried)
625 {
626 	struct htype *h = set->data;
627 	struct htable *t, *orig;
628 	u8 pos, htable_bits;
629 	size_t hsize, dsize = set->dsize;
630 #ifdef IP_SET_HASH_WITH_NETS
631 	u8 flags;
632 	struct mtype_elem *tmp;
633 #endif
634 	struct mtype_elem *data;
635 	struct mtype_elem *d;
636 	struct hbucket *n, *m;
637 	struct list_head *l, *lt;
638 	struct mtype_resize_ad *x;
639 	u32 i, j, r, nr, key;
640 	int ret;
641 
642 #ifdef IP_SET_HASH_WITH_NETS
643 	tmp = kmalloc(dsize, GFP_KERNEL);
644 	if (!tmp)
645 		return -ENOMEM;
646 #endif
647 	orig = ipset_dereference_bh_nfnl(h->table);
648 	htable_bits = orig->htable_bits;
649 
650 retry:
651 	ret = 0;
652 	htable_bits++;
653 	if (!htable_bits)
654 		goto hbwarn;
655 	hsize = htable_size(htable_bits);
656 	if (!hsize)
657 		goto hbwarn;
658 	t = ip_set_alloc(hsize);
659 	if (!t) {
660 		ret = -ENOMEM;
661 		goto out;
662 	}
663 	t->hregion = ip_set_alloc(ahash_sizeof_regions(htable_bits));
664 	if (!t->hregion) {
665 		ip_set_free(t);
666 		ret = -ENOMEM;
667 		goto out;
668 	}
669 	t->htable_bits = htable_bits;
670 	t->maxelem = h->maxelem / ahash_numof_locks(htable_bits);
671 	for (i = 0; i < ahash_numof_locks(htable_bits); i++)
672 		spin_lock_init(&t->hregion[i].lock);
673 
674 	/* There can't be another parallel resizing,
675 	 * but dumping, gc, kernel side add/del are possible
676 	 */
677 	orig = ipset_dereference_bh_nfnl(h->table);
678 	atomic_set(&orig->ref, 1);
679 	atomic_inc(&orig->uref);
680 	pr_debug("attempt to resize set %s from %u to %u, t %p\n",
681 		 set->name, orig->htable_bits, htable_bits, orig);
682 	for (r = 0; r < ahash_numof_locks(orig->htable_bits); r++) {
683 		/* Expire may replace a hbucket with another one */
684 		rcu_read_lock_bh();
685 		for (i = ahash_bucket_start(r, orig->htable_bits);
686 		     i < ahash_bucket_end(r, orig->htable_bits); i++) {
687 			n = __ipset_dereference(hbucket(orig, i));
688 			if (!n)
689 				continue;
690 			pos = smp_load_acquire(&n->pos);
691 			for (j = 0; j < pos; j++) {
692 				if (!test_bit_acquire(j, n->used))
693 					continue;
694 				data = ahash_data(n, j, dsize);
695 				if (SET_ELEM_EXPIRED(set, data))
696 					continue;
697 #ifdef IP_SET_HASH_WITH_NETS
698 				/* We have readers running parallel with us,
699 				 * so the live data cannot be modified.
700 				 */
701 				flags = 0;
702 				memcpy(tmp, data, dsize);
703 				data = tmp;
704 				mtype_data_reset_flags(data, &flags);
705 #endif
706 				key = HKEY(data, h->initval, htable_bits);
707 				m = __ipset_dereference(hbucket(t, key));
708 				nr = ahash_region(key);
709 				if (!m) {
710 					m = kzalloc(sizeof(*m) +
711 					    AHASH_INIT_SIZE * dsize,
712 					    GFP_ATOMIC);
713 					if (!m) {
714 						ret = -ENOMEM;
715 						goto cleanup;
716 					}
717 					m->size = AHASH_INIT_SIZE;
718 					t->hregion[nr].ext_size +=
719 						ext_size(AHASH_INIT_SIZE,
720 							 dsize);
721 					RCU_INIT_POINTER(hbucket(t, key), m);
722 				} else if (m->pos >= m->size) {
723 					struct hbucket *ht;
724 
725 					if (m->size >= AHASH_MAX(h)) {
726 						ret = -EAGAIN;
727 					} else {
728 						ht = kzalloc(sizeof(*ht) +
729 						(m->size + AHASH_INIT_SIZE)
730 						* dsize,
731 						GFP_ATOMIC);
732 						if (!ht)
733 							ret = -ENOMEM;
734 					}
735 					if (ret < 0)
736 						goto cleanup;
737 					memcpy(ht, m, sizeof(struct hbucket) +
738 					       m->size * dsize);
739 					ht->size = m->size + AHASH_INIT_SIZE;
740 					t->hregion[nr].ext_size +=
741 						ext_size(AHASH_INIT_SIZE,
742 							 dsize);
743 					kfree(m);
744 					m = ht;
745 					RCU_INIT_POINTER(hbucket(t, key), ht);
746 				}
747 				d = ahash_data(m, m->pos, dsize);
748 				memcpy(d, data, dsize);
749 				set_bit(m->pos++, m->used);
750 				t->hregion[nr].elements++;
751 #ifdef IP_SET_HASH_WITH_NETS
752 				mtype_data_reset_flags(d, &flags);
753 #endif
754 			}
755 		}
756 		rcu_read_unlock_bh();
757 	}
758 
759 	/* There can't be any other writer. */
760 	rcu_assign_pointer(h->table, t);
761 
762 	/* Give time to other readers of the set */
763 	synchronize_rcu();
764 
765 	pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
766 		 orig->htable_bits, orig, t->htable_bits, t);
767 	/* Add/delete elements processed by the SET target during resize.
768 	 * Kernel-side add cannot trigger a resize and userspace actions
769 	 * are serialized by the mutex.
770 	 */
771 	list_for_each_safe(l, lt, &h->ad) {
772 		x = list_entry(l, struct mtype_resize_ad, list);
773 		if (x->ad == IPSET_ADD) {
774 			mtype_add(set, &x->d, &x->ext, &x->mext, x->flags);
775 		} else {
776 			mtype_del(set, &x->d, NULL, NULL, 0);
777 		}
778 		list_del(l);
779 		kfree(l);
780 	}
781 	/* If there's nobody else using the table, destroy it */
782 	if (atomic_dec_and_test(&orig->uref)) {
783 		pr_debug("Table destroy by resize %p\n", orig);
784 		mtype_ahash_destroy(set, orig, false);
785 	}
786 
787 out:
788 #ifdef IP_SET_HASH_WITH_NETS
789 	kfree(tmp);
790 #endif
791 	return ret;
792 
793 cleanup:
794 	rcu_read_unlock_bh();
795 	atomic_set(&orig->ref, 0);
796 	atomic_dec(&orig->uref);
797 	mtype_ahash_destroy(set, t, false);
798 	if (ret == -EAGAIN)
799 		goto retry;
800 	goto out;
801 
802 hbwarn:
803 	/* In case we have plenty of memory :-) */
804 	pr_warn("Cannot increase the hashsize of set %s further\n", set->name);
805 	ret = -IPSET_ERR_HASH_FULL;
806 	goto out;
807 }
808 
809 /* Get the current number of elements and ext_size in the set  */
810 static void
811 mtype_ext_size(struct ip_set *set, u32 *elements, size_t *ext_size)
812 {
813 	struct htype *h = set->data;
814 	const struct htable *t;
815 	struct hbucket *n;
816 	struct mtype_elem *data;
817 	u32 i, j, r;
818 	u8 pos;
819 
820 	t = rcu_dereference_bh(h->table);
821 	for (r = 0; r < ahash_numof_locks(t->htable_bits); r++) {
822 		for (i = ahash_bucket_start(r, t->htable_bits);
823 		     i < ahash_bucket_end(r, t->htable_bits); i++) {
824 			n = rcu_dereference_bh(hbucket(t, i));
825 			if (!n)
826 				continue;
827 			pos = smp_load_acquire(&n->pos);
828 			for (j = 0; j < pos; j++) {
829 				if (!test_bit_acquire(j, n->used))
830 					continue;
831 				data = ahash_data(n, j, set->dsize);
832 				if (!SET_ELEM_EXPIRED(set, data))
833 					(*elements)++;
834 			}
835 		}
836 		*ext_size += t->hregion[r].ext_size;
837 	}
838 }
839 
840 /* Add an element to a hash and update the internal counters when succeeded,
841  * otherwise report the proper error code.
842  */
843 static int
844 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
845 	  struct ip_set_ext *mext, u32 flags)
846 {
847 	struct htype *h = set->data;
848 	struct htable *t;
849 	const struct mtype_elem *d = value;
850 	struct mtype_elem *data;
851 	struct hbucket *n, *old = ERR_PTR(-ENOENT);
852 	int i, j = -1, ret;
853 	bool flag_exist = flags & IPSET_FLAG_EXIST;
854 	bool deleted = false, forceadd = false, reuse = false;
855 	u32 r, key, multi = 0, elements, maxelem;
856 	u8 npos = 0;
857 
858 	rcu_read_lock_bh();
859 	t = rcu_dereference_bh(h->table);
860 	key = HKEY(value, h->initval, t->htable_bits);
861 	r = ahash_region(key);
862 	atomic_inc(&t->uref);
863 	elements = t->hregion[r].elements;
864 	maxelem = t->maxelem;
865 	if (elements >= maxelem) {
866 		u32 e;
867 		if (SET_WITH_TIMEOUT(set)) {
868 			rcu_read_unlock_bh();
869 			mtype_gc_do(set, h, t, r);
870 			rcu_read_lock_bh();
871 		}
872 		maxelem = h->maxelem;
873 		elements = 0;
874 		for (e = 0; e < ahash_numof_locks(t->htable_bits); e++)
875 			elements += t->hregion[e].elements;
876 		if (elements >= maxelem && SET_WITH_FORCEADD(set))
877 			forceadd = true;
878 	}
879 	rcu_read_unlock_bh();
880 
881 	spin_lock_bh(&t->hregion[r].lock);
882 	n = rcu_dereference_bh(hbucket(t, key));
883 	if (!n) {
884 		if (forceadd || elements >= maxelem)
885 			goto set_full;
886 		old = NULL;
887 		n = kzalloc(sizeof(*n) + AHASH_INIT_SIZE * set->dsize,
888 			    GFP_ATOMIC);
889 		if (!n) {
890 			ret = -ENOMEM;
891 			goto unlock;
892 		}
893 		n->size = AHASH_INIT_SIZE;
894 		t->hregion[r].ext_size +=
895 			ext_size(AHASH_INIT_SIZE, set->dsize);
896 		goto copy_elem;
897 	}
898 	npos = smp_load_acquire(&n->pos);
899 	for (i = 0; i < npos; i++) {
900 		if (!test_bit(i, n->used)) {
901 			/* Reuse first deleted entry */
902 			if (j == -1) {
903 				deleted = reuse = true;
904 				j = i;
905 			}
906 			continue;
907 		}
908 		data = ahash_data(n, i, set->dsize);
909 		if (mtype_data_equal(data, d, &multi)) {
910 			if (flag_exist || SET_ELEM_EXPIRED(set, data)) {
911 				/* Just the extensions could be overwritten */
912 				j = i;
913 				goto overwrite_extensions;
914 			}
915 			ret = -IPSET_ERR_EXIST;
916 			goto unlock;
917 		}
918 		/* Reuse first timed out entry */
919 		if (SET_ELEM_EXPIRED(set, data) && j == -1) {
920 			j = i;
921 			reuse = true;
922 		}
923 	}
924 	if (reuse || forceadd) {
925 		if (j == -1)
926 			j = 0;
927 		data = ahash_data(n, j, set->dsize);
928 		if (!deleted) {
929 #ifdef IP_SET_HASH_WITH_NETS
930 			for (i = 0; i < IPSET_NET_COUNT; i++)
931 				mtype_del_cidr(set, h,
932 					NCIDR_PUT(DCIDR_GET(data->cidr, i)),
933 					i);
934 #endif
935 			ip_set_ext_destroy(set, data);
936 			t->hregion[r].elements--;
937 		}
938 		goto copy_data;
939 	}
940 	if (elements >= maxelem)
941 		goto set_full;
942 	/* Create a new slot */
943 	if (npos >= n->size) {
944 #ifdef IP_SET_HASH_WITH_MULTI
945 		if (h->bucketsize >= AHASH_MAX_TUNED)
946 			goto set_full;
947 		else if (h->bucketsize <= multi)
948 			h->bucketsize += AHASH_INIT_SIZE;
949 #endif
950 		if (n->size >= AHASH_MAX(h)) {
951 			/* Trigger rehashing */
952 			mtype_data_next(&h->next, d);
953 			ret = -EAGAIN;
954 			goto resize;
955 		}
956 		old = n;
957 		n = kzalloc(sizeof(*n) +
958 			    (old->size + AHASH_INIT_SIZE) * set->dsize,
959 			    GFP_ATOMIC);
960 		if (!n) {
961 			ret = -ENOMEM;
962 			goto unlock;
963 		}
964 		memcpy(n, old, sizeof(struct hbucket) +
965 		       old->size * set->dsize);
966 		n->size = old->size + AHASH_INIT_SIZE;
967 		t->hregion[r].ext_size +=
968 			ext_size(AHASH_INIT_SIZE, set->dsize);
969 	}
970 
971 copy_elem:
972 	j = npos++;
973 	data = ahash_data(n, j, set->dsize);
974 copy_data:
975 	t->hregion[r].elements++;
976 #ifdef IP_SET_HASH_WITH_NETS
977 	for (i = 0; i < IPSET_NET_COUNT; i++)
978 		mtype_add_cidr(set, h, NCIDR_PUT(DCIDR_GET(d->cidr, i)), i);
979 #endif
980 	memcpy(data, d, sizeof(struct mtype_elem));
981 overwrite_extensions:
982 #ifdef IP_SET_HASH_WITH_NETS
983 	mtype_data_set_flags(data, flags);
984 #endif
985 	if (SET_WITH_COUNTER(set))
986 		ip_set_init_counter(ext_counter(data, set), ext);
987 	if (SET_WITH_COMMENT(set))
988 		ip_set_init_comment(set, ext_comment(data, set), ext);
989 	if (SET_WITH_SKBINFO(set))
990 		ip_set_init_skbinfo(ext_skbinfo(data, set), ext);
991 	/* Must come last for the case when timed out entry is reused */
992 	if (SET_WITH_TIMEOUT(set))
993 		ip_set_timeout_set(ext_timeout(data, set), ext->timeout);
994 	smp_mb__before_atomic();
995 	/* Ensure all data writes are visible before updating position */
996 	smp_store_release(&n->pos, npos);
997 	set_bit(j, n->used);
998 	if (old != ERR_PTR(-ENOENT)) {
999 		rcu_assign_pointer(hbucket(t, key), n);
1000 		if (old)
1001 			kfree_rcu(old, rcu);
1002 	}
1003 	ret = 0;
1004 resize:
1005 	spin_unlock_bh(&t->hregion[r].lock);
1006 	if (atomic_read(&t->ref) && ext->target) {
1007 		/* Resize is in process and kernel side add, save values */
1008 		struct mtype_resize_ad *x;
1009 
1010 		x = kzalloc_obj(struct mtype_resize_ad, GFP_ATOMIC);
1011 		if (!x)
1012 			/* Don't bother */
1013 			goto out;
1014 		x->ad = IPSET_ADD;
1015 		memcpy(&x->d, value, sizeof(struct mtype_elem));
1016 		memcpy(&x->ext, ext, sizeof(struct ip_set_ext));
1017 		memcpy(&x->mext, mext, sizeof(struct ip_set_ext));
1018 		x->flags = flags;
1019 		spin_lock_bh(&set->lock);
1020 		list_add_tail(&x->list, &h->ad);
1021 		spin_unlock_bh(&set->lock);
1022 	}
1023 	goto out;
1024 
1025 set_full:
1026 	if (net_ratelimit())
1027 		pr_warn("Set %s is full, maxelem %u reached\n",
1028 			set->name, maxelem);
1029 	ret = -IPSET_ERR_HASH_FULL;
1030 unlock:
1031 	spin_unlock_bh(&t->hregion[r].lock);
1032 out:
1033 	if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1034 		pr_debug("Table destroy after resize by add: %p\n", t);
1035 		mtype_ahash_destroy(set, t, false);
1036 	}
1037 	return ret;
1038 }
1039 
1040 /* Delete an element from the hash and free up space if possible.
1041  */
1042 static int
1043 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1044 	  struct ip_set_ext *mext, u32 flags)
1045 {
1046 	struct htype *h = set->data;
1047 	struct htable *t;
1048 	const struct mtype_elem *d = value;
1049 	struct mtype_elem *data;
1050 	struct hbucket *n;
1051 	struct mtype_resize_ad *x = NULL;
1052 	int i, j, k, r, ret = -IPSET_ERR_EXIST;
1053 	u32 key, multi = 0;
1054 	size_t dsize = set->dsize;
1055 	u8 pos;
1056 
1057 	/* Userspace add and resize is excluded by the mutex.
1058 	 * Kernespace add does not trigger resize.
1059 	 */
1060 	rcu_read_lock_bh();
1061 	t = rcu_dereference_bh(h->table);
1062 	key = HKEY(value, h->initval, t->htable_bits);
1063 	r = ahash_region(key);
1064 	atomic_inc(&t->uref);
1065 	rcu_read_unlock_bh();
1066 
1067 	spin_lock_bh(&t->hregion[r].lock);
1068 	n = rcu_dereference_bh(hbucket(t, key));
1069 	if (!n)
1070 		goto out;
1071 	pos = smp_load_acquire(&n->pos);
1072 	for (i = 0, k = 0; i < pos; i++) {
1073 		if (!test_bit(i, n->used)) {
1074 			k++;
1075 			continue;
1076 		}
1077 		data = ahash_data(n, i, dsize);
1078 		if (!mtype_data_equal(data, d, &multi))
1079 			continue;
1080 		if (SET_ELEM_EXPIRED(set, data))
1081 			goto out;
1082 
1083 		ret = 0;
1084 		clear_bit(i, n->used);
1085 		smp_mb__after_atomic();
1086 		if (i + 1 == pos)
1087 			smp_store_release(&n->pos, --pos);
1088 		t->hregion[r].elements--;
1089 #ifdef IP_SET_HASH_WITH_NETS
1090 		for (j = 0; j < IPSET_NET_COUNT; j++)
1091 			mtype_del_cidr(set, h,
1092 				       NCIDR_PUT(DCIDR_GET(d->cidr, j)), j);
1093 #endif
1094 		ip_set_ext_destroy(set, data);
1095 
1096 		if (atomic_read(&t->ref) && ext->target) {
1097 			/* Resize is in process and kernel side del,
1098 			 * save values
1099 			 */
1100 			x = kzalloc_obj(struct mtype_resize_ad, GFP_ATOMIC);
1101 			if (x) {
1102 				x->ad = IPSET_DEL;
1103 				memcpy(&x->d, value,
1104 				       sizeof(struct mtype_elem));
1105 				x->flags = flags;
1106 			}
1107 		}
1108 		for (; i < pos; i++) {
1109 			if (!test_bit(i, n->used))
1110 				k++;
1111 		}
1112 		if (k == pos) {
1113 			t->hregion[r].ext_size -= ext_size(n->size, dsize);
1114 			rcu_assign_pointer(hbucket(t, key), NULL);
1115 			kfree_rcu(n, rcu);
1116 		} else if (k >= AHASH_INIT_SIZE) {
1117 			struct hbucket *tmp = kzalloc(sizeof(*tmp) +
1118 					(n->size - AHASH_INIT_SIZE) * dsize,
1119 					GFP_ATOMIC);
1120 			if (!tmp)
1121 				goto out;
1122 			tmp->size = n->size - AHASH_INIT_SIZE;
1123 			for (j = 0, k = 0; j < pos; j++) {
1124 				if (!test_bit(j, n->used))
1125 					continue;
1126 				data = ahash_data(n, j, dsize);
1127 				memcpy(tmp->value + k * dsize, data, dsize);
1128 				set_bit(k, tmp->used);
1129 				k++;
1130 			}
1131 			tmp->pos = k;
1132 			t->hregion[r].ext_size -=
1133 				ext_size(AHASH_INIT_SIZE, dsize);
1134 			rcu_assign_pointer(hbucket(t, key), tmp);
1135 			kfree_rcu(n, rcu);
1136 		}
1137 		goto out;
1138 	}
1139 
1140 out:
1141 	spin_unlock_bh(&t->hregion[r].lock);
1142 	if (x) {
1143 		spin_lock_bh(&set->lock);
1144 		list_add(&x->list, &h->ad);
1145 		spin_unlock_bh(&set->lock);
1146 	}
1147 	if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1148 		pr_debug("Table destroy after resize by del: %p\n", t);
1149 		mtype_ahash_destroy(set, t, false);
1150 	}
1151 	return ret;
1152 }
1153 
1154 static int
1155 mtype_data_match(struct mtype_elem *data, const struct ip_set_ext *ext,
1156 		 struct ip_set_ext *mext, struct ip_set *set, u32 flags)
1157 {
1158 	if (!ip_set_match_extensions(set, ext, mext, flags, data))
1159 		return 0;
1160 	/* nomatch entries return -ENOTEMPTY */
1161 	return mtype_do_data_match(data);
1162 }
1163 
1164 #ifdef IP_SET_HASH_WITH_NETS
1165 /* Special test function which takes into account the different network
1166  * sizes added to the set
1167  */
1168 static int
1169 mtype_test_cidrs(struct ip_set *set, struct mtype_elem *d,
1170 		 const struct ip_set_ext *ext,
1171 		 struct ip_set_ext *mext, u32 flags)
1172 {
1173 	struct htype *h = set->data;
1174 	struct htable *t = rcu_dereference_bh(h->table);
1175 	struct hbucket *n;
1176 	struct mtype_elem *data;
1177 #if IPSET_NET_COUNT == 2
1178 	struct mtype_elem orig = *d;
1179 	int ret, i, j = 0, k;
1180 #else
1181 	int ret, i, j = 0;
1182 #endif
1183 	u32 key, multi = 0;
1184 	u8 pos;
1185 
1186 	pr_debug("test by nets\n");
1187 	for (; j < NLEN && h->nets[j].cidr[0] && !multi; j++) {
1188 #if IPSET_NET_COUNT == 2
1189 		mtype_data_reset_elem(d, &orig);
1190 		mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]), false);
1191 		for (k = 0; k < NLEN && h->nets[k].cidr[1] && !multi;
1192 		     k++) {
1193 			mtype_data_netmask(d, NCIDR_GET(h->nets[k].cidr[1]),
1194 					   true);
1195 #else
1196 		mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]));
1197 #endif
1198 		key = HKEY(d, h->initval, t->htable_bits);
1199 		n = rcu_dereference_bh(hbucket(t, key));
1200 		if (!n)
1201 			continue;
1202 		pos = smp_load_acquire(&n->pos);
1203 		for (i = 0; i < pos; i++) {
1204 			if (!test_bit_acquire(i, n->used))
1205 				continue;
1206 			data = ahash_data(n, i, set->dsize);
1207 			if (!mtype_data_equal(data, d, &multi))
1208 				continue;
1209 			ret = mtype_data_match(data, ext, mext, set, flags);
1210 			if (ret != 0)
1211 				return ret;
1212 #ifdef IP_SET_HASH_WITH_MULTI
1213 			/* No match, reset multiple match flag */
1214 			multi = 0;
1215 #endif
1216 		}
1217 #if IPSET_NET_COUNT == 2
1218 		}
1219 #endif
1220 	}
1221 	return 0;
1222 }
1223 #endif
1224 
1225 /* Test whether the element is added to the set */
1226 static int
1227 mtype_test(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1228 	   struct ip_set_ext *mext, u32 flags)
1229 {
1230 	struct htype *h = set->data;
1231 	struct htable *t;
1232 	struct mtype_elem *d = value;
1233 	struct hbucket *n;
1234 	struct mtype_elem *data;
1235 	int i, ret = 0;
1236 	u32 key, multi = 0;
1237 	u8 pos;
1238 
1239 	rcu_read_lock_bh();
1240 	t = rcu_dereference_bh(h->table);
1241 #ifdef IP_SET_HASH_WITH_NETS
1242 	/* If we test an IP address and not a network address,
1243 	 * try all possible network sizes
1244 	 */
1245 	for (i = 0; i < IPSET_NET_COUNT; i++)
1246 		if (DCIDR_GET(d->cidr, i) != HOST_MASK)
1247 			break;
1248 	if (i == IPSET_NET_COUNT) {
1249 		ret = mtype_test_cidrs(set, d, ext, mext, flags);
1250 		goto out;
1251 	}
1252 #endif
1253 
1254 	key = HKEY(d, h->initval, t->htable_bits);
1255 	n = rcu_dereference_bh(hbucket(t, key));
1256 	if (!n) {
1257 		ret = 0;
1258 		goto out;
1259 	}
1260 	pos = smp_load_acquire(&n->pos);
1261 	for (i = 0; i < pos; i++) {
1262 		if (!test_bit_acquire(i, n->used))
1263 			continue;
1264 		data = ahash_data(n, i, set->dsize);
1265 		if (!mtype_data_equal(data, d, &multi))
1266 			continue;
1267 		ret = mtype_data_match(data, ext, mext, set, flags);
1268 		if (ret != 0)
1269 			goto out;
1270 	}
1271 out:
1272 	rcu_read_unlock_bh();
1273 	return ret;
1274 }
1275 
1276 /* Reply a HEADER request: fill out the header part of the set */
1277 static int
1278 mtype_head(struct ip_set *set, struct sk_buff *skb)
1279 {
1280 	struct htype *h = set->data;
1281 	const struct htable *t;
1282 	struct nlattr *nested;
1283 	size_t memsize;
1284 	u32 elements = 0;
1285 	size_t ext_size = 0;
1286 	u8 htable_bits;
1287 
1288 	rcu_read_lock_bh();
1289 	t = rcu_dereference_bh(h->table);
1290 	mtype_ext_size(set, &elements, &ext_size);
1291 	memsize = mtype_ahash_memsize(h, t) + ext_size + set->ext_size;
1292 	htable_bits = t->htable_bits;
1293 	rcu_read_unlock_bh();
1294 
1295 	nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1296 	if (!nested)
1297 		goto nla_put_failure;
1298 	if (nla_put_net32(skb, IPSET_ATTR_HASHSIZE,
1299 			  htonl(jhash_size(htable_bits))) ||
1300 	    nla_put_net32(skb, IPSET_ATTR_MAXELEM, htonl(h->maxelem)))
1301 		goto nla_put_failure;
1302 #ifdef IP_SET_HASH_WITH_BITMASK
1303 	/* if netmask is set to anything other than HOST_MASK we know that the user supplied netmask
1304 	 * and not bitmask. These two are mutually exclusive. */
1305 	if (h->netmask == HOST_MASK && !nf_inet_addr_cmp(&onesmask, &h->bitmask)) {
1306 		if (set->family == NFPROTO_IPV4) {
1307 			if (nla_put_ipaddr4(skb, IPSET_ATTR_BITMASK, h->bitmask.ip))
1308 				goto nla_put_failure;
1309 		} else if (set->family == NFPROTO_IPV6) {
1310 			if (nla_put_ipaddr6(skb, IPSET_ATTR_BITMASK, &h->bitmask.in6))
1311 				goto nla_put_failure;
1312 		}
1313 	}
1314 #endif
1315 #ifdef IP_SET_HASH_WITH_NETMASK
1316 	if (h->netmask != HOST_MASK && nla_put_u8(skb, IPSET_ATTR_NETMASK, h->netmask))
1317 		goto nla_put_failure;
1318 #endif
1319 #ifdef IP_SET_HASH_WITH_MARKMASK
1320 	if (nla_put_u32(skb, IPSET_ATTR_MARKMASK, h->markmask))
1321 		goto nla_put_failure;
1322 #endif
1323 	if (set->flags & IPSET_CREATE_FLAG_BUCKETSIZE) {
1324 		if (nla_put_u8(skb, IPSET_ATTR_BUCKETSIZE, h->bucketsize) ||
1325 		    nla_put_net32(skb, IPSET_ATTR_INITVAL, htonl(h->initval)))
1326 			goto nla_put_failure;
1327 	}
1328 	if (nla_put_net32(skb, IPSET_ATTR_REFERENCES, htonl(set->ref)) ||
1329 	    nla_put_net32(skb, IPSET_ATTR_MEMSIZE, htonl(memsize)) ||
1330 	    nla_put_net32(skb, IPSET_ATTR_ELEMENTS, htonl(elements)))
1331 		goto nla_put_failure;
1332 	if (unlikely(ip_set_put_flags(skb, set)))
1333 		goto nla_put_failure;
1334 	nla_nest_end(skb, nested);
1335 
1336 	return 0;
1337 nla_put_failure:
1338 	return -EMSGSIZE;
1339 }
1340 
1341 /* Make possible to run dumping parallel with resizing */
1342 static void
1343 mtype_uref(struct ip_set *set, struct netlink_callback *cb, bool start)
1344 {
1345 	struct htype *h = set->data;
1346 	struct htable *t;
1347 
1348 	if (start) {
1349 		rcu_read_lock_bh();
1350 		t = ipset_dereference_bh_nfnl(h->table);
1351 		atomic_inc(&t->uref);
1352 		cb->args[IPSET_CB_PRIVATE] = (unsigned long)t;
1353 		rcu_read_unlock_bh();
1354 	} else if (cb->args[IPSET_CB_PRIVATE]) {
1355 		t = (struct htable *)cb->args[IPSET_CB_PRIVATE];
1356 		if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1357 			pr_debug("Table destroy after resize "
1358 				 " by dump: %p\n", t);
1359 			mtype_ahash_destroy(set, t, false);
1360 		}
1361 		cb->args[IPSET_CB_PRIVATE] = 0;
1362 	}
1363 }
1364 
1365 /* Reply a LIST/SAVE request: dump the elements of the specified set */
1366 static int
1367 mtype_list(const struct ip_set *set,
1368 	   struct sk_buff *skb, struct netlink_callback *cb)
1369 {
1370 	const struct htable *t;
1371 	struct nlattr *atd, *nested;
1372 	const struct hbucket *n;
1373 	const struct mtype_elem *e;
1374 	u32 first = cb->args[IPSET_CB_ARG0];
1375 	/* We assume that one hash bucket fills into one page */
1376 	void *incomplete;
1377 	int i, ret = 0;
1378 	u8 pos;
1379 
1380 	atd = nla_nest_start(skb, IPSET_ATTR_ADT);
1381 	if (!atd)
1382 		return -EMSGSIZE;
1383 
1384 	pr_debug("list hash set %s\n", set->name);
1385 	t = (const struct htable *)cb->args[IPSET_CB_PRIVATE];
1386 	/* Expire may replace a hbucket with another one */
1387 	rcu_read_lock();
1388 	for (; cb->args[IPSET_CB_ARG0] < jhash_size(t->htable_bits);
1389 	     cb->args[IPSET_CB_ARG0]++) {
1390 		cond_resched_rcu();
1391 		incomplete = skb_tail_pointer(skb);
1392 		n = rcu_dereference(hbucket(t, cb->args[IPSET_CB_ARG0]));
1393 		pr_debug("cb->arg bucket: %lu, t %p n %p\n",
1394 			 cb->args[IPSET_CB_ARG0], t, n);
1395 		if (!n)
1396 			continue;
1397 		pos = smp_load_acquire(&n->pos);
1398 		for (i = 0; i < pos; i++) {
1399 			if (!test_bit_acquire(i, n->used))
1400 				continue;
1401 			e = ahash_data(n, i, set->dsize);
1402 			if (SET_ELEM_EXPIRED(set, e))
1403 				continue;
1404 			pr_debug("list hash %lu hbucket %p i %u, data %p\n",
1405 				 cb->args[IPSET_CB_ARG0], n, i, e);
1406 			nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1407 			if (!nested) {
1408 				if (cb->args[IPSET_CB_ARG0] == first) {
1409 					nla_nest_cancel(skb, atd);
1410 					ret = -EMSGSIZE;
1411 					goto out;
1412 				}
1413 				goto nla_put_failure;
1414 			}
1415 			if (mtype_data_list(skb, e))
1416 				goto nla_put_failure;
1417 			if (ip_set_put_extensions(skb, set, e, true))
1418 				goto nla_put_failure;
1419 			nla_nest_end(skb, nested);
1420 		}
1421 	}
1422 	nla_nest_end(skb, atd);
1423 	/* Set listing finished */
1424 	cb->args[IPSET_CB_ARG0] = 0;
1425 
1426 	goto out;
1427 
1428 nla_put_failure:
1429 	nlmsg_trim(skb, incomplete);
1430 	if (unlikely(first == cb->args[IPSET_CB_ARG0])) {
1431 		pr_warn("Can't list set %s: one bucket does not fit into a message. Please report it!\n",
1432 			set->name);
1433 		cb->args[IPSET_CB_ARG0] = 0;
1434 		ret = -EMSGSIZE;
1435 	} else {
1436 		nla_nest_end(skb, atd);
1437 	}
1438 out:
1439 	rcu_read_unlock();
1440 	return ret;
1441 }
1442 
1443 static int
1444 IPSET_TOKEN(MTYPE, _kadt)(struct ip_set *set, const struct sk_buff *skb,
1445 			  const struct xt_action_param *par,
1446 			  enum ipset_adt adt, struct ip_set_adt_opt *opt);
1447 
1448 static int
1449 IPSET_TOKEN(MTYPE, _uadt)(struct ip_set *set, struct nlattr *tb[],
1450 			  enum ipset_adt adt, u32 *lineno, u32 flags,
1451 			  bool retried);
1452 
1453 static const struct ip_set_type_variant mtype_variant = {
1454 	.kadt	= mtype_kadt,
1455 	.uadt	= mtype_uadt,
1456 	.adt	= {
1457 		[IPSET_ADD] = mtype_add,
1458 		[IPSET_DEL] = mtype_del,
1459 		[IPSET_TEST] = mtype_test,
1460 	},
1461 	.destroy = mtype_destroy,
1462 	.flush	= mtype_flush,
1463 	.head	= mtype_head,
1464 	.list	= mtype_list,
1465 	.uref	= mtype_uref,
1466 	.resize	= mtype_resize,
1467 	.same_set = mtype_same_set,
1468 	.cancel_gc = mtype_cancel_gc,
1469 	.region_lock = true,
1470 };
1471 
1472 #ifdef IP_SET_EMIT_CREATE
1473 static int
1474 IPSET_TOKEN(HTYPE, _create)(struct net *net, struct ip_set *set,
1475 			    struct nlattr *tb[], u32 flags)
1476 {
1477 	u32 hashsize = IPSET_DEFAULT_HASHSIZE, maxelem = IPSET_DEFAULT_MAXELEM;
1478 #ifdef IP_SET_HASH_WITH_MARKMASK
1479 	u32 markmask;
1480 #endif
1481 	u8 hbits;
1482 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
1483 	int ret __attribute__((unused)) = 0;
1484 	u8 netmask = set->family == NFPROTO_IPV4 ? 32 : 128;
1485 	union nf_inet_addr bitmask = onesmask;
1486 #endif
1487 	size_t hsize;
1488 	struct htype *h;
1489 	struct htable *t;
1490 	u32 i;
1491 
1492 	pr_debug("Create set %s with family %s\n",
1493 		 set->name, set->family == NFPROTO_IPV4 ? "inet" : "inet6");
1494 
1495 #ifdef IP_SET_PROTO_UNDEF
1496 	if (set->family != NFPROTO_UNSPEC)
1497 		return -IPSET_ERR_INVALID_FAMILY;
1498 #else
1499 	if (!(set->family == NFPROTO_IPV4 || set->family == NFPROTO_IPV6))
1500 		return -IPSET_ERR_INVALID_FAMILY;
1501 #endif
1502 
1503 	if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_HASHSIZE) ||
1504 		     !ip_set_optattr_netorder(tb, IPSET_ATTR_MAXELEM) ||
1505 		     !ip_set_optattr_netorder(tb, IPSET_ATTR_TIMEOUT) ||
1506 		     !ip_set_optattr_netorder(tb, IPSET_ATTR_CADT_FLAGS)))
1507 		return -IPSET_ERR_PROTOCOL;
1508 
1509 #ifdef IP_SET_HASH_WITH_MARKMASK
1510 	/* Separated condition in order to avoid directive in argument list */
1511 	if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_MARKMASK)))
1512 		return -IPSET_ERR_PROTOCOL;
1513 
1514 	markmask = 0xffffffff;
1515 	if (tb[IPSET_ATTR_MARKMASK]) {
1516 		markmask = ntohl(nla_get_be32(tb[IPSET_ATTR_MARKMASK]));
1517 		if (markmask == 0)
1518 			return -IPSET_ERR_INVALID_MARKMASK;
1519 	}
1520 #endif
1521 
1522 #ifdef IP_SET_HASH_WITH_NETMASK
1523 	if (tb[IPSET_ATTR_NETMASK]) {
1524 		netmask = nla_get_u8(tb[IPSET_ATTR_NETMASK]);
1525 
1526 		if ((set->family == NFPROTO_IPV4 && netmask > 32) ||
1527 		    (set->family == NFPROTO_IPV6 && netmask > 128) ||
1528 		    netmask == 0)
1529 			return -IPSET_ERR_INVALID_NETMASK;
1530 
1531 		/* we convert netmask to bitmask and store it */
1532 		if (set->family == NFPROTO_IPV4)
1533 			bitmask.ip = ip_set_netmask(netmask);
1534 		else
1535 			ip6_netmask(&bitmask, netmask);
1536 	}
1537 #endif
1538 
1539 #ifdef IP_SET_HASH_WITH_BITMASK
1540 	if (tb[IPSET_ATTR_BITMASK]) {
1541 		/* bitmask and netmask do the same thing, allow only one of these options */
1542 		if (tb[IPSET_ATTR_NETMASK])
1543 			return -IPSET_ERR_BITMASK_NETMASK_EXCL;
1544 
1545 		if (set->family == NFPROTO_IPV4) {
1546 			ret = ip_set_get_ipaddr4(tb[IPSET_ATTR_BITMASK], &bitmask.ip);
1547 			if (ret || !bitmask.ip)
1548 				return -IPSET_ERR_INVALID_NETMASK;
1549 		} else if (set->family == NFPROTO_IPV6) {
1550 			ret = ip_set_get_ipaddr6(tb[IPSET_ATTR_BITMASK], &bitmask);
1551 			if (ret || ipv6_addr_any(&bitmask.in6))
1552 				return -IPSET_ERR_INVALID_NETMASK;
1553 		}
1554 
1555 		if (nf_inet_addr_cmp(&bitmask, &zeromask))
1556 			return -IPSET_ERR_INVALID_NETMASK;
1557 	}
1558 #endif
1559 
1560 	if (tb[IPSET_ATTR_HASHSIZE]) {
1561 		hashsize = ip_set_get_h32(tb[IPSET_ATTR_HASHSIZE]);
1562 		if (hashsize < IPSET_MIMINAL_HASHSIZE)
1563 			hashsize = IPSET_MIMINAL_HASHSIZE;
1564 	}
1565 
1566 	if (tb[IPSET_ATTR_MAXELEM])
1567 		maxelem = ip_set_get_h32(tb[IPSET_ATTR_MAXELEM]);
1568 
1569 	hsize = sizeof(*h);
1570 	h = kzalloc(hsize, GFP_KERNEL);
1571 	if (!h)
1572 		return -ENOMEM;
1573 
1574 	/* Compute htable_bits from the user input parameter hashsize.
1575 	 * Assume that hashsize == 2^htable_bits,
1576 	 * otherwise round up to the first 2^n value.
1577 	 */
1578 	hbits = fls(hashsize - 1);
1579 	hsize = htable_size(hbits);
1580 	if (hsize == 0) {
1581 		kfree(h);
1582 		return -ENOMEM;
1583 	}
1584 	t = ip_set_alloc(hsize);
1585 	if (!t) {
1586 		kfree(h);
1587 		return -ENOMEM;
1588 	}
1589 	t->hregion = ip_set_alloc(ahash_sizeof_regions(hbits));
1590 	if (!t->hregion) {
1591 		ip_set_free(t);
1592 		kfree(h);
1593 		return -ENOMEM;
1594 	}
1595 	h->gc.set = set;
1596 	for (i = 0; i < ahash_numof_locks(hbits); i++)
1597 		spin_lock_init(&t->hregion[i].lock);
1598 	h->maxelem = maxelem;
1599 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
1600 	h->bitmask = bitmask;
1601 	h->netmask = netmask;
1602 #endif
1603 #ifdef IP_SET_HASH_WITH_MARKMASK
1604 	h->markmask = markmask;
1605 #endif
1606 	if (tb[IPSET_ATTR_INITVAL])
1607 		h->initval = ntohl(nla_get_be32(tb[IPSET_ATTR_INITVAL]));
1608 	else
1609 		get_random_bytes(&h->initval, sizeof(h->initval));
1610 	h->bucketsize = AHASH_MAX_SIZE;
1611 	if (tb[IPSET_ATTR_BUCKETSIZE]) {
1612 		h->bucketsize = nla_get_u8(tb[IPSET_ATTR_BUCKETSIZE]);
1613 		if (h->bucketsize < AHASH_INIT_SIZE)
1614 			h->bucketsize = AHASH_INIT_SIZE;
1615 		else if (h->bucketsize > AHASH_MAX_SIZE)
1616 			h->bucketsize = AHASH_MAX_SIZE;
1617 		else if (h->bucketsize % 2)
1618 			h->bucketsize += 1;
1619 	}
1620 	t->htable_bits = hbits;
1621 	t->maxelem = h->maxelem / ahash_numof_locks(hbits);
1622 	RCU_INIT_POINTER(h->table, t);
1623 
1624 	INIT_LIST_HEAD(&h->ad);
1625 	set->data = h;
1626 #ifndef IP_SET_PROTO_UNDEF
1627 	if (set->family == NFPROTO_IPV4) {
1628 #endif
1629 		set->variant = &IPSET_TOKEN(HTYPE, 4_variant);
1630 		set->dsize = ip_set_elem_len(set, tb,
1631 			sizeof(struct IPSET_TOKEN(HTYPE, 4_elem)),
1632 			__alignof__(struct IPSET_TOKEN(HTYPE, 4_elem)));
1633 #ifndef IP_SET_PROTO_UNDEF
1634 	} else {
1635 		set->variant = &IPSET_TOKEN(HTYPE, 6_variant);
1636 		set->dsize = ip_set_elem_len(set, tb,
1637 			sizeof(struct IPSET_TOKEN(HTYPE, 6_elem)),
1638 			__alignof__(struct IPSET_TOKEN(HTYPE, 6_elem)));
1639 	}
1640 #endif
1641 	set->timeout = IPSET_NO_TIMEOUT;
1642 	if (tb[IPSET_ATTR_TIMEOUT]) {
1643 		set->timeout = ip_set_timeout_uget(tb[IPSET_ATTR_TIMEOUT]);
1644 #ifndef IP_SET_PROTO_UNDEF
1645 		if (set->family == NFPROTO_IPV4)
1646 #endif
1647 			IPSET_TOKEN(HTYPE, 4_gc_init)(&h->gc);
1648 #ifndef IP_SET_PROTO_UNDEF
1649 		else
1650 			IPSET_TOKEN(HTYPE, 6_gc_init)(&h->gc);
1651 #endif
1652 	}
1653 	pr_debug("create %s hashsize %u (%u) maxelem %u: %p(%p)\n",
1654 		 set->name, jhash_size(t->htable_bits),
1655 		 t->htable_bits, h->maxelem, set->data, t);
1656 
1657 	return 0;
1658 }
1659 #endif /* IP_SET_EMIT_CREATE */
1660 
1661 #undef HKEY_DATALEN
1662