xref: /linux/net/netfilter/ipset/ip_set_hash_gen.h (revision 79790b6818e96c58fe2bffee1b418c16e64e7b80)
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, htable_bits)		\
67 	((n) % ahash_numof_locks(htable_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
htable_size(u8 hbits)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
mtype_add_cidr(struct ip_set * set,struct htype * h,u8 cidr,u8 n)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
mtype_del_cidr(struct ip_set * set,struct htype * h,u8 cidr,u8 n)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
mtype_ahash_memsize(const struct htype * h,const struct htable * 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
mtype_ext_cleanup(struct ip_set * set,struct hbucket * n)386 mtype_ext_cleanup(struct ip_set *set, struct hbucket *n)
387 {
388 	int i;
389 
390 	for (i = 0; i < n->pos; i++)
391 		if (test_bit(i, n->used))
392 			ip_set_ext_destroy(set, ahash_data(n, i, set->dsize));
393 }
394 
395 /* Flush a hash type of set: destroy all elements */
396 static void
mtype_flush(struct ip_set * set)397 mtype_flush(struct ip_set *set)
398 {
399 	struct htype *h = set->data;
400 	struct htable *t;
401 	struct hbucket *n;
402 	u32 r, i;
403 
404 	t = ipset_dereference_nfnl(h->table);
405 	for (r = 0; r < ahash_numof_locks(t->htable_bits); r++) {
406 		spin_lock_bh(&t->hregion[r].lock);
407 		for (i = ahash_bucket_start(r, t->htable_bits);
408 		     i < ahash_bucket_end(r, t->htable_bits); i++) {
409 			n = __ipset_dereference(hbucket(t, i));
410 			if (!n)
411 				continue;
412 			if (set->extensions & IPSET_EXT_DESTROY)
413 				mtype_ext_cleanup(set, n);
414 			/* FIXME: use slab cache */
415 			rcu_assign_pointer(hbucket(t, i), NULL);
416 			kfree_rcu(n, rcu);
417 		}
418 		t->hregion[r].ext_size = 0;
419 		t->hregion[r].elements = 0;
420 		spin_unlock_bh(&t->hregion[r].lock);
421 	}
422 #ifdef IP_SET_HASH_WITH_NETS
423 	memset(h->nets, 0, sizeof(h->nets));
424 #endif
425 }
426 
427 /* Destroy the hashtable part of the set */
428 static void
mtype_ahash_destroy(struct ip_set * set,struct htable * t,bool ext_destroy)429 mtype_ahash_destroy(struct ip_set *set, struct htable *t, bool ext_destroy)
430 {
431 	struct hbucket *n;
432 	u32 i;
433 
434 	for (i = 0; i < jhash_size(t->htable_bits); i++) {
435 		n = (__force struct hbucket *)hbucket(t, i);
436 		if (!n)
437 			continue;
438 		if (set->extensions & IPSET_EXT_DESTROY && ext_destroy)
439 			mtype_ext_cleanup(set, n);
440 		/* FIXME: use slab cache */
441 		kfree(n);
442 	}
443 
444 	ip_set_free(t->hregion);
445 	ip_set_free(t);
446 }
447 
448 /* Destroy a hash type of set */
449 static void
mtype_destroy(struct ip_set * set)450 mtype_destroy(struct ip_set *set)
451 {
452 	struct htype *h = set->data;
453 	struct list_head *l, *lt;
454 
455 	mtype_ahash_destroy(set, (__force struct htable *)h->table, true);
456 	list_for_each_safe(l, lt, &h->ad) {
457 		list_del(l);
458 		kfree(l);
459 	}
460 	kfree(h);
461 
462 	set->data = NULL;
463 }
464 
465 static bool
mtype_same_set(const struct ip_set * a,const struct ip_set * b)466 mtype_same_set(const struct ip_set *a, const struct ip_set *b)
467 {
468 	const struct htype *x = a->data;
469 	const struct htype *y = b->data;
470 
471 	/* Resizing changes htable_bits, so we ignore it */
472 	return x->maxelem == y->maxelem &&
473 	       a->timeout == b->timeout &&
474 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
475 	       nf_inet_addr_cmp(&x->bitmask, &y->bitmask) &&
476 #endif
477 #ifdef IP_SET_HASH_WITH_MARKMASK
478 	       x->markmask == y->markmask &&
479 #endif
480 	       a->extensions == b->extensions;
481 }
482 
483 static void
mtype_gc_do(struct ip_set * set,struct htype * h,struct htable * t,u32 r)484 mtype_gc_do(struct ip_set *set, struct htype *h, struct htable *t, u32 r)
485 {
486 	struct hbucket *n, *tmp;
487 	struct mtype_elem *data;
488 	u32 i, j, d;
489 	size_t dsize = set->dsize;
490 #ifdef IP_SET_HASH_WITH_NETS
491 	u8 k;
492 #endif
493 	u8 htable_bits = t->htable_bits;
494 
495 	spin_lock_bh(&t->hregion[r].lock);
496 	for (i = ahash_bucket_start(r, htable_bits);
497 	     i < ahash_bucket_end(r, htable_bits); i++) {
498 		n = __ipset_dereference(hbucket(t, i));
499 		if (!n)
500 			continue;
501 		for (j = 0, d = 0; j < n->pos; j++) {
502 			if (!test_bit(j, n->used)) {
503 				d++;
504 				continue;
505 			}
506 			data = ahash_data(n, j, dsize);
507 			if (!ip_set_timeout_expired(ext_timeout(data, set)))
508 				continue;
509 			pr_debug("expired %u/%u\n", i, j);
510 			clear_bit(j, n->used);
511 			smp_mb__after_atomic();
512 #ifdef IP_SET_HASH_WITH_NETS
513 			for (k = 0; k < IPSET_NET_COUNT; k++)
514 				mtype_del_cidr(set, h,
515 					NCIDR_PUT(DCIDR_GET(data->cidr, k)),
516 					k);
517 #endif
518 			t->hregion[r].elements--;
519 			ip_set_ext_destroy(set, data);
520 			d++;
521 		}
522 		if (d >= AHASH_INIT_SIZE) {
523 			if (d >= n->size) {
524 				t->hregion[r].ext_size -=
525 					ext_size(n->size, dsize);
526 				rcu_assign_pointer(hbucket(t, i), NULL);
527 				kfree_rcu(n, rcu);
528 				continue;
529 			}
530 			tmp = kzalloc(sizeof(*tmp) +
531 				(n->size - AHASH_INIT_SIZE) * dsize,
532 				GFP_ATOMIC);
533 			if (!tmp)
534 				/* Still try to delete expired elements. */
535 				continue;
536 			tmp->size = n->size - AHASH_INIT_SIZE;
537 			for (j = 0, d = 0; j < n->pos; j++) {
538 				if (!test_bit(j, n->used))
539 					continue;
540 				data = ahash_data(n, j, dsize);
541 				memcpy(tmp->value + d * dsize,
542 				       data, dsize);
543 				set_bit(d, tmp->used);
544 				d++;
545 			}
546 			tmp->pos = d;
547 			t->hregion[r].ext_size -=
548 				ext_size(AHASH_INIT_SIZE, dsize);
549 			rcu_assign_pointer(hbucket(t, i), tmp);
550 			kfree_rcu(n, rcu);
551 		}
552 	}
553 	spin_unlock_bh(&t->hregion[r].lock);
554 }
555 
556 static void
mtype_gc(struct work_struct * work)557 mtype_gc(struct work_struct *work)
558 {
559 	struct htable_gc *gc;
560 	struct ip_set *set;
561 	struct htype *h;
562 	struct htable *t;
563 	u32 r, numof_locks;
564 	unsigned int next_run;
565 
566 	gc = container_of(work, struct htable_gc, dwork.work);
567 	set = gc->set;
568 	h = set->data;
569 
570 	spin_lock_bh(&set->lock);
571 	t = ipset_dereference_set(h->table, set);
572 	atomic_inc(&t->uref);
573 	numof_locks = ahash_numof_locks(t->htable_bits);
574 	r = gc->region++;
575 	if (r >= numof_locks) {
576 		r = gc->region = 0;
577 	}
578 	next_run = (IPSET_GC_PERIOD(set->timeout) * HZ) / numof_locks;
579 	if (next_run < HZ/10)
580 		next_run = HZ/10;
581 	spin_unlock_bh(&set->lock);
582 
583 	mtype_gc_do(set, h, t, r);
584 
585 	if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
586 		pr_debug("Table destroy after resize by expire: %p\n", t);
587 		mtype_ahash_destroy(set, t, false);
588 	}
589 
590 	queue_delayed_work(system_power_efficient_wq, &gc->dwork, next_run);
591 
592 }
593 
594 static void
mtype_gc_init(struct htable_gc * gc)595 mtype_gc_init(struct htable_gc *gc)
596 {
597 	INIT_DEFERRABLE_WORK(&gc->dwork, mtype_gc);
598 	queue_delayed_work(system_power_efficient_wq, &gc->dwork, HZ);
599 }
600 
601 static void
mtype_cancel_gc(struct ip_set * set)602 mtype_cancel_gc(struct ip_set *set)
603 {
604 	struct htype *h = set->data;
605 
606 	if (SET_WITH_TIMEOUT(set))
607 		cancel_delayed_work_sync(&h->gc.dwork);
608 }
609 
610 static int
611 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
612 	  struct ip_set_ext *mext, u32 flags);
613 static int
614 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
615 	  struct ip_set_ext *mext, u32 flags);
616 
617 /* Resize a hash: create a new hash table with doubling the hashsize
618  * and inserting the elements to it. Repeat until we succeed or
619  * fail due to memory pressures.
620  */
621 static int
mtype_resize(struct ip_set * set,bool retried)622 mtype_resize(struct ip_set *set, bool retried)
623 {
624 	struct htype *h = set->data;
625 	struct htable *t, *orig;
626 	u8 htable_bits;
627 	size_t hsize, dsize = set->dsize;
628 #ifdef IP_SET_HASH_WITH_NETS
629 	u8 flags;
630 	struct mtype_elem *tmp;
631 #endif
632 	struct mtype_elem *data;
633 	struct mtype_elem *d;
634 	struct hbucket *n, *m;
635 	struct list_head *l, *lt;
636 	struct mtype_resize_ad *x;
637 	u32 i, j, r, nr, key;
638 	int ret;
639 
640 #ifdef IP_SET_HASH_WITH_NETS
641 	tmp = kmalloc(dsize, GFP_KERNEL);
642 	if (!tmp)
643 		return -ENOMEM;
644 #endif
645 	orig = ipset_dereference_bh_nfnl(h->table);
646 	htable_bits = orig->htable_bits;
647 
648 retry:
649 	ret = 0;
650 	htable_bits++;
651 	if (!htable_bits)
652 		goto hbwarn;
653 	hsize = htable_size(htable_bits);
654 	if (!hsize)
655 		goto hbwarn;
656 	t = ip_set_alloc(hsize);
657 	if (!t) {
658 		ret = -ENOMEM;
659 		goto out;
660 	}
661 	t->hregion = ip_set_alloc(ahash_sizeof_regions(htable_bits));
662 	if (!t->hregion) {
663 		ip_set_free(t);
664 		ret = -ENOMEM;
665 		goto out;
666 	}
667 	t->htable_bits = htable_bits;
668 	t->maxelem = h->maxelem / ahash_numof_locks(htable_bits);
669 	for (i = 0; i < ahash_numof_locks(htable_bits); i++)
670 		spin_lock_init(&t->hregion[i].lock);
671 
672 	/* There can't be another parallel resizing,
673 	 * but dumping, gc, kernel side add/del are possible
674 	 */
675 	orig = ipset_dereference_bh_nfnl(h->table);
676 	atomic_set(&orig->ref, 1);
677 	atomic_inc(&orig->uref);
678 	pr_debug("attempt to resize set %s from %u to %u, t %p\n",
679 		 set->name, orig->htable_bits, htable_bits, orig);
680 	for (r = 0; r < ahash_numof_locks(orig->htable_bits); r++) {
681 		/* Expire may replace a hbucket with another one */
682 		rcu_read_lock_bh();
683 		for (i = ahash_bucket_start(r, orig->htable_bits);
684 		     i < ahash_bucket_end(r, orig->htable_bits); i++) {
685 			n = __ipset_dereference(hbucket(orig, i));
686 			if (!n)
687 				continue;
688 			for (j = 0; j < n->pos; j++) {
689 				if (!test_bit(j, n->used))
690 					continue;
691 				data = ahash_data(n, j, dsize);
692 				if (SET_ELEM_EXPIRED(set, data))
693 					continue;
694 #ifdef IP_SET_HASH_WITH_NETS
695 				/* We have readers running parallel with us,
696 				 * so the live data cannot be modified.
697 				 */
698 				flags = 0;
699 				memcpy(tmp, data, dsize);
700 				data = tmp;
701 				mtype_data_reset_flags(data, &flags);
702 #endif
703 				key = HKEY(data, h->initval, htable_bits);
704 				m = __ipset_dereference(hbucket(t, key));
705 				nr = ahash_region(key, htable_bits);
706 				if (!m) {
707 					m = kzalloc(sizeof(*m) +
708 					    AHASH_INIT_SIZE * dsize,
709 					    GFP_ATOMIC);
710 					if (!m) {
711 						ret = -ENOMEM;
712 						goto cleanup;
713 					}
714 					m->size = AHASH_INIT_SIZE;
715 					t->hregion[nr].ext_size +=
716 						ext_size(AHASH_INIT_SIZE,
717 							 dsize);
718 					RCU_INIT_POINTER(hbucket(t, key), m);
719 				} else if (m->pos >= m->size) {
720 					struct hbucket *ht;
721 
722 					if (m->size >= AHASH_MAX(h)) {
723 						ret = -EAGAIN;
724 					} else {
725 						ht = kzalloc(sizeof(*ht) +
726 						(m->size + AHASH_INIT_SIZE)
727 						* dsize,
728 						GFP_ATOMIC);
729 						if (!ht)
730 							ret = -ENOMEM;
731 					}
732 					if (ret < 0)
733 						goto cleanup;
734 					memcpy(ht, m, sizeof(struct hbucket) +
735 					       m->size * dsize);
736 					ht->size = m->size + AHASH_INIT_SIZE;
737 					t->hregion[nr].ext_size +=
738 						ext_size(AHASH_INIT_SIZE,
739 							 dsize);
740 					kfree(m);
741 					m = ht;
742 					RCU_INIT_POINTER(hbucket(t, key), ht);
743 				}
744 				d = ahash_data(m, m->pos, dsize);
745 				memcpy(d, data, dsize);
746 				set_bit(m->pos++, m->used);
747 				t->hregion[nr].elements++;
748 #ifdef IP_SET_HASH_WITH_NETS
749 				mtype_data_reset_flags(d, &flags);
750 #endif
751 			}
752 		}
753 		rcu_read_unlock_bh();
754 	}
755 
756 	/* There can't be any other writer. */
757 	rcu_assign_pointer(h->table, t);
758 
759 	/* Give time to other readers of the set */
760 	synchronize_rcu();
761 
762 	pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
763 		 orig->htable_bits, orig, t->htable_bits, t);
764 	/* Add/delete elements processed by the SET target during resize.
765 	 * Kernel-side add cannot trigger a resize and userspace actions
766 	 * are serialized by the mutex.
767 	 */
768 	list_for_each_safe(l, lt, &h->ad) {
769 		x = list_entry(l, struct mtype_resize_ad, list);
770 		if (x->ad == IPSET_ADD) {
771 			mtype_add(set, &x->d, &x->ext, &x->mext, x->flags);
772 		} else {
773 			mtype_del(set, &x->d, NULL, NULL, 0);
774 		}
775 		list_del(l);
776 		kfree(l);
777 	}
778 	/* If there's nobody else using the table, destroy it */
779 	if (atomic_dec_and_test(&orig->uref)) {
780 		pr_debug("Table destroy by resize %p\n", orig);
781 		mtype_ahash_destroy(set, orig, false);
782 	}
783 
784 out:
785 #ifdef IP_SET_HASH_WITH_NETS
786 	kfree(tmp);
787 #endif
788 	return ret;
789 
790 cleanup:
791 	rcu_read_unlock_bh();
792 	atomic_set(&orig->ref, 0);
793 	atomic_dec(&orig->uref);
794 	mtype_ahash_destroy(set, t, false);
795 	if (ret == -EAGAIN)
796 		goto retry;
797 	goto out;
798 
799 hbwarn:
800 	/* In case we have plenty of memory :-) */
801 	pr_warn("Cannot increase the hashsize of set %s further\n", set->name);
802 	ret = -IPSET_ERR_HASH_FULL;
803 	goto out;
804 }
805 
806 /* Get the current number of elements and ext_size in the set  */
807 static void
mtype_ext_size(struct ip_set * set,u32 * elements,size_t * ext_size)808 mtype_ext_size(struct ip_set *set, u32 *elements, size_t *ext_size)
809 {
810 	struct htype *h = set->data;
811 	const struct htable *t;
812 	u32 i, j, r;
813 	struct hbucket *n;
814 	struct mtype_elem *data;
815 
816 	t = rcu_dereference_bh(h->table);
817 	for (r = 0; r < ahash_numof_locks(t->htable_bits); r++) {
818 		for (i = ahash_bucket_start(r, t->htable_bits);
819 		     i < ahash_bucket_end(r, t->htable_bits); i++) {
820 			n = rcu_dereference_bh(hbucket(t, i));
821 			if (!n)
822 				continue;
823 			for (j = 0; j < n->pos; j++) {
824 				if (!test_bit(j, n->used))
825 					continue;
826 				data = ahash_data(n, j, set->dsize);
827 				if (!SET_ELEM_EXPIRED(set, data))
828 					(*elements)++;
829 			}
830 		}
831 		*ext_size += t->hregion[r].ext_size;
832 	}
833 }
834 
835 /* Add an element to a hash and update the internal counters when succeeded,
836  * otherwise report the proper error code.
837  */
838 static int
mtype_add(struct ip_set * set,void * value,const struct ip_set_ext * ext,struct ip_set_ext * mext,u32 flags)839 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
840 	  struct ip_set_ext *mext, u32 flags)
841 {
842 	struct htype *h = set->data;
843 	struct htable *t;
844 	const struct mtype_elem *d = value;
845 	struct mtype_elem *data;
846 	struct hbucket *n, *old = ERR_PTR(-ENOENT);
847 	int i, j = -1, ret;
848 	bool flag_exist = flags & IPSET_FLAG_EXIST;
849 	bool deleted = false, forceadd = false, reuse = false;
850 	u32 r, key, multi = 0, elements, maxelem;
851 
852 	rcu_read_lock_bh();
853 	t = rcu_dereference_bh(h->table);
854 	key = HKEY(value, h->initval, t->htable_bits);
855 	r = ahash_region(key, t->htable_bits);
856 	atomic_inc(&t->uref);
857 	elements = t->hregion[r].elements;
858 	maxelem = t->maxelem;
859 	if (elements >= maxelem) {
860 		u32 e;
861 		if (SET_WITH_TIMEOUT(set)) {
862 			rcu_read_unlock_bh();
863 			mtype_gc_do(set, h, t, r);
864 			rcu_read_lock_bh();
865 		}
866 		maxelem = h->maxelem;
867 		elements = 0;
868 		for (e = 0; e < ahash_numof_locks(t->htable_bits); e++)
869 			elements += t->hregion[e].elements;
870 		if (elements >= maxelem && SET_WITH_FORCEADD(set))
871 			forceadd = true;
872 	}
873 	rcu_read_unlock_bh();
874 
875 	spin_lock_bh(&t->hregion[r].lock);
876 	n = rcu_dereference_bh(hbucket(t, key));
877 	if (!n) {
878 		if (forceadd || elements >= maxelem)
879 			goto set_full;
880 		old = NULL;
881 		n = kzalloc(sizeof(*n) + AHASH_INIT_SIZE * set->dsize,
882 			    GFP_ATOMIC);
883 		if (!n) {
884 			ret = -ENOMEM;
885 			goto unlock;
886 		}
887 		n->size = AHASH_INIT_SIZE;
888 		t->hregion[r].ext_size +=
889 			ext_size(AHASH_INIT_SIZE, set->dsize);
890 		goto copy_elem;
891 	}
892 	for (i = 0; i < n->pos; i++) {
893 		if (!test_bit(i, n->used)) {
894 			/* Reuse first deleted entry */
895 			if (j == -1) {
896 				deleted = reuse = true;
897 				j = i;
898 			}
899 			continue;
900 		}
901 		data = ahash_data(n, i, set->dsize);
902 		if (mtype_data_equal(data, d, &multi)) {
903 			if (flag_exist || SET_ELEM_EXPIRED(set, data)) {
904 				/* Just the extensions could be overwritten */
905 				j = i;
906 				goto overwrite_extensions;
907 			}
908 			ret = -IPSET_ERR_EXIST;
909 			goto unlock;
910 		}
911 		/* Reuse first timed out entry */
912 		if (SET_ELEM_EXPIRED(set, data) && j == -1) {
913 			j = i;
914 			reuse = true;
915 		}
916 	}
917 	if (reuse || forceadd) {
918 		if (j == -1)
919 			j = 0;
920 		data = ahash_data(n, j, set->dsize);
921 		if (!deleted) {
922 #ifdef IP_SET_HASH_WITH_NETS
923 			for (i = 0; i < IPSET_NET_COUNT; i++)
924 				mtype_del_cidr(set, h,
925 					NCIDR_PUT(DCIDR_GET(data->cidr, i)),
926 					i);
927 #endif
928 			ip_set_ext_destroy(set, data);
929 			t->hregion[r].elements--;
930 		}
931 		goto copy_data;
932 	}
933 	if (elements >= maxelem)
934 		goto set_full;
935 	/* Create a new slot */
936 	if (n->pos >= n->size) {
937 #ifdef IP_SET_HASH_WITH_MULTI
938 		if (h->bucketsize >= AHASH_MAX_TUNED)
939 			goto set_full;
940 		else if (h->bucketsize <= multi)
941 			h->bucketsize += AHASH_INIT_SIZE;
942 #endif
943 		if (n->size >= AHASH_MAX(h)) {
944 			/* Trigger rehashing */
945 			mtype_data_next(&h->next, d);
946 			ret = -EAGAIN;
947 			goto resize;
948 		}
949 		old = n;
950 		n = kzalloc(sizeof(*n) +
951 			    (old->size + AHASH_INIT_SIZE) * set->dsize,
952 			    GFP_ATOMIC);
953 		if (!n) {
954 			ret = -ENOMEM;
955 			goto unlock;
956 		}
957 		memcpy(n, old, sizeof(struct hbucket) +
958 		       old->size * set->dsize);
959 		n->size = old->size + AHASH_INIT_SIZE;
960 		t->hregion[r].ext_size +=
961 			ext_size(AHASH_INIT_SIZE, set->dsize);
962 	}
963 
964 copy_elem:
965 	j = n->pos++;
966 	data = ahash_data(n, j, set->dsize);
967 copy_data:
968 	t->hregion[r].elements++;
969 #ifdef IP_SET_HASH_WITH_NETS
970 	for (i = 0; i < IPSET_NET_COUNT; i++)
971 		mtype_add_cidr(set, h, NCIDR_PUT(DCIDR_GET(d->cidr, i)), i);
972 #endif
973 	memcpy(data, d, sizeof(struct mtype_elem));
974 overwrite_extensions:
975 #ifdef IP_SET_HASH_WITH_NETS
976 	mtype_data_set_flags(data, flags);
977 #endif
978 	if (SET_WITH_COUNTER(set))
979 		ip_set_init_counter(ext_counter(data, set), ext);
980 	if (SET_WITH_COMMENT(set))
981 		ip_set_init_comment(set, ext_comment(data, set), ext);
982 	if (SET_WITH_SKBINFO(set))
983 		ip_set_init_skbinfo(ext_skbinfo(data, set), ext);
984 	/* Must come last for the case when timed out entry is reused */
985 	if (SET_WITH_TIMEOUT(set))
986 		ip_set_timeout_set(ext_timeout(data, set), ext->timeout);
987 	smp_mb__before_atomic();
988 	set_bit(j, n->used);
989 	if (old != ERR_PTR(-ENOENT)) {
990 		rcu_assign_pointer(hbucket(t, key), n);
991 		if (old)
992 			kfree_rcu(old, rcu);
993 	}
994 	ret = 0;
995 resize:
996 	spin_unlock_bh(&t->hregion[r].lock);
997 	if (atomic_read(&t->ref) && ext->target) {
998 		/* Resize is in process and kernel side add, save values */
999 		struct mtype_resize_ad *x;
1000 
1001 		x = kzalloc(sizeof(struct mtype_resize_ad), GFP_ATOMIC);
1002 		if (!x)
1003 			/* Don't bother */
1004 			goto out;
1005 		x->ad = IPSET_ADD;
1006 		memcpy(&x->d, value, sizeof(struct mtype_elem));
1007 		memcpy(&x->ext, ext, sizeof(struct ip_set_ext));
1008 		memcpy(&x->mext, mext, sizeof(struct ip_set_ext));
1009 		x->flags = flags;
1010 		spin_lock_bh(&set->lock);
1011 		list_add_tail(&x->list, &h->ad);
1012 		spin_unlock_bh(&set->lock);
1013 	}
1014 	goto out;
1015 
1016 set_full:
1017 	if (net_ratelimit())
1018 		pr_warn("Set %s is full, maxelem %u reached\n",
1019 			set->name, maxelem);
1020 	ret = -IPSET_ERR_HASH_FULL;
1021 unlock:
1022 	spin_unlock_bh(&t->hregion[r].lock);
1023 out:
1024 	if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1025 		pr_debug("Table destroy after resize by add: %p\n", t);
1026 		mtype_ahash_destroy(set, t, false);
1027 	}
1028 	return ret;
1029 }
1030 
1031 /* Delete an element from the hash and free up space if possible.
1032  */
1033 static int
mtype_del(struct ip_set * set,void * value,const struct ip_set_ext * ext,struct ip_set_ext * mext,u32 flags)1034 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1035 	  struct ip_set_ext *mext, u32 flags)
1036 {
1037 	struct htype *h = set->data;
1038 	struct htable *t;
1039 	const struct mtype_elem *d = value;
1040 	struct mtype_elem *data;
1041 	struct hbucket *n;
1042 	struct mtype_resize_ad *x = NULL;
1043 	int i, j, k, r, ret = -IPSET_ERR_EXIST;
1044 	u32 key, multi = 0;
1045 	size_t dsize = set->dsize;
1046 
1047 	/* Userspace add and resize is excluded by the mutex.
1048 	 * Kernespace add does not trigger resize.
1049 	 */
1050 	rcu_read_lock_bh();
1051 	t = rcu_dereference_bh(h->table);
1052 	key = HKEY(value, h->initval, t->htable_bits);
1053 	r = ahash_region(key, t->htable_bits);
1054 	atomic_inc(&t->uref);
1055 	rcu_read_unlock_bh();
1056 
1057 	spin_lock_bh(&t->hregion[r].lock);
1058 	n = rcu_dereference_bh(hbucket(t, key));
1059 	if (!n)
1060 		goto out;
1061 	for (i = 0, k = 0; i < n->pos; i++) {
1062 		if (!test_bit(i, n->used)) {
1063 			k++;
1064 			continue;
1065 		}
1066 		data = ahash_data(n, i, dsize);
1067 		if (!mtype_data_equal(data, d, &multi))
1068 			continue;
1069 		if (SET_ELEM_EXPIRED(set, data))
1070 			goto out;
1071 
1072 		ret = 0;
1073 		clear_bit(i, n->used);
1074 		smp_mb__after_atomic();
1075 		if (i + 1 == n->pos)
1076 			n->pos--;
1077 		t->hregion[r].elements--;
1078 #ifdef IP_SET_HASH_WITH_NETS
1079 		for (j = 0; j < IPSET_NET_COUNT; j++)
1080 			mtype_del_cidr(set, h,
1081 				       NCIDR_PUT(DCIDR_GET(d->cidr, j)), j);
1082 #endif
1083 		ip_set_ext_destroy(set, data);
1084 
1085 		if (atomic_read(&t->ref) && ext->target) {
1086 			/* Resize is in process and kernel side del,
1087 			 * save values
1088 			 */
1089 			x = kzalloc(sizeof(struct mtype_resize_ad),
1090 				    GFP_ATOMIC);
1091 			if (x) {
1092 				x->ad = IPSET_DEL;
1093 				memcpy(&x->d, value,
1094 				       sizeof(struct mtype_elem));
1095 				x->flags = flags;
1096 			}
1097 		}
1098 		for (; i < n->pos; i++) {
1099 			if (!test_bit(i, n->used))
1100 				k++;
1101 		}
1102 		if (n->pos == 0 && k == 0) {
1103 			t->hregion[r].ext_size -= ext_size(n->size, dsize);
1104 			rcu_assign_pointer(hbucket(t, key), NULL);
1105 			kfree_rcu(n, rcu);
1106 		} else if (k >= AHASH_INIT_SIZE) {
1107 			struct hbucket *tmp = kzalloc(sizeof(*tmp) +
1108 					(n->size - AHASH_INIT_SIZE) * dsize,
1109 					GFP_ATOMIC);
1110 			if (!tmp)
1111 				goto out;
1112 			tmp->size = n->size - AHASH_INIT_SIZE;
1113 			for (j = 0, k = 0; j < n->pos; j++) {
1114 				if (!test_bit(j, n->used))
1115 					continue;
1116 				data = ahash_data(n, j, dsize);
1117 				memcpy(tmp->value + k * dsize, data, dsize);
1118 				set_bit(k, tmp->used);
1119 				k++;
1120 			}
1121 			tmp->pos = k;
1122 			t->hregion[r].ext_size -=
1123 				ext_size(AHASH_INIT_SIZE, dsize);
1124 			rcu_assign_pointer(hbucket(t, key), tmp);
1125 			kfree_rcu(n, rcu);
1126 		}
1127 		goto out;
1128 	}
1129 
1130 out:
1131 	spin_unlock_bh(&t->hregion[r].lock);
1132 	if (x) {
1133 		spin_lock_bh(&set->lock);
1134 		list_add(&x->list, &h->ad);
1135 		spin_unlock_bh(&set->lock);
1136 	}
1137 	if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1138 		pr_debug("Table destroy after resize by del: %p\n", t);
1139 		mtype_ahash_destroy(set, t, false);
1140 	}
1141 	return ret;
1142 }
1143 
1144 static int
mtype_data_match(struct mtype_elem * data,const struct ip_set_ext * ext,struct ip_set_ext * mext,struct ip_set * set,u32 flags)1145 mtype_data_match(struct mtype_elem *data, const struct ip_set_ext *ext,
1146 		 struct ip_set_ext *mext, struct ip_set *set, u32 flags)
1147 {
1148 	if (!ip_set_match_extensions(set, ext, mext, flags, data))
1149 		return 0;
1150 	/* nomatch entries return -ENOTEMPTY */
1151 	return mtype_do_data_match(data);
1152 }
1153 
1154 #ifdef IP_SET_HASH_WITH_NETS
1155 /* Special test function which takes into account the different network
1156  * sizes added to the set
1157  */
1158 static int
mtype_test_cidrs(struct ip_set * set,struct mtype_elem * d,const struct ip_set_ext * ext,struct ip_set_ext * mext,u32 flags)1159 mtype_test_cidrs(struct ip_set *set, struct mtype_elem *d,
1160 		 const struct ip_set_ext *ext,
1161 		 struct ip_set_ext *mext, u32 flags)
1162 {
1163 	struct htype *h = set->data;
1164 	struct htable *t = rcu_dereference_bh(h->table);
1165 	struct hbucket *n;
1166 	struct mtype_elem *data;
1167 #if IPSET_NET_COUNT == 2
1168 	struct mtype_elem orig = *d;
1169 	int ret, i, j = 0, k;
1170 #else
1171 	int ret, i, j = 0;
1172 #endif
1173 	u32 key, multi = 0;
1174 
1175 	pr_debug("test by nets\n");
1176 	for (; j < NLEN && h->nets[j].cidr[0] && !multi; j++) {
1177 #if IPSET_NET_COUNT == 2
1178 		mtype_data_reset_elem(d, &orig);
1179 		mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]), false);
1180 		for (k = 0; k < NLEN && h->nets[k].cidr[1] && !multi;
1181 		     k++) {
1182 			mtype_data_netmask(d, NCIDR_GET(h->nets[k].cidr[1]),
1183 					   true);
1184 #else
1185 		mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]));
1186 #endif
1187 		key = HKEY(d, h->initval, t->htable_bits);
1188 		n = rcu_dereference_bh(hbucket(t, key));
1189 		if (!n)
1190 			continue;
1191 		for (i = 0; i < n->pos; i++) {
1192 			if (!test_bit(i, n->used))
1193 				continue;
1194 			data = ahash_data(n, i, set->dsize);
1195 			if (!mtype_data_equal(data, d, &multi))
1196 				continue;
1197 			ret = mtype_data_match(data, ext, mext, set, flags);
1198 			if (ret != 0)
1199 				return ret;
1200 #ifdef IP_SET_HASH_WITH_MULTI
1201 			/* No match, reset multiple match flag */
1202 			multi = 0;
1203 #endif
1204 		}
1205 #if IPSET_NET_COUNT == 2
1206 		}
1207 #endif
1208 	}
1209 	return 0;
1210 }
1211 #endif
1212 
1213 /* Test whether the element is added to the set */
1214 static int
mtype_test(struct ip_set * set,void * value,const struct ip_set_ext * ext,struct ip_set_ext * mext,u32 flags)1215 mtype_test(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1216 	   struct ip_set_ext *mext, u32 flags)
1217 {
1218 	struct htype *h = set->data;
1219 	struct htable *t;
1220 	struct mtype_elem *d = value;
1221 	struct hbucket *n;
1222 	struct mtype_elem *data;
1223 	int i, ret = 0;
1224 	u32 key, multi = 0;
1225 
1226 	rcu_read_lock_bh();
1227 	t = rcu_dereference_bh(h->table);
1228 #ifdef IP_SET_HASH_WITH_NETS
1229 	/* If we test an IP address and not a network address,
1230 	 * try all possible network sizes
1231 	 */
1232 	for (i = 0; i < IPSET_NET_COUNT; i++)
1233 		if (DCIDR_GET(d->cidr, i) != HOST_MASK)
1234 			break;
1235 	if (i == IPSET_NET_COUNT) {
1236 		ret = mtype_test_cidrs(set, d, ext, mext, flags);
1237 		goto out;
1238 	}
1239 #endif
1240 
1241 	key = HKEY(d, h->initval, t->htable_bits);
1242 	n = rcu_dereference_bh(hbucket(t, key));
1243 	if (!n) {
1244 		ret = 0;
1245 		goto out;
1246 	}
1247 	for (i = 0; i < n->pos; i++) {
1248 		if (!test_bit(i, n->used))
1249 			continue;
1250 		data = ahash_data(n, i, set->dsize);
1251 		if (!mtype_data_equal(data, d, &multi))
1252 			continue;
1253 		ret = mtype_data_match(data, ext, mext, set, flags);
1254 		if (ret != 0)
1255 			goto out;
1256 	}
1257 out:
1258 	rcu_read_unlock_bh();
1259 	return ret;
1260 }
1261 
1262 /* Reply a HEADER request: fill out the header part of the set */
1263 static int
mtype_head(struct ip_set * set,struct sk_buff * skb)1264 mtype_head(struct ip_set *set, struct sk_buff *skb)
1265 {
1266 	struct htype *h = set->data;
1267 	const struct htable *t;
1268 	struct nlattr *nested;
1269 	size_t memsize;
1270 	u32 elements = 0;
1271 	size_t ext_size = 0;
1272 	u8 htable_bits;
1273 
1274 	rcu_read_lock_bh();
1275 	t = rcu_dereference_bh(h->table);
1276 	mtype_ext_size(set, &elements, &ext_size);
1277 	memsize = mtype_ahash_memsize(h, t) + ext_size + set->ext_size;
1278 	htable_bits = t->htable_bits;
1279 	rcu_read_unlock_bh();
1280 
1281 	nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1282 	if (!nested)
1283 		goto nla_put_failure;
1284 	if (nla_put_net32(skb, IPSET_ATTR_HASHSIZE,
1285 			  htonl(jhash_size(htable_bits))) ||
1286 	    nla_put_net32(skb, IPSET_ATTR_MAXELEM, htonl(h->maxelem)))
1287 		goto nla_put_failure;
1288 #ifdef IP_SET_HASH_WITH_BITMASK
1289 	/* if netmask is set to anything other than HOST_MASK we know that the user supplied netmask
1290 	 * and not bitmask. These two are mutually exclusive. */
1291 	if (h->netmask == HOST_MASK && !nf_inet_addr_cmp(&onesmask, &h->bitmask)) {
1292 		if (set->family == NFPROTO_IPV4) {
1293 			if (nla_put_ipaddr4(skb, IPSET_ATTR_BITMASK, h->bitmask.ip))
1294 				goto nla_put_failure;
1295 		} else if (set->family == NFPROTO_IPV6) {
1296 			if (nla_put_ipaddr6(skb, IPSET_ATTR_BITMASK, &h->bitmask.in6))
1297 				goto nla_put_failure;
1298 		}
1299 	}
1300 #endif
1301 #ifdef IP_SET_HASH_WITH_NETMASK
1302 	if (h->netmask != HOST_MASK && nla_put_u8(skb, IPSET_ATTR_NETMASK, h->netmask))
1303 		goto nla_put_failure;
1304 #endif
1305 #ifdef IP_SET_HASH_WITH_MARKMASK
1306 	if (nla_put_u32(skb, IPSET_ATTR_MARKMASK, h->markmask))
1307 		goto nla_put_failure;
1308 #endif
1309 	if (set->flags & IPSET_CREATE_FLAG_BUCKETSIZE) {
1310 		if (nla_put_u8(skb, IPSET_ATTR_BUCKETSIZE, h->bucketsize) ||
1311 		    nla_put_net32(skb, IPSET_ATTR_INITVAL, htonl(h->initval)))
1312 			goto nla_put_failure;
1313 	}
1314 	if (nla_put_net32(skb, IPSET_ATTR_REFERENCES, htonl(set->ref)) ||
1315 	    nla_put_net32(skb, IPSET_ATTR_MEMSIZE, htonl(memsize)) ||
1316 	    nla_put_net32(skb, IPSET_ATTR_ELEMENTS, htonl(elements)))
1317 		goto nla_put_failure;
1318 	if (unlikely(ip_set_put_flags(skb, set)))
1319 		goto nla_put_failure;
1320 	nla_nest_end(skb, nested);
1321 
1322 	return 0;
1323 nla_put_failure:
1324 	return -EMSGSIZE;
1325 }
1326 
1327 /* Make possible to run dumping parallel with resizing */
1328 static void
mtype_uref(struct ip_set * set,struct netlink_callback * cb,bool start)1329 mtype_uref(struct ip_set *set, struct netlink_callback *cb, bool start)
1330 {
1331 	struct htype *h = set->data;
1332 	struct htable *t;
1333 
1334 	if (start) {
1335 		rcu_read_lock_bh();
1336 		t = ipset_dereference_bh_nfnl(h->table);
1337 		atomic_inc(&t->uref);
1338 		cb->args[IPSET_CB_PRIVATE] = (unsigned long)t;
1339 		rcu_read_unlock_bh();
1340 	} else if (cb->args[IPSET_CB_PRIVATE]) {
1341 		t = (struct htable *)cb->args[IPSET_CB_PRIVATE];
1342 		if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1343 			pr_debug("Table destroy after resize "
1344 				 " by dump: %p\n", t);
1345 			mtype_ahash_destroy(set, t, false);
1346 		}
1347 		cb->args[IPSET_CB_PRIVATE] = 0;
1348 	}
1349 }
1350 
1351 /* Reply a LIST/SAVE request: dump the elements of the specified set */
1352 static int
mtype_list(const struct ip_set * set,struct sk_buff * skb,struct netlink_callback * cb)1353 mtype_list(const struct ip_set *set,
1354 	   struct sk_buff *skb, struct netlink_callback *cb)
1355 {
1356 	const struct htable *t;
1357 	struct nlattr *atd, *nested;
1358 	const struct hbucket *n;
1359 	const struct mtype_elem *e;
1360 	u32 first = cb->args[IPSET_CB_ARG0];
1361 	/* We assume that one hash bucket fills into one page */
1362 	void *incomplete;
1363 	int i, ret = 0;
1364 
1365 	atd = nla_nest_start(skb, IPSET_ATTR_ADT);
1366 	if (!atd)
1367 		return -EMSGSIZE;
1368 
1369 	pr_debug("list hash set %s\n", set->name);
1370 	t = (const struct htable *)cb->args[IPSET_CB_PRIVATE];
1371 	/* Expire may replace a hbucket with another one */
1372 	rcu_read_lock();
1373 	for (; cb->args[IPSET_CB_ARG0] < jhash_size(t->htable_bits);
1374 	     cb->args[IPSET_CB_ARG0]++) {
1375 		cond_resched_rcu();
1376 		incomplete = skb_tail_pointer(skb);
1377 		n = rcu_dereference(hbucket(t, cb->args[IPSET_CB_ARG0]));
1378 		pr_debug("cb->arg bucket: %lu, t %p n %p\n",
1379 			 cb->args[IPSET_CB_ARG0], t, n);
1380 		if (!n)
1381 			continue;
1382 		for (i = 0; i < n->pos; i++) {
1383 			if (!test_bit(i, n->used))
1384 				continue;
1385 			e = ahash_data(n, i, set->dsize);
1386 			if (SET_ELEM_EXPIRED(set, e))
1387 				continue;
1388 			pr_debug("list hash %lu hbucket %p i %u, data %p\n",
1389 				 cb->args[IPSET_CB_ARG0], n, i, e);
1390 			nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1391 			if (!nested) {
1392 				if (cb->args[IPSET_CB_ARG0] == first) {
1393 					nla_nest_cancel(skb, atd);
1394 					ret = -EMSGSIZE;
1395 					goto out;
1396 				}
1397 				goto nla_put_failure;
1398 			}
1399 			if (mtype_data_list(skb, e))
1400 				goto nla_put_failure;
1401 			if (ip_set_put_extensions(skb, set, e, true))
1402 				goto nla_put_failure;
1403 			nla_nest_end(skb, nested);
1404 		}
1405 	}
1406 	nla_nest_end(skb, atd);
1407 	/* Set listing finished */
1408 	cb->args[IPSET_CB_ARG0] = 0;
1409 
1410 	goto out;
1411 
1412 nla_put_failure:
1413 	nlmsg_trim(skb, incomplete);
1414 	if (unlikely(first == cb->args[IPSET_CB_ARG0])) {
1415 		pr_warn("Can't list set %s: one bucket does not fit into a message. Please report it!\n",
1416 			set->name);
1417 		cb->args[IPSET_CB_ARG0] = 0;
1418 		ret = -EMSGSIZE;
1419 	} else {
1420 		nla_nest_end(skb, atd);
1421 	}
1422 out:
1423 	rcu_read_unlock();
1424 	return ret;
1425 }
1426 
1427 static int
1428 IPSET_TOKEN(MTYPE, _kadt)(struct ip_set *set, const struct sk_buff *skb,
1429 			  const struct xt_action_param *par,
1430 			  enum ipset_adt adt, struct ip_set_adt_opt *opt);
1431 
1432 static int
1433 IPSET_TOKEN(MTYPE, _uadt)(struct ip_set *set, struct nlattr *tb[],
1434 			  enum ipset_adt adt, u32 *lineno, u32 flags,
1435 			  bool retried);
1436 
1437 static const struct ip_set_type_variant mtype_variant = {
1438 	.kadt	= mtype_kadt,
1439 	.uadt	= mtype_uadt,
1440 	.adt	= {
1441 		[IPSET_ADD] = mtype_add,
1442 		[IPSET_DEL] = mtype_del,
1443 		[IPSET_TEST] = mtype_test,
1444 	},
1445 	.destroy = mtype_destroy,
1446 	.flush	= mtype_flush,
1447 	.head	= mtype_head,
1448 	.list	= mtype_list,
1449 	.uref	= mtype_uref,
1450 	.resize	= mtype_resize,
1451 	.same_set = mtype_same_set,
1452 	.cancel_gc = mtype_cancel_gc,
1453 	.region_lock = true,
1454 };
1455 
1456 #ifdef IP_SET_EMIT_CREATE
1457 static int
IPSET_TOKEN(HTYPE,_create)1458 IPSET_TOKEN(HTYPE, _create)(struct net *net, struct ip_set *set,
1459 			    struct nlattr *tb[], u32 flags)
1460 {
1461 	u32 hashsize = IPSET_DEFAULT_HASHSIZE, maxelem = IPSET_DEFAULT_MAXELEM;
1462 #ifdef IP_SET_HASH_WITH_MARKMASK
1463 	u32 markmask;
1464 #endif
1465 	u8 hbits;
1466 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
1467 	int ret __attribute__((unused)) = 0;
1468 	u8 netmask = set->family == NFPROTO_IPV4 ? 32 : 128;
1469 	union nf_inet_addr bitmask = onesmask;
1470 #endif
1471 	size_t hsize;
1472 	struct htype *h;
1473 	struct htable *t;
1474 	u32 i;
1475 
1476 	pr_debug("Create set %s with family %s\n",
1477 		 set->name, set->family == NFPROTO_IPV4 ? "inet" : "inet6");
1478 
1479 #ifdef IP_SET_PROTO_UNDEF
1480 	if (set->family != NFPROTO_UNSPEC)
1481 		return -IPSET_ERR_INVALID_FAMILY;
1482 #else
1483 	if (!(set->family == NFPROTO_IPV4 || set->family == NFPROTO_IPV6))
1484 		return -IPSET_ERR_INVALID_FAMILY;
1485 #endif
1486 
1487 	if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_HASHSIZE) ||
1488 		     !ip_set_optattr_netorder(tb, IPSET_ATTR_MAXELEM) ||
1489 		     !ip_set_optattr_netorder(tb, IPSET_ATTR_TIMEOUT) ||
1490 		     !ip_set_optattr_netorder(tb, IPSET_ATTR_CADT_FLAGS)))
1491 		return -IPSET_ERR_PROTOCOL;
1492 
1493 #ifdef IP_SET_HASH_WITH_MARKMASK
1494 	/* Separated condition in order to avoid directive in argument list */
1495 	if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_MARKMASK)))
1496 		return -IPSET_ERR_PROTOCOL;
1497 
1498 	markmask = 0xffffffff;
1499 	if (tb[IPSET_ATTR_MARKMASK]) {
1500 		markmask = ntohl(nla_get_be32(tb[IPSET_ATTR_MARKMASK]));
1501 		if (markmask == 0)
1502 			return -IPSET_ERR_INVALID_MARKMASK;
1503 	}
1504 #endif
1505 
1506 #ifdef IP_SET_HASH_WITH_NETMASK
1507 	if (tb[IPSET_ATTR_NETMASK]) {
1508 		netmask = nla_get_u8(tb[IPSET_ATTR_NETMASK]);
1509 
1510 		if ((set->family == NFPROTO_IPV4 && netmask > 32) ||
1511 		    (set->family == NFPROTO_IPV6 && netmask > 128) ||
1512 		    netmask == 0)
1513 			return -IPSET_ERR_INVALID_NETMASK;
1514 
1515 		/* we convert netmask to bitmask and store it */
1516 		if (set->family == NFPROTO_IPV4)
1517 			bitmask.ip = ip_set_netmask(netmask);
1518 		else
1519 			ip6_netmask(&bitmask, netmask);
1520 	}
1521 #endif
1522 
1523 #ifdef IP_SET_HASH_WITH_BITMASK
1524 	if (tb[IPSET_ATTR_BITMASK]) {
1525 		/* bitmask and netmask do the same thing, allow only one of these options */
1526 		if (tb[IPSET_ATTR_NETMASK])
1527 			return -IPSET_ERR_BITMASK_NETMASK_EXCL;
1528 
1529 		if (set->family == NFPROTO_IPV4) {
1530 			ret = ip_set_get_ipaddr4(tb[IPSET_ATTR_BITMASK], &bitmask.ip);
1531 			if (ret || !bitmask.ip)
1532 				return -IPSET_ERR_INVALID_NETMASK;
1533 		} else if (set->family == NFPROTO_IPV6) {
1534 			ret = ip_set_get_ipaddr6(tb[IPSET_ATTR_BITMASK], &bitmask);
1535 			if (ret || ipv6_addr_any(&bitmask.in6))
1536 				return -IPSET_ERR_INVALID_NETMASK;
1537 		}
1538 
1539 		if (nf_inet_addr_cmp(&bitmask, &zeromask))
1540 			return -IPSET_ERR_INVALID_NETMASK;
1541 	}
1542 #endif
1543 
1544 	if (tb[IPSET_ATTR_HASHSIZE]) {
1545 		hashsize = ip_set_get_h32(tb[IPSET_ATTR_HASHSIZE]);
1546 		if (hashsize < IPSET_MIMINAL_HASHSIZE)
1547 			hashsize = IPSET_MIMINAL_HASHSIZE;
1548 	}
1549 
1550 	if (tb[IPSET_ATTR_MAXELEM])
1551 		maxelem = ip_set_get_h32(tb[IPSET_ATTR_MAXELEM]);
1552 
1553 	hsize = sizeof(*h);
1554 	h = kzalloc(hsize, GFP_KERNEL);
1555 	if (!h)
1556 		return -ENOMEM;
1557 
1558 	/* Compute htable_bits from the user input parameter hashsize.
1559 	 * Assume that hashsize == 2^htable_bits,
1560 	 * otherwise round up to the first 2^n value.
1561 	 */
1562 	hbits = fls(hashsize - 1);
1563 	hsize = htable_size(hbits);
1564 	if (hsize == 0) {
1565 		kfree(h);
1566 		return -ENOMEM;
1567 	}
1568 	t = ip_set_alloc(hsize);
1569 	if (!t) {
1570 		kfree(h);
1571 		return -ENOMEM;
1572 	}
1573 	t->hregion = ip_set_alloc(ahash_sizeof_regions(hbits));
1574 	if (!t->hregion) {
1575 		ip_set_free(t);
1576 		kfree(h);
1577 		return -ENOMEM;
1578 	}
1579 	h->gc.set = set;
1580 	for (i = 0; i < ahash_numof_locks(hbits); i++)
1581 		spin_lock_init(&t->hregion[i].lock);
1582 	h->maxelem = maxelem;
1583 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
1584 	h->bitmask = bitmask;
1585 	h->netmask = netmask;
1586 #endif
1587 #ifdef IP_SET_HASH_WITH_MARKMASK
1588 	h->markmask = markmask;
1589 #endif
1590 	if (tb[IPSET_ATTR_INITVAL])
1591 		h->initval = ntohl(nla_get_be32(tb[IPSET_ATTR_INITVAL]));
1592 	else
1593 		get_random_bytes(&h->initval, sizeof(h->initval));
1594 	h->bucketsize = AHASH_MAX_SIZE;
1595 	if (tb[IPSET_ATTR_BUCKETSIZE]) {
1596 		h->bucketsize = nla_get_u8(tb[IPSET_ATTR_BUCKETSIZE]);
1597 		if (h->bucketsize < AHASH_INIT_SIZE)
1598 			h->bucketsize = AHASH_INIT_SIZE;
1599 		else if (h->bucketsize > AHASH_MAX_SIZE)
1600 			h->bucketsize = AHASH_MAX_SIZE;
1601 		else if (h->bucketsize % 2)
1602 			h->bucketsize += 1;
1603 	}
1604 	t->htable_bits = hbits;
1605 	t->maxelem = h->maxelem / ahash_numof_locks(hbits);
1606 	RCU_INIT_POINTER(h->table, t);
1607 
1608 	INIT_LIST_HEAD(&h->ad);
1609 	set->data = h;
1610 #ifndef IP_SET_PROTO_UNDEF
1611 	if (set->family == NFPROTO_IPV4) {
1612 #endif
1613 		set->variant = &IPSET_TOKEN(HTYPE, 4_variant);
1614 		set->dsize = ip_set_elem_len(set, tb,
1615 			sizeof(struct IPSET_TOKEN(HTYPE, 4_elem)),
1616 			__alignof__(struct IPSET_TOKEN(HTYPE, 4_elem)));
1617 #ifndef IP_SET_PROTO_UNDEF
1618 	} else {
1619 		set->variant = &IPSET_TOKEN(HTYPE, 6_variant);
1620 		set->dsize = ip_set_elem_len(set, tb,
1621 			sizeof(struct IPSET_TOKEN(HTYPE, 6_elem)),
1622 			__alignof__(struct IPSET_TOKEN(HTYPE, 6_elem)));
1623 	}
1624 #endif
1625 	set->timeout = IPSET_NO_TIMEOUT;
1626 	if (tb[IPSET_ATTR_TIMEOUT]) {
1627 		set->timeout = ip_set_timeout_uget(tb[IPSET_ATTR_TIMEOUT]);
1628 #ifndef IP_SET_PROTO_UNDEF
1629 		if (set->family == NFPROTO_IPV4)
1630 #endif
1631 			IPSET_TOKEN(HTYPE, 4_gc_init)(&h->gc);
1632 #ifndef IP_SET_PROTO_UNDEF
1633 		else
1634 			IPSET_TOKEN(HTYPE, 6_gc_init)(&h->gc);
1635 #endif
1636 	}
1637 	pr_debug("create %s hashsize %u (%u) maxelem %u: %p(%p)\n",
1638 		 set->name, jhash_size(t->htable_bits),
1639 		 t->htable_bits, h->maxelem, set->data, t);
1640 
1641 	return 0;
1642 }
1643 #endif /* IP_SET_EMIT_CREATE */
1644 
1645 #undef HKEY_DATALEN
1646