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
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);
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);
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_obj(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);
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_obj(struct mtype_resize_ad, GFP_ATOMIC);
1090 if (x) {
1091 x->ad = IPSET_DEL;
1092 memcpy(&x->d, value,
1093 sizeof(struct mtype_elem));
1094 x->flags = flags;
1095 }
1096 }
1097 for (; i < n->pos; i++) {
1098 if (!test_bit(i, n->used))
1099 k++;
1100 }
1101 if (k == n->pos) {
1102 t->hregion[r].ext_size -= ext_size(n->size, dsize);
1103 rcu_assign_pointer(hbucket(t, key), NULL);
1104 kfree_rcu(n, rcu);
1105 } else if (k >= AHASH_INIT_SIZE) {
1106 struct hbucket *tmp = kzalloc(sizeof(*tmp) +
1107 (n->size - AHASH_INIT_SIZE) * dsize,
1108 GFP_ATOMIC);
1109 if (!tmp)
1110 goto out;
1111 tmp->size = n->size - AHASH_INIT_SIZE;
1112 for (j = 0, k = 0; j < n->pos; j++) {
1113 if (!test_bit(j, n->used))
1114 continue;
1115 data = ahash_data(n, j, dsize);
1116 memcpy(tmp->value + k * dsize, data, dsize);
1117 set_bit(k, tmp->used);
1118 k++;
1119 }
1120 tmp->pos = k;
1121 t->hregion[r].ext_size -=
1122 ext_size(AHASH_INIT_SIZE, dsize);
1123 rcu_assign_pointer(hbucket(t, key), tmp);
1124 kfree_rcu(n, rcu);
1125 }
1126 goto out;
1127 }
1128
1129 out:
1130 spin_unlock_bh(&t->hregion[r].lock);
1131 if (x) {
1132 spin_lock_bh(&set->lock);
1133 list_add(&x->list, &h->ad);
1134 spin_unlock_bh(&set->lock);
1135 }
1136 if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1137 pr_debug("Table destroy after resize by del: %p\n", t);
1138 mtype_ahash_destroy(set, t, false);
1139 }
1140 return ret;
1141 }
1142
1143 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)1144 mtype_data_match(struct mtype_elem *data, const struct ip_set_ext *ext,
1145 struct ip_set_ext *mext, struct ip_set *set, u32 flags)
1146 {
1147 if (!ip_set_match_extensions(set, ext, mext, flags, data))
1148 return 0;
1149 /* nomatch entries return -ENOTEMPTY */
1150 return mtype_do_data_match(data);
1151 }
1152
1153 #ifdef IP_SET_HASH_WITH_NETS
1154 /* Special test function which takes into account the different network
1155 * sizes added to the set
1156 */
1157 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)1158 mtype_test_cidrs(struct ip_set *set, struct mtype_elem *d,
1159 const struct ip_set_ext *ext,
1160 struct ip_set_ext *mext, u32 flags)
1161 {
1162 struct htype *h = set->data;
1163 struct htable *t = rcu_dereference_bh(h->table);
1164 struct hbucket *n;
1165 struct mtype_elem *data;
1166 #if IPSET_NET_COUNT == 2
1167 struct mtype_elem orig = *d;
1168 int ret, i, j = 0, k;
1169 #else
1170 int ret, i, j = 0;
1171 #endif
1172 u32 key, multi = 0;
1173
1174 pr_debug("test by nets\n");
1175 for (; j < NLEN && h->nets[j].cidr[0] && !multi; j++) {
1176 #if IPSET_NET_COUNT == 2
1177 mtype_data_reset_elem(d, &orig);
1178 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]), false);
1179 for (k = 0; k < NLEN && h->nets[k].cidr[1] && !multi;
1180 k++) {
1181 mtype_data_netmask(d, NCIDR_GET(h->nets[k].cidr[1]),
1182 true);
1183 #else
1184 mtype_data_netmask(d, NCIDR_GET(h->nets[j].cidr[0]));
1185 #endif
1186 key = HKEY(d, h->initval, t->htable_bits);
1187 n = rcu_dereference_bh(hbucket(t, key));
1188 if (!n)
1189 continue;
1190 for (i = 0; i < n->pos; i++) {
1191 if (!test_bit(i, n->used))
1192 continue;
1193 data = ahash_data(n, i, set->dsize);
1194 if (!mtype_data_equal(data, d, &multi))
1195 continue;
1196 ret = mtype_data_match(data, ext, mext, set, flags);
1197 if (ret != 0)
1198 return ret;
1199 #ifdef IP_SET_HASH_WITH_MULTI
1200 /* No match, reset multiple match flag */
1201 multi = 0;
1202 #endif
1203 }
1204 #if IPSET_NET_COUNT == 2
1205 }
1206 #endif
1207 }
1208 return 0;
1209 }
1210 #endif
1211
1212 /* Test whether the element is added to the set */
1213 static int
mtype_test(struct ip_set * set,void * value,const struct ip_set_ext * ext,struct ip_set_ext * mext,u32 flags)1214 mtype_test(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1215 struct ip_set_ext *mext, u32 flags)
1216 {
1217 struct htype *h = set->data;
1218 struct htable *t;
1219 struct mtype_elem *d = value;
1220 struct hbucket *n;
1221 struct mtype_elem *data;
1222 int i, ret = 0;
1223 u32 key, multi = 0;
1224
1225 rcu_read_lock_bh();
1226 t = rcu_dereference_bh(h->table);
1227 #ifdef IP_SET_HASH_WITH_NETS
1228 /* If we test an IP address and not a network address,
1229 * try all possible network sizes
1230 */
1231 for (i = 0; i < IPSET_NET_COUNT; i++)
1232 if (DCIDR_GET(d->cidr, i) != HOST_MASK)
1233 break;
1234 if (i == IPSET_NET_COUNT) {
1235 ret = mtype_test_cidrs(set, d, ext, mext, flags);
1236 goto out;
1237 }
1238 #endif
1239
1240 key = HKEY(d, h->initval, t->htable_bits);
1241 n = rcu_dereference_bh(hbucket(t, key));
1242 if (!n) {
1243 ret = 0;
1244 goto out;
1245 }
1246 for (i = 0; i < n->pos; i++) {
1247 if (!test_bit(i, n->used))
1248 continue;
1249 data = ahash_data(n, i, set->dsize);
1250 if (!mtype_data_equal(data, d, &multi))
1251 continue;
1252 ret = mtype_data_match(data, ext, mext, set, flags);
1253 if (ret != 0)
1254 goto out;
1255 }
1256 out:
1257 rcu_read_unlock_bh();
1258 return ret;
1259 }
1260
1261 /* Reply a HEADER request: fill out the header part of the set */
1262 static int
mtype_head(struct ip_set * set,struct sk_buff * skb)1263 mtype_head(struct ip_set *set, struct sk_buff *skb)
1264 {
1265 struct htype *h = set->data;
1266 const struct htable *t;
1267 struct nlattr *nested;
1268 size_t memsize;
1269 u32 elements = 0;
1270 size_t ext_size = 0;
1271 u8 htable_bits;
1272
1273 rcu_read_lock_bh();
1274 t = rcu_dereference_bh(h->table);
1275 mtype_ext_size(set, &elements, &ext_size);
1276 memsize = mtype_ahash_memsize(h, t) + ext_size + set->ext_size;
1277 htable_bits = t->htable_bits;
1278 rcu_read_unlock_bh();
1279
1280 nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1281 if (!nested)
1282 goto nla_put_failure;
1283 if (nla_put_net32(skb, IPSET_ATTR_HASHSIZE,
1284 htonl(jhash_size(htable_bits))) ||
1285 nla_put_net32(skb, IPSET_ATTR_MAXELEM, htonl(h->maxelem)))
1286 goto nla_put_failure;
1287 #ifdef IP_SET_HASH_WITH_BITMASK
1288 /* if netmask is set to anything other than HOST_MASK we know that the user supplied netmask
1289 * and not bitmask. These two are mutually exclusive. */
1290 if (h->netmask == HOST_MASK && !nf_inet_addr_cmp(&onesmask, &h->bitmask)) {
1291 if (set->family == NFPROTO_IPV4) {
1292 if (nla_put_ipaddr4(skb, IPSET_ATTR_BITMASK, h->bitmask.ip))
1293 goto nla_put_failure;
1294 } else if (set->family == NFPROTO_IPV6) {
1295 if (nla_put_ipaddr6(skb, IPSET_ATTR_BITMASK, &h->bitmask.in6))
1296 goto nla_put_failure;
1297 }
1298 }
1299 #endif
1300 #ifdef IP_SET_HASH_WITH_NETMASK
1301 if (h->netmask != HOST_MASK && nla_put_u8(skb, IPSET_ATTR_NETMASK, h->netmask))
1302 goto nla_put_failure;
1303 #endif
1304 #ifdef IP_SET_HASH_WITH_MARKMASK
1305 if (nla_put_u32(skb, IPSET_ATTR_MARKMASK, h->markmask))
1306 goto nla_put_failure;
1307 #endif
1308 if (set->flags & IPSET_CREATE_FLAG_BUCKETSIZE) {
1309 if (nla_put_u8(skb, IPSET_ATTR_BUCKETSIZE, h->bucketsize) ||
1310 nla_put_net32(skb, IPSET_ATTR_INITVAL, htonl(h->initval)))
1311 goto nla_put_failure;
1312 }
1313 if (nla_put_net32(skb, IPSET_ATTR_REFERENCES, htonl(set->ref)) ||
1314 nla_put_net32(skb, IPSET_ATTR_MEMSIZE, htonl(memsize)) ||
1315 nla_put_net32(skb, IPSET_ATTR_ELEMENTS, htonl(elements)))
1316 goto nla_put_failure;
1317 if (unlikely(ip_set_put_flags(skb, set)))
1318 goto nla_put_failure;
1319 nla_nest_end(skb, nested);
1320
1321 return 0;
1322 nla_put_failure:
1323 return -EMSGSIZE;
1324 }
1325
1326 /* Make possible to run dumping parallel with resizing */
1327 static void
mtype_uref(struct ip_set * set,struct netlink_callback * cb,bool start)1328 mtype_uref(struct ip_set *set, struct netlink_callback *cb, bool start)
1329 {
1330 struct htype *h = set->data;
1331 struct htable *t;
1332
1333 if (start) {
1334 rcu_read_lock_bh();
1335 t = ipset_dereference_bh_nfnl(h->table);
1336 atomic_inc(&t->uref);
1337 cb->args[IPSET_CB_PRIVATE] = (unsigned long)t;
1338 rcu_read_unlock_bh();
1339 } else if (cb->args[IPSET_CB_PRIVATE]) {
1340 t = (struct htable *)cb->args[IPSET_CB_PRIVATE];
1341 if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1342 pr_debug("Table destroy after resize "
1343 " by dump: %p\n", t);
1344 mtype_ahash_destroy(set, t, false);
1345 }
1346 cb->args[IPSET_CB_PRIVATE] = 0;
1347 }
1348 }
1349
1350 /* Reply a LIST/SAVE request: dump the elements of the specified set */
1351 static int
mtype_list(const struct ip_set * set,struct sk_buff * skb,struct netlink_callback * cb)1352 mtype_list(const struct ip_set *set,
1353 struct sk_buff *skb, struct netlink_callback *cb)
1354 {
1355 const struct htable *t;
1356 struct nlattr *atd, *nested;
1357 const struct hbucket *n;
1358 const struct mtype_elem *e;
1359 u32 first = cb->args[IPSET_CB_ARG0];
1360 /* We assume that one hash bucket fills into one page */
1361 void *incomplete;
1362 int i, ret = 0;
1363
1364 atd = nla_nest_start(skb, IPSET_ATTR_ADT);
1365 if (!atd)
1366 return -EMSGSIZE;
1367
1368 pr_debug("list hash set %s\n", set->name);
1369 t = (const struct htable *)cb->args[IPSET_CB_PRIVATE];
1370 /* Expire may replace a hbucket with another one */
1371 rcu_read_lock();
1372 for (; cb->args[IPSET_CB_ARG0] < jhash_size(t->htable_bits);
1373 cb->args[IPSET_CB_ARG0]++) {
1374 cond_resched_rcu();
1375 incomplete = skb_tail_pointer(skb);
1376 n = rcu_dereference(hbucket(t, cb->args[IPSET_CB_ARG0]));
1377 pr_debug("cb->arg bucket: %lu, t %p n %p\n",
1378 cb->args[IPSET_CB_ARG0], t, n);
1379 if (!n)
1380 continue;
1381 for (i = 0; i < n->pos; i++) {
1382 if (!test_bit(i, n->used))
1383 continue;
1384 e = ahash_data(n, i, set->dsize);
1385 if (SET_ELEM_EXPIRED(set, e))
1386 continue;
1387 pr_debug("list hash %lu hbucket %p i %u, data %p\n",
1388 cb->args[IPSET_CB_ARG0], n, i, e);
1389 nested = nla_nest_start(skb, IPSET_ATTR_DATA);
1390 if (!nested) {
1391 if (cb->args[IPSET_CB_ARG0] == first) {
1392 nla_nest_cancel(skb, atd);
1393 ret = -EMSGSIZE;
1394 goto out;
1395 }
1396 goto nla_put_failure;
1397 }
1398 if (mtype_data_list(skb, e))
1399 goto nla_put_failure;
1400 if (ip_set_put_extensions(skb, set, e, true))
1401 goto nla_put_failure;
1402 nla_nest_end(skb, nested);
1403 }
1404 }
1405 nla_nest_end(skb, atd);
1406 /* Set listing finished */
1407 cb->args[IPSET_CB_ARG0] = 0;
1408
1409 goto out;
1410
1411 nla_put_failure:
1412 nlmsg_trim(skb, incomplete);
1413 if (unlikely(first == cb->args[IPSET_CB_ARG0])) {
1414 pr_warn("Can't list set %s: one bucket does not fit into a message. Please report it!\n",
1415 set->name);
1416 cb->args[IPSET_CB_ARG0] = 0;
1417 ret = -EMSGSIZE;
1418 } else {
1419 nla_nest_end(skb, atd);
1420 }
1421 out:
1422 rcu_read_unlock();
1423 return ret;
1424 }
1425
1426 static int
1427 IPSET_TOKEN(MTYPE, _kadt)(struct ip_set *set, const struct sk_buff *skb,
1428 const struct xt_action_param *par,
1429 enum ipset_adt adt, struct ip_set_adt_opt *opt);
1430
1431 static int
1432 IPSET_TOKEN(MTYPE, _uadt)(struct ip_set *set, struct nlattr *tb[],
1433 enum ipset_adt adt, u32 *lineno, u32 flags,
1434 bool retried);
1435
1436 static const struct ip_set_type_variant mtype_variant = {
1437 .kadt = mtype_kadt,
1438 .uadt = mtype_uadt,
1439 .adt = {
1440 [IPSET_ADD] = mtype_add,
1441 [IPSET_DEL] = mtype_del,
1442 [IPSET_TEST] = mtype_test,
1443 },
1444 .destroy = mtype_destroy,
1445 .flush = mtype_flush,
1446 .head = mtype_head,
1447 .list = mtype_list,
1448 .uref = mtype_uref,
1449 .resize = mtype_resize,
1450 .same_set = mtype_same_set,
1451 .cancel_gc = mtype_cancel_gc,
1452 .region_lock = true,
1453 };
1454
1455 #ifdef IP_SET_EMIT_CREATE
1456 static int
IPSET_TOKEN(HTYPE,_create)1457 IPSET_TOKEN(HTYPE, _create)(struct net *net, struct ip_set *set,
1458 struct nlattr *tb[], u32 flags)
1459 {
1460 u32 hashsize = IPSET_DEFAULT_HASHSIZE, maxelem = IPSET_DEFAULT_MAXELEM;
1461 #ifdef IP_SET_HASH_WITH_MARKMASK
1462 u32 markmask;
1463 #endif
1464 u8 hbits;
1465 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
1466 int ret __attribute__((unused)) = 0;
1467 u8 netmask = set->family == NFPROTO_IPV4 ? 32 : 128;
1468 union nf_inet_addr bitmask = onesmask;
1469 #endif
1470 size_t hsize;
1471 struct htype *h;
1472 struct htable *t;
1473 u32 i;
1474
1475 pr_debug("Create set %s with family %s\n",
1476 set->name, set->family == NFPROTO_IPV4 ? "inet" : "inet6");
1477
1478 #ifdef IP_SET_PROTO_UNDEF
1479 if (set->family != NFPROTO_UNSPEC)
1480 return -IPSET_ERR_INVALID_FAMILY;
1481 #else
1482 if (!(set->family == NFPROTO_IPV4 || set->family == NFPROTO_IPV6))
1483 return -IPSET_ERR_INVALID_FAMILY;
1484 #endif
1485
1486 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_HASHSIZE) ||
1487 !ip_set_optattr_netorder(tb, IPSET_ATTR_MAXELEM) ||
1488 !ip_set_optattr_netorder(tb, IPSET_ATTR_TIMEOUT) ||
1489 !ip_set_optattr_netorder(tb, IPSET_ATTR_CADT_FLAGS)))
1490 return -IPSET_ERR_PROTOCOL;
1491
1492 #ifdef IP_SET_HASH_WITH_MARKMASK
1493 /* Separated condition in order to avoid directive in argument list */
1494 if (unlikely(!ip_set_optattr_netorder(tb, IPSET_ATTR_MARKMASK)))
1495 return -IPSET_ERR_PROTOCOL;
1496
1497 markmask = 0xffffffff;
1498 if (tb[IPSET_ATTR_MARKMASK]) {
1499 markmask = ntohl(nla_get_be32(tb[IPSET_ATTR_MARKMASK]));
1500 if (markmask == 0)
1501 return -IPSET_ERR_INVALID_MARKMASK;
1502 }
1503 #endif
1504
1505 #ifdef IP_SET_HASH_WITH_NETMASK
1506 if (tb[IPSET_ATTR_NETMASK]) {
1507 netmask = nla_get_u8(tb[IPSET_ATTR_NETMASK]);
1508
1509 if ((set->family == NFPROTO_IPV4 && netmask > 32) ||
1510 (set->family == NFPROTO_IPV6 && netmask > 128) ||
1511 netmask == 0)
1512 return -IPSET_ERR_INVALID_NETMASK;
1513
1514 /* we convert netmask to bitmask and store it */
1515 if (set->family == NFPROTO_IPV4)
1516 bitmask.ip = ip_set_netmask(netmask);
1517 else
1518 ip6_netmask(&bitmask, netmask);
1519 }
1520 #endif
1521
1522 #ifdef IP_SET_HASH_WITH_BITMASK
1523 if (tb[IPSET_ATTR_BITMASK]) {
1524 /* bitmask and netmask do the same thing, allow only one of these options */
1525 if (tb[IPSET_ATTR_NETMASK])
1526 return -IPSET_ERR_BITMASK_NETMASK_EXCL;
1527
1528 if (set->family == NFPROTO_IPV4) {
1529 ret = ip_set_get_ipaddr4(tb[IPSET_ATTR_BITMASK], &bitmask.ip);
1530 if (ret || !bitmask.ip)
1531 return -IPSET_ERR_INVALID_NETMASK;
1532 } else if (set->family == NFPROTO_IPV6) {
1533 ret = ip_set_get_ipaddr6(tb[IPSET_ATTR_BITMASK], &bitmask);
1534 if (ret || ipv6_addr_any(&bitmask.in6))
1535 return -IPSET_ERR_INVALID_NETMASK;
1536 }
1537
1538 if (nf_inet_addr_cmp(&bitmask, &zeromask))
1539 return -IPSET_ERR_INVALID_NETMASK;
1540 }
1541 #endif
1542
1543 if (tb[IPSET_ATTR_HASHSIZE]) {
1544 hashsize = ip_set_get_h32(tb[IPSET_ATTR_HASHSIZE]);
1545 if (hashsize < IPSET_MIMINAL_HASHSIZE)
1546 hashsize = IPSET_MIMINAL_HASHSIZE;
1547 }
1548
1549 if (tb[IPSET_ATTR_MAXELEM])
1550 maxelem = ip_set_get_h32(tb[IPSET_ATTR_MAXELEM]);
1551
1552 hsize = sizeof(*h);
1553 h = kzalloc(hsize, GFP_KERNEL);
1554 if (!h)
1555 return -ENOMEM;
1556
1557 /* Compute htable_bits from the user input parameter hashsize.
1558 * Assume that hashsize == 2^htable_bits,
1559 * otherwise round up to the first 2^n value.
1560 */
1561 hbits = fls(hashsize - 1);
1562 hsize = htable_size(hbits);
1563 if (hsize == 0) {
1564 kfree(h);
1565 return -ENOMEM;
1566 }
1567 t = ip_set_alloc(hsize);
1568 if (!t) {
1569 kfree(h);
1570 return -ENOMEM;
1571 }
1572 t->hregion = ip_set_alloc(ahash_sizeof_regions(hbits));
1573 if (!t->hregion) {
1574 ip_set_free(t);
1575 kfree(h);
1576 return -ENOMEM;
1577 }
1578 h->gc.set = set;
1579 for (i = 0; i < ahash_numof_locks(hbits); i++)
1580 spin_lock_init(&t->hregion[i].lock);
1581 h->maxelem = maxelem;
1582 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
1583 h->bitmask = bitmask;
1584 h->netmask = netmask;
1585 #endif
1586 #ifdef IP_SET_HASH_WITH_MARKMASK
1587 h->markmask = markmask;
1588 #endif
1589 if (tb[IPSET_ATTR_INITVAL])
1590 h->initval = ntohl(nla_get_be32(tb[IPSET_ATTR_INITVAL]));
1591 else
1592 get_random_bytes(&h->initval, sizeof(h->initval));
1593 h->bucketsize = AHASH_MAX_SIZE;
1594 if (tb[IPSET_ATTR_BUCKETSIZE]) {
1595 h->bucketsize = nla_get_u8(tb[IPSET_ATTR_BUCKETSIZE]);
1596 if (h->bucketsize < AHASH_INIT_SIZE)
1597 h->bucketsize = AHASH_INIT_SIZE;
1598 else if (h->bucketsize > AHASH_MAX_SIZE)
1599 h->bucketsize = AHASH_MAX_SIZE;
1600 else if (h->bucketsize % 2)
1601 h->bucketsize += 1;
1602 }
1603 t->htable_bits = hbits;
1604 t->maxelem = h->maxelem / ahash_numof_locks(hbits);
1605 RCU_INIT_POINTER(h->table, t);
1606
1607 INIT_LIST_HEAD(&h->ad);
1608 set->data = h;
1609 #ifndef IP_SET_PROTO_UNDEF
1610 if (set->family == NFPROTO_IPV4) {
1611 #endif
1612 set->variant = &IPSET_TOKEN(HTYPE, 4_variant);
1613 set->dsize = ip_set_elem_len(set, tb,
1614 sizeof(struct IPSET_TOKEN(HTYPE, 4_elem)),
1615 __alignof__(struct IPSET_TOKEN(HTYPE, 4_elem)));
1616 #ifndef IP_SET_PROTO_UNDEF
1617 } else {
1618 set->variant = &IPSET_TOKEN(HTYPE, 6_variant);
1619 set->dsize = ip_set_elem_len(set, tb,
1620 sizeof(struct IPSET_TOKEN(HTYPE, 6_elem)),
1621 __alignof__(struct IPSET_TOKEN(HTYPE, 6_elem)));
1622 }
1623 #endif
1624 set->timeout = IPSET_NO_TIMEOUT;
1625 if (tb[IPSET_ATTR_TIMEOUT]) {
1626 set->timeout = ip_set_timeout_uget(tb[IPSET_ATTR_TIMEOUT]);
1627 #ifndef IP_SET_PROTO_UNDEF
1628 if (set->family == NFPROTO_IPV4)
1629 #endif
1630 IPSET_TOKEN(HTYPE, 4_gc_init)(&h->gc);
1631 #ifndef IP_SET_PROTO_UNDEF
1632 else
1633 IPSET_TOKEN(HTYPE, 6_gc_init)(&h->gc);
1634 #endif
1635 }
1636 pr_debug("create %s hashsize %u (%u) maxelem %u: %p(%p)\n",
1637 set->name, jhash_size(t->htable_bits),
1638 t->htable_bits, h->maxelem, set->data, t);
1639
1640 return 0;
1641 }
1642 #endif /* IP_SET_EMIT_CREATE */
1643
1644 #undef HKEY_DATALEN
1645