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/jhash.h>
9 #include <linux/types.h>
10 #include <linux/netfilter/nfnetlink.h>
11 #include <linux/netfilter/ipset/ip_set.h>
12
13 #define __ipset_dereference(p) \
14 rcu_dereference_protected(p, 1)
15 #define ipset_dereference_nfnl(p) \
16 rcu_dereference_protected(p, \
17 lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET))
18 #define ipset_dereference_set(p, set) \
19 rcu_dereference_protected(p, \
20 lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET) || \
21 lockdep_is_held(&(set)->lock))
22 #define ipset_dereference_bh_nfnl(p) \
23 rcu_dereference_bh_check(p, \
24 lockdep_nfnl_is_held(NFNL_SUBSYS_IPSET))
25
26 /* Hashing which uses arrays to resolve clashing. The hash table is resized
27 * (doubled) when searching becomes too long.
28 * Internally jhash is used with the assumption that the size of the
29 * stored data is a multiple of sizeof(u32).
30 *
31 * Readers and resizing
32 *
33 * Resizing can be triggered by userspace command only, and those
34 * are serialized by the nfnl mutex. During resizing the set is
35 * read-locked, so the only possible concurrent operations are
36 * the kernel side readers. Those must be protected by proper RCU locking.
37 */
38
39 /* Number of elements to store in an initial array block */
40 #define AHASH_INIT_SIZE 2
41 /* Max number of elements to store in an array block */
42 #define AHASH_MAX_SIZE (6 * AHASH_INIT_SIZE)
43 /* Max muber of elements in the array block when tuned */
44 #define AHASH_MAX_TUNED 64
45 #define AHASH_MAX(h) ((h)->bucketsize)
46
47 /* A hash bucket */
48 struct hbucket {
49 struct rcu_head rcu; /* for call_rcu */
50 /* Which positions are used in the array */
51 DECLARE_BITMAP(used, AHASH_MAX_TUNED);
52 u8 size; /* size of the array */
53 u8 pos; /* position of the first free entry */
54 unsigned char value[] /* the array of the values */
55 __aligned(__alignof__(u64));
56 };
57
58 /* Region size for locking == 2^HTABLE_REGION_BITS */
59 #define HTABLE_REGION_BITS 10
60 #define ahash_numof_locks(htable_bits) \
61 ((htable_bits) < HTABLE_REGION_BITS ? 1 \
62 : jhash_size((htable_bits) - HTABLE_REGION_BITS))
63 #define ahash_sizeof_regions(htable_bits) \
64 (ahash_numof_locks(htable_bits) * sizeof(struct ip_set_region))
65 #define ahash_region(n, htable_bits) \
66 ((n) % ahash_numof_locks(htable_bits))
67 #define ahash_bucket_start(h, htable_bits) \
68 ((htable_bits) < HTABLE_REGION_BITS ? 0 \
69 : (h) * jhash_size(HTABLE_REGION_BITS))
70 #define ahash_bucket_end(h, htable_bits) \
71 ((htable_bits) < HTABLE_REGION_BITS ? jhash_size(htable_bits) \
72 : ((h) + 1) * jhash_size(HTABLE_REGION_BITS))
73
74 struct htable_gc {
75 struct delayed_work dwork;
76 struct ip_set *set; /* Set the gc belongs to */
77 u32 region; /* Last gc run position */
78 };
79
80 /* The hash table: the table size stored here in order to make resizing easy */
81 struct htable {
82 atomic_t ref; /* References for resizing */
83 atomic_t uref; /* References for dumping and gc */
84 u8 htable_bits; /* size of hash table == 2^htable_bits */
85 u32 maxelem; /* Maxelem per region */
86 struct ip_set_region *hregion; /* Region locks and ext sizes */
87 struct hbucket __rcu *bucket[]; /* hashtable buckets */
88 };
89
90 #define hbucket(h, i) ((h)->bucket[i])
91 #define ext_size(n, dsize) \
92 (sizeof(struct hbucket) + (n) * (dsize))
93
94 #ifndef IPSET_NET_COUNT
95 #define IPSET_NET_COUNT 1
96 #endif
97
98 /* Book-keeping of the prefixes added to the set */
99 struct net_prefixes {
100 u32 nets[IPSET_NET_COUNT]; /* number of elements for this cidr */
101 u8 cidr[IPSET_NET_COUNT]; /* the cidr value */
102 };
103
104 /* Compute the hash table size */
105 static size_t
htable_size(u8 hbits)106 htable_size(u8 hbits)
107 {
108 size_t hsize;
109
110 /* We must fit both into u32 in jhash and INT_MAX in kvmalloc_node() */
111 if (hbits > 31)
112 return 0;
113 hsize = jhash_size(hbits);
114 if ((INT_MAX - sizeof(struct htable)) / sizeof(struct hbucket *)
115 < hsize)
116 return 0;
117
118 return hsize * sizeof(struct hbucket *) + sizeof(struct htable);
119 }
120
121 #ifdef IP_SET_HASH_WITH_NETS
122 #if IPSET_NET_COUNT > 1
123 #define __CIDR(cidr, i) (cidr[i])
124 #else
125 #define __CIDR(cidr, i) (cidr)
126 #endif
127
128 /* cidr + 1 is stored in net_prefixes to support /0 */
129 #define NCIDR_PUT(cidr) ((cidr) + 1)
130 #define NCIDR_GET(cidr) ((cidr) - 1)
131
132 #ifdef IP_SET_HASH_WITH_NETS_PACKED
133 /* When cidr is packed with nomatch, cidr - 1 is stored in the data entry */
134 #define DCIDR_PUT(cidr) ((cidr) - 1)
135 #define DCIDR_GET(cidr, i) (__CIDR(cidr, i) + 1)
136 #else
137 #define DCIDR_PUT(cidr) (cidr)
138 #define DCIDR_GET(cidr, i) __CIDR(cidr, i)
139 #endif
140
141 #define INIT_CIDR(cidr, host_mask) \
142 DCIDR_PUT(((cidr) ? NCIDR_GET(cidr) : host_mask))
143
144 #ifdef IP_SET_HASH_WITH_NET0
145 /* cidr from 0 to HOST_MASK value and c = cidr + 1 */
146 #define NLEN (HOST_MASK + 1)
147 #define CIDR_POS(c) ((c) - 1)
148 #else
149 /* cidr from 1 to HOST_MASK value and c = cidr + 1 */
150 #define NLEN HOST_MASK
151 #define CIDR_POS(c) ((c) - 2)
152 #endif
153
154 #else
155 #define NLEN 0
156 #endif /* IP_SET_HASH_WITH_NETS */
157
158 #define SET_ELEM_EXPIRED(set, d) \
159 (SET_WITH_TIMEOUT(set) && \
160 ip_set_timeout_expired(ext_timeout(d, set)))
161
162 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
163 static const union nf_inet_addr onesmask = {
164 .all[0] = 0xffffffff,
165 .all[1] = 0xffffffff,
166 .all[2] = 0xffffffff,
167 .all[3] = 0xffffffff
168 };
169
170 static const union nf_inet_addr zeromask = {};
171 #endif
172
173 #endif /* _IP_SET_HASH_GEN_H */
174
175 #ifndef MTYPE
176 #error "MTYPE is not defined!"
177 #endif
178
179 #ifndef HTYPE
180 #error "HTYPE is not defined!"
181 #endif
182
183 #ifndef HOST_MASK
184 #error "HOST_MASK is not defined!"
185 #endif
186
187 /* Family dependent templates */
188
189 #undef ahash_data
190 #undef mtype_data_equal
191 #undef mtype_do_data_match
192 #undef mtype_data_set_flags
193 #undef mtype_data_reset_elem
194 #undef mtype_data_reset_flags
195 #undef mtype_data_netmask
196 #undef mtype_data_list
197 #undef mtype_data_next
198 #undef mtype_elem
199
200 #undef mtype_ahash_destroy
201 #undef mtype_ext_cleanup
202 #undef mtype_add_cidr
203 #undef mtype_del_cidr
204 #undef mtype_ahash_memsize
205 #undef mtype_flush
206 #undef mtype_destroy
207 #undef mtype_same_set
208 #undef mtype_kadt
209 #undef mtype_uadt
210
211 #undef mtype_add
212 #undef mtype_del
213 #undef mtype_test_cidrs
214 #undef mtype_test
215 #undef mtype_uref
216 #undef mtype_resize
217 #undef mtype_ext_size
218 #undef mtype_resize_ad
219 #undef mtype_head
220 #undef mtype_list
221 #undef mtype_gc_do
222 #undef mtype_gc
223 #undef mtype_gc_init
224 #undef mtype_cancel_gc
225 #undef mtype_variant
226 #undef mtype_data_match
227
228 #undef htype
229 #undef HKEY
230
231 #define mtype_data_equal IPSET_TOKEN(MTYPE, _data_equal)
232 #ifdef IP_SET_HASH_WITH_NETS
233 #define mtype_do_data_match IPSET_TOKEN(MTYPE, _do_data_match)
234 #else
235 #define mtype_do_data_match(d) 1
236 #endif
237 #define mtype_data_set_flags IPSET_TOKEN(MTYPE, _data_set_flags)
238 #define mtype_data_reset_elem IPSET_TOKEN(MTYPE, _data_reset_elem)
239 #define mtype_data_reset_flags IPSET_TOKEN(MTYPE, _data_reset_flags)
240 #define mtype_data_netmask IPSET_TOKEN(MTYPE, _data_netmask)
241 #define mtype_data_list IPSET_TOKEN(MTYPE, _data_list)
242 #define mtype_data_next IPSET_TOKEN(MTYPE, _data_next)
243 #define mtype_elem IPSET_TOKEN(MTYPE, _elem)
244
245 #define mtype_ahash_destroy IPSET_TOKEN(MTYPE, _ahash_destroy)
246 #define mtype_ext_cleanup IPSET_TOKEN(MTYPE, _ext_cleanup)
247 #define mtype_add_cidr IPSET_TOKEN(MTYPE, _add_cidr)
248 #define mtype_del_cidr IPSET_TOKEN(MTYPE, _del_cidr)
249 #define mtype_ahash_memsize IPSET_TOKEN(MTYPE, _ahash_memsize)
250 #define mtype_flush IPSET_TOKEN(MTYPE, _flush)
251 #define mtype_destroy IPSET_TOKEN(MTYPE, _destroy)
252 #define mtype_same_set IPSET_TOKEN(MTYPE, _same_set)
253 #define mtype_kadt IPSET_TOKEN(MTYPE, _kadt)
254 #define mtype_uadt IPSET_TOKEN(MTYPE, _uadt)
255
256 #define mtype_add IPSET_TOKEN(MTYPE, _add)
257 #define mtype_del IPSET_TOKEN(MTYPE, _del)
258 #define mtype_test_cidrs IPSET_TOKEN(MTYPE, _test_cidrs)
259 #define mtype_test IPSET_TOKEN(MTYPE, _test)
260 #define mtype_uref IPSET_TOKEN(MTYPE, _uref)
261 #define mtype_resize IPSET_TOKEN(MTYPE, _resize)
262 #define mtype_ext_size IPSET_TOKEN(MTYPE, _ext_size)
263 #define mtype_resize_ad IPSET_TOKEN(MTYPE, _resize_ad)
264 #define mtype_head IPSET_TOKEN(MTYPE, _head)
265 #define mtype_list IPSET_TOKEN(MTYPE, _list)
266 #define mtype_gc_do IPSET_TOKEN(MTYPE, _gc_do)
267 #define mtype_gc IPSET_TOKEN(MTYPE, _gc)
268 #define mtype_gc_init IPSET_TOKEN(MTYPE, _gc_init)
269 #define mtype_cancel_gc IPSET_TOKEN(MTYPE, _cancel_gc)
270 #define mtype_variant IPSET_TOKEN(MTYPE, _variant)
271 #define mtype_data_match IPSET_TOKEN(MTYPE, _data_match)
272
273 #ifndef HKEY_DATALEN
274 #define HKEY_DATALEN sizeof(struct mtype_elem)
275 #endif
276
277 #define htype MTYPE
278
279 #define HKEY(data, initval, htable_bits) \
280 ({ \
281 const u32 *__k = (const u32 *)data; \
282 u32 __l = HKEY_DATALEN / sizeof(u32); \
283 \
284 BUILD_BUG_ON(HKEY_DATALEN % sizeof(u32) != 0); \
285 \
286 jhash2(__k, __l, initval) & jhash_mask(htable_bits); \
287 })
288
289 /* The generic hash structure */
290 struct htype {
291 struct htable __rcu *table; /* the hash table */
292 struct htable_gc gc; /* gc workqueue */
293 u32 maxelem; /* max elements in the hash */
294 u32 initval; /* random jhash init value */
295 #ifdef IP_SET_HASH_WITH_MARKMASK
296 u32 markmask; /* markmask value for mark mask to store */
297 #endif
298 u8 bucketsize; /* max elements in an array block */
299 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
300 u8 netmask; /* netmask value for subnets to store */
301 union nf_inet_addr bitmask; /* stores bitmask */
302 #endif
303 struct list_head ad; /* Resize add|del backlist */
304 struct mtype_elem next; /* temporary storage for uadd */
305 #ifdef IP_SET_HASH_WITH_NETS
306 struct net_prefixes nets[NLEN]; /* book-keeping of prefixes */
307 #endif
308 };
309
310 /* ADD|DEL entries saved during resize */
311 struct mtype_resize_ad {
312 struct list_head list;
313 enum ipset_adt ad; /* ADD|DEL element */
314 struct mtype_elem d; /* Element value */
315 struct ip_set_ext ext; /* Extensions for ADD */
316 struct ip_set_ext mext; /* Target extensions for ADD */
317 u32 flags; /* Flags for ADD */
318 };
319
320 #ifdef IP_SET_HASH_WITH_NETS
321 /* Network cidr size book keeping when the hash stores different
322 * sized networks. cidr == real cidr + 1 to support /0.
323 */
324 static void
mtype_add_cidr(struct ip_set * set,struct htype * h,u8 cidr,u8 n)325 mtype_add_cidr(struct ip_set *set, struct htype *h, u8 cidr, u8 n)
326 {
327 int i, j;
328
329 spin_lock_bh(&set->lock);
330 /* Add in increasing prefix order, so larger cidr first */
331 for (i = 0, j = -1; i < NLEN && h->nets[i].cidr[n]; i++) {
332 if (j != -1) {
333 continue;
334 } else if (h->nets[i].cidr[n] < cidr) {
335 j = i;
336 } else if (h->nets[i].cidr[n] == cidr) {
337 h->nets[CIDR_POS(cidr)].nets[n]++;
338 goto unlock;
339 }
340 }
341 if (j != -1) {
342 for (; i > j; i--)
343 h->nets[i].cidr[n] = h->nets[i - 1].cidr[n];
344 }
345 h->nets[i].cidr[n] = cidr;
346 h->nets[CIDR_POS(cidr)].nets[n] = 1;
347 unlock:
348 spin_unlock_bh(&set->lock);
349 }
350
351 static void
mtype_del_cidr(struct ip_set * set,struct htype * h,u8 cidr,u8 n)352 mtype_del_cidr(struct ip_set *set, struct htype *h, u8 cidr, u8 n)
353 {
354 u8 i, j, net_end = NLEN - 1;
355
356 spin_lock_bh(&set->lock);
357 for (i = 0; i < NLEN; i++) {
358 if (h->nets[i].cidr[n] != cidr)
359 continue;
360 h->nets[CIDR_POS(cidr)].nets[n]--;
361 if (h->nets[CIDR_POS(cidr)].nets[n] > 0)
362 goto unlock;
363 for (j = i; j < net_end && h->nets[j].cidr[n]; j++)
364 h->nets[j].cidr[n] = h->nets[j + 1].cidr[n];
365 h->nets[j].cidr[n] = 0;
366 goto unlock;
367 }
368 unlock:
369 spin_unlock_bh(&set->lock);
370 }
371 #endif
372
373 /* Calculate the actual memory size of the set data */
374 static size_t
mtype_ahash_memsize(const struct htype * h,const struct htable * t)375 mtype_ahash_memsize(const struct htype *h, const struct htable *t)
376 {
377 return sizeof(*h) + sizeof(*t) + ahash_sizeof_regions(t->htable_bits);
378 }
379
380 /* Get the ith element from the array block n */
381 #define ahash_data(n, i, dsize) \
382 ((struct mtype_elem *)((n)->value + ((i) * (dsize))))
383
384 static void
mtype_ext_cleanup(struct ip_set * set,struct hbucket * n)385 mtype_ext_cleanup(struct ip_set *set, struct hbucket *n)
386 {
387 int i;
388
389 for (i = 0; i < n->pos; i++)
390 if (test_bit(i, n->used))
391 ip_set_ext_destroy(set, ahash_data(n, i, set->dsize));
392 }
393
394 /* Flush a hash type of set: destroy all elements */
395 static void
mtype_flush(struct ip_set * set)396 mtype_flush(struct ip_set *set)
397 {
398 struct htype *h = set->data;
399 struct htable *t;
400 struct hbucket *n;
401 u32 r, i;
402
403 t = ipset_dereference_nfnl(h->table);
404 for (r = 0; r < ahash_numof_locks(t->htable_bits); r++) {
405 spin_lock_bh(&t->hregion[r].lock);
406 for (i = ahash_bucket_start(r, t->htable_bits);
407 i < ahash_bucket_end(r, t->htable_bits); i++) {
408 n = __ipset_dereference(hbucket(t, i));
409 if (!n)
410 continue;
411 if (set->extensions & IPSET_EXT_DESTROY)
412 mtype_ext_cleanup(set, n);
413 /* FIXME: use slab cache */
414 rcu_assign_pointer(hbucket(t, i), NULL);
415 kfree_rcu(n, rcu);
416 }
417 t->hregion[r].ext_size = 0;
418 t->hregion[r].elements = 0;
419 spin_unlock_bh(&t->hregion[r].lock);
420 }
421 #ifdef IP_SET_HASH_WITH_NETS
422 memset(h->nets, 0, sizeof(h->nets));
423 #endif
424 }
425
426 /* Destroy the hashtable part of the set */
427 static void
mtype_ahash_destroy(struct ip_set * set,struct htable * t,bool ext_destroy)428 mtype_ahash_destroy(struct ip_set *set, struct htable *t, bool ext_destroy)
429 {
430 struct hbucket *n;
431 u32 i;
432
433 for (i = 0; i < jhash_size(t->htable_bits); i++) {
434 n = (__force struct hbucket *)hbucket(t, i);
435 if (!n)
436 continue;
437 if (set->extensions & IPSET_EXT_DESTROY && ext_destroy)
438 mtype_ext_cleanup(set, n);
439 /* FIXME: use slab cache */
440 kfree(n);
441 }
442
443 ip_set_free(t->hregion);
444 ip_set_free(t);
445 }
446
447 /* Destroy a hash type of set */
448 static void
mtype_destroy(struct ip_set * set)449 mtype_destroy(struct ip_set *set)
450 {
451 struct htype *h = set->data;
452 struct list_head *l, *lt;
453
454 mtype_ahash_destroy(set, (__force struct htable *)h->table, true);
455 list_for_each_safe(l, lt, &h->ad) {
456 list_del(l);
457 kfree(l);
458 }
459 kfree(h);
460
461 set->data = NULL;
462 }
463
464 static bool
mtype_same_set(const struct ip_set * a,const struct ip_set * b)465 mtype_same_set(const struct ip_set *a, const struct ip_set *b)
466 {
467 const struct htype *x = a->data;
468 const struct htype *y = b->data;
469
470 /* Resizing changes htable_bits, so we ignore it */
471 return x->maxelem == y->maxelem &&
472 a->timeout == b->timeout &&
473 #if defined(IP_SET_HASH_WITH_NETMASK) || defined(IP_SET_HASH_WITH_BITMASK)
474 nf_inet_addr_cmp(&x->bitmask, &y->bitmask) &&
475 #endif
476 #ifdef IP_SET_HASH_WITH_MARKMASK
477 x->markmask == y->markmask &&
478 #endif
479 a->extensions == b->extensions;
480 }
481
482 static void
mtype_gc_do(struct ip_set * set,struct htype * h,struct htable * t,u32 r)483 mtype_gc_do(struct ip_set *set, struct htype *h, struct htable *t, u32 r)
484 {
485 struct hbucket *n, *tmp;
486 struct mtype_elem *data;
487 u32 i, j, d;
488 size_t dsize = set->dsize;
489 #ifdef IP_SET_HASH_WITH_NETS
490 u8 k;
491 #endif
492 u8 htable_bits = t->htable_bits;
493
494 spin_lock_bh(&t->hregion[r].lock);
495 for (i = ahash_bucket_start(r, htable_bits);
496 i < ahash_bucket_end(r, htable_bits); i++) {
497 n = __ipset_dereference(hbucket(t, i));
498 if (!n)
499 continue;
500 for (j = 0, d = 0; j < n->pos; j++) {
501 if (!test_bit(j, n->used)) {
502 d++;
503 continue;
504 }
505 data = ahash_data(n, j, dsize);
506 if (!ip_set_timeout_expired(ext_timeout(data, set)))
507 continue;
508 pr_debug("expired %u/%u\n", i, j);
509 clear_bit(j, n->used);
510 smp_mb__after_atomic();
511 #ifdef IP_SET_HASH_WITH_NETS
512 for (k = 0; k < IPSET_NET_COUNT; k++)
513 mtype_del_cidr(set, h,
514 NCIDR_PUT(DCIDR_GET(data->cidr, k)),
515 k);
516 #endif
517 t->hregion[r].elements--;
518 ip_set_ext_destroy(set, data);
519 d++;
520 }
521 if (d >= AHASH_INIT_SIZE) {
522 if (d >= n->size) {
523 t->hregion[r].ext_size -=
524 ext_size(n->size, dsize);
525 rcu_assign_pointer(hbucket(t, i), NULL);
526 kfree_rcu(n, rcu);
527 continue;
528 }
529 tmp = kzalloc(sizeof(*tmp) +
530 (n->size - AHASH_INIT_SIZE) * dsize,
531 GFP_ATOMIC);
532 if (!tmp)
533 /* Still try to delete expired elements. */
534 continue;
535 tmp->size = n->size - AHASH_INIT_SIZE;
536 for (j = 0, d = 0; j < n->pos; j++) {
537 if (!test_bit(j, n->used))
538 continue;
539 data = ahash_data(n, j, dsize);
540 memcpy(tmp->value + d * dsize,
541 data, dsize);
542 set_bit(d, tmp->used);
543 d++;
544 }
545 tmp->pos = d;
546 t->hregion[r].ext_size -=
547 ext_size(AHASH_INIT_SIZE, dsize);
548 rcu_assign_pointer(hbucket(t, i), tmp);
549 kfree_rcu(n, rcu);
550 }
551 }
552 spin_unlock_bh(&t->hregion[r].lock);
553 }
554
555 static void
mtype_gc(struct work_struct * work)556 mtype_gc(struct work_struct *work)
557 {
558 struct htable_gc *gc;
559 struct ip_set *set;
560 struct htype *h;
561 struct htable *t;
562 u32 r, numof_locks;
563 unsigned int next_run;
564
565 gc = container_of(work, struct htable_gc, dwork.work);
566 set = gc->set;
567 h = set->data;
568
569 spin_lock_bh(&set->lock);
570 t = ipset_dereference_set(h->table, set);
571 atomic_inc(&t->uref);
572 numof_locks = ahash_numof_locks(t->htable_bits);
573 r = gc->region++;
574 if (r >= numof_locks) {
575 r = gc->region = 0;
576 }
577 next_run = (IPSET_GC_PERIOD(set->timeout) * HZ) / numof_locks;
578 if (next_run < HZ/10)
579 next_run = HZ/10;
580 spin_unlock_bh(&set->lock);
581
582 mtype_gc_do(set, h, t, r);
583
584 if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
585 pr_debug("Table destroy after resize by expire: %p\n", t);
586 mtype_ahash_destroy(set, t, false);
587 }
588
589 queue_delayed_work(system_power_efficient_wq, &gc->dwork, next_run);
590
591 }
592
593 static void
mtype_gc_init(struct htable_gc * gc)594 mtype_gc_init(struct htable_gc *gc)
595 {
596 INIT_DEFERRABLE_WORK(&gc->dwork, mtype_gc);
597 queue_delayed_work(system_power_efficient_wq, &gc->dwork, HZ);
598 }
599
600 static void
mtype_cancel_gc(struct ip_set * set)601 mtype_cancel_gc(struct ip_set *set)
602 {
603 struct htype *h = set->data;
604
605 if (SET_WITH_TIMEOUT(set))
606 cancel_delayed_work_sync(&h->gc.dwork);
607 }
608
609 static int
610 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
611 struct ip_set_ext *mext, u32 flags);
612 static int
613 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
614 struct ip_set_ext *mext, u32 flags);
615
616 /* Resize a hash: create a new hash table with doubling the hashsize
617 * and inserting the elements to it. Repeat until we succeed or
618 * fail due to memory pressures.
619 */
620 static int
mtype_resize(struct ip_set * set,bool retried)621 mtype_resize(struct ip_set *set, bool retried)
622 {
623 struct htype *h = set->data;
624 struct htable *t, *orig;
625 u8 htable_bits;
626 size_t hsize, dsize = set->dsize;
627 #ifdef IP_SET_HASH_WITH_NETS
628 u8 flags;
629 struct mtype_elem *tmp;
630 #endif
631 struct mtype_elem *data;
632 struct mtype_elem *d;
633 struct hbucket *n, *m;
634 struct list_head *l, *lt;
635 struct mtype_resize_ad *x;
636 u32 i, j, r, nr, key;
637 int ret;
638
639 #ifdef IP_SET_HASH_WITH_NETS
640 tmp = kmalloc(dsize, GFP_KERNEL);
641 if (!tmp)
642 return -ENOMEM;
643 #endif
644 orig = ipset_dereference_bh_nfnl(h->table);
645 htable_bits = orig->htable_bits;
646
647 retry:
648 ret = 0;
649 htable_bits++;
650 if (!htable_bits)
651 goto hbwarn;
652 hsize = htable_size(htable_bits);
653 if (!hsize)
654 goto hbwarn;
655 t = ip_set_alloc(hsize);
656 if (!t) {
657 ret = -ENOMEM;
658 goto out;
659 }
660 t->hregion = ip_set_alloc(ahash_sizeof_regions(htable_bits));
661 if (!t->hregion) {
662 ip_set_free(t);
663 ret = -ENOMEM;
664 goto out;
665 }
666 t->htable_bits = htable_bits;
667 t->maxelem = h->maxelem / ahash_numof_locks(htable_bits);
668 for (i = 0; i < ahash_numof_locks(htable_bits); i++)
669 spin_lock_init(&t->hregion[i].lock);
670
671 /* There can't be another parallel resizing,
672 * but dumping, gc, kernel side add/del are possible
673 */
674 orig = ipset_dereference_bh_nfnl(h->table);
675 atomic_set(&orig->ref, 1);
676 atomic_inc(&orig->uref);
677 pr_debug("attempt to resize set %s from %u to %u, t %p\n",
678 set->name, orig->htable_bits, htable_bits, orig);
679 for (r = 0; r < ahash_numof_locks(orig->htable_bits); r++) {
680 /* Expire may replace a hbucket with another one */
681 rcu_read_lock_bh();
682 for (i = ahash_bucket_start(r, orig->htable_bits);
683 i < ahash_bucket_end(r, orig->htable_bits); i++) {
684 n = __ipset_dereference(hbucket(orig, i));
685 if (!n)
686 continue;
687 for (j = 0; j < n->pos; j++) {
688 if (!test_bit(j, n->used))
689 continue;
690 data = ahash_data(n, j, dsize);
691 if (SET_ELEM_EXPIRED(set, data))
692 continue;
693 #ifdef IP_SET_HASH_WITH_NETS
694 /* We have readers running parallel with us,
695 * so the live data cannot be modified.
696 */
697 flags = 0;
698 memcpy(tmp, data, dsize);
699 data = tmp;
700 mtype_data_reset_flags(data, &flags);
701 #endif
702 key = HKEY(data, h->initval, htable_bits);
703 m = __ipset_dereference(hbucket(t, key));
704 nr = ahash_region(key, htable_bits);
705 if (!m) {
706 m = kzalloc(sizeof(*m) +
707 AHASH_INIT_SIZE * dsize,
708 GFP_ATOMIC);
709 if (!m) {
710 ret = -ENOMEM;
711 goto cleanup;
712 }
713 m->size = AHASH_INIT_SIZE;
714 t->hregion[nr].ext_size +=
715 ext_size(AHASH_INIT_SIZE,
716 dsize);
717 RCU_INIT_POINTER(hbucket(t, key), m);
718 } else if (m->pos >= m->size) {
719 struct hbucket *ht;
720
721 if (m->size >= AHASH_MAX(h)) {
722 ret = -EAGAIN;
723 } else {
724 ht = kzalloc(sizeof(*ht) +
725 (m->size + AHASH_INIT_SIZE)
726 * dsize,
727 GFP_ATOMIC);
728 if (!ht)
729 ret = -ENOMEM;
730 }
731 if (ret < 0)
732 goto cleanup;
733 memcpy(ht, m, sizeof(struct hbucket) +
734 m->size * dsize);
735 ht->size = m->size + AHASH_INIT_SIZE;
736 t->hregion[nr].ext_size +=
737 ext_size(AHASH_INIT_SIZE,
738 dsize);
739 kfree(m);
740 m = ht;
741 RCU_INIT_POINTER(hbucket(t, key), ht);
742 }
743 d = ahash_data(m, m->pos, dsize);
744 memcpy(d, data, dsize);
745 set_bit(m->pos++, m->used);
746 t->hregion[nr].elements++;
747 #ifdef IP_SET_HASH_WITH_NETS
748 mtype_data_reset_flags(d, &flags);
749 #endif
750 }
751 }
752 rcu_read_unlock_bh();
753 }
754
755 /* There can't be any other writer. */
756 rcu_assign_pointer(h->table, t);
757
758 /* Give time to other readers of the set */
759 synchronize_rcu();
760
761 pr_debug("set %s resized from %u (%p) to %u (%p)\n", set->name,
762 orig->htable_bits, orig, t->htable_bits, t);
763 /* Add/delete elements processed by the SET target during resize.
764 * Kernel-side add cannot trigger a resize and userspace actions
765 * are serialized by the mutex.
766 */
767 list_for_each_safe(l, lt, &h->ad) {
768 x = list_entry(l, struct mtype_resize_ad, list);
769 if (x->ad == IPSET_ADD) {
770 mtype_add(set, &x->d, &x->ext, &x->mext, x->flags);
771 } else {
772 mtype_del(set, &x->d, NULL, NULL, 0);
773 }
774 list_del(l);
775 kfree(l);
776 }
777 /* If there's nobody else using the table, destroy it */
778 if (atomic_dec_and_test(&orig->uref)) {
779 pr_debug("Table destroy by resize %p\n", orig);
780 mtype_ahash_destroy(set, orig, false);
781 }
782
783 out:
784 #ifdef IP_SET_HASH_WITH_NETS
785 kfree(tmp);
786 #endif
787 return ret;
788
789 cleanup:
790 rcu_read_unlock_bh();
791 atomic_set(&orig->ref, 0);
792 atomic_dec(&orig->uref);
793 mtype_ahash_destroy(set, t, false);
794 if (ret == -EAGAIN)
795 goto retry;
796 goto out;
797
798 hbwarn:
799 /* In case we have plenty of memory :-) */
800 pr_warn("Cannot increase the hashsize of set %s further\n", set->name);
801 ret = -IPSET_ERR_HASH_FULL;
802 goto out;
803 }
804
805 /* Get the current number of elements and ext_size in the set */
806 static void
mtype_ext_size(struct ip_set * set,u32 * elements,size_t * ext_size)807 mtype_ext_size(struct ip_set *set, u32 *elements, size_t *ext_size)
808 {
809 struct htype *h = set->data;
810 const struct htable *t;
811 u32 i, j, r;
812 struct hbucket *n;
813 struct mtype_elem *data;
814
815 t = rcu_dereference_bh(h->table);
816 for (r = 0; r < ahash_numof_locks(t->htable_bits); r++) {
817 for (i = ahash_bucket_start(r, t->htable_bits);
818 i < ahash_bucket_end(r, t->htable_bits); i++) {
819 n = rcu_dereference_bh(hbucket(t, i));
820 if (!n)
821 continue;
822 for (j = 0; j < n->pos; j++) {
823 if (!test_bit(j, n->used))
824 continue;
825 data = ahash_data(n, j, set->dsize);
826 if (!SET_ELEM_EXPIRED(set, data))
827 (*elements)++;
828 }
829 }
830 *ext_size += t->hregion[r].ext_size;
831 }
832 }
833
834 /* Add an element to a hash and update the internal counters when succeeded,
835 * otherwise report the proper error code.
836 */
837 static int
mtype_add(struct ip_set * set,void * value,const struct ip_set_ext * ext,struct ip_set_ext * mext,u32 flags)838 mtype_add(struct ip_set *set, void *value, const struct ip_set_ext *ext,
839 struct ip_set_ext *mext, u32 flags)
840 {
841 struct htype *h = set->data;
842 struct htable *t;
843 const struct mtype_elem *d = value;
844 struct mtype_elem *data;
845 struct hbucket *n, *old = ERR_PTR(-ENOENT);
846 int i, j = -1, ret;
847 bool flag_exist = flags & IPSET_FLAG_EXIST;
848 bool deleted = false, forceadd = false, reuse = false;
849 u32 r, key, multi = 0, elements, maxelem;
850
851 rcu_read_lock_bh();
852 t = rcu_dereference_bh(h->table);
853 key = HKEY(value, h->initval, t->htable_bits);
854 r = ahash_region(key, t->htable_bits);
855 atomic_inc(&t->uref);
856 elements = t->hregion[r].elements;
857 maxelem = t->maxelem;
858 if (elements >= maxelem) {
859 u32 e;
860 if (SET_WITH_TIMEOUT(set)) {
861 rcu_read_unlock_bh();
862 mtype_gc_do(set, h, t, r);
863 rcu_read_lock_bh();
864 }
865 maxelem = h->maxelem;
866 elements = 0;
867 for (e = 0; e < ahash_numof_locks(t->htable_bits); e++)
868 elements += t->hregion[e].elements;
869 if (elements >= maxelem && SET_WITH_FORCEADD(set))
870 forceadd = true;
871 }
872 rcu_read_unlock_bh();
873
874 spin_lock_bh(&t->hregion[r].lock);
875 n = rcu_dereference_bh(hbucket(t, key));
876 if (!n) {
877 if (forceadd || elements >= maxelem)
878 goto set_full;
879 old = NULL;
880 n = kzalloc(sizeof(*n) + AHASH_INIT_SIZE * set->dsize,
881 GFP_ATOMIC);
882 if (!n) {
883 ret = -ENOMEM;
884 goto unlock;
885 }
886 n->size = AHASH_INIT_SIZE;
887 t->hregion[r].ext_size +=
888 ext_size(AHASH_INIT_SIZE, set->dsize);
889 goto copy_elem;
890 }
891 for (i = 0; i < n->pos; i++) {
892 if (!test_bit(i, n->used)) {
893 /* Reuse first deleted entry */
894 if (j == -1) {
895 deleted = reuse = true;
896 j = i;
897 }
898 continue;
899 }
900 data = ahash_data(n, i, set->dsize);
901 if (mtype_data_equal(data, d, &multi)) {
902 if (flag_exist || SET_ELEM_EXPIRED(set, data)) {
903 /* Just the extensions could be overwritten */
904 j = i;
905 goto overwrite_extensions;
906 }
907 ret = -IPSET_ERR_EXIST;
908 goto unlock;
909 }
910 /* Reuse first timed out entry */
911 if (SET_ELEM_EXPIRED(set, data) && j == -1) {
912 j = i;
913 reuse = true;
914 }
915 }
916 if (reuse || forceadd) {
917 if (j == -1)
918 j = 0;
919 data = ahash_data(n, j, set->dsize);
920 if (!deleted) {
921 #ifdef IP_SET_HASH_WITH_NETS
922 for (i = 0; i < IPSET_NET_COUNT; i++)
923 mtype_del_cidr(set, h,
924 NCIDR_PUT(DCIDR_GET(data->cidr, i)),
925 i);
926 #endif
927 ip_set_ext_destroy(set, data);
928 t->hregion[r].elements--;
929 }
930 goto copy_data;
931 }
932 if (elements >= maxelem)
933 goto set_full;
934 /* Create a new slot */
935 if (n->pos >= n->size) {
936 #ifdef IP_SET_HASH_WITH_MULTI
937 if (h->bucketsize >= AHASH_MAX_TUNED)
938 goto set_full;
939 else if (h->bucketsize <= multi)
940 h->bucketsize += AHASH_INIT_SIZE;
941 #endif
942 if (n->size >= AHASH_MAX(h)) {
943 /* Trigger rehashing */
944 mtype_data_next(&h->next, d);
945 ret = -EAGAIN;
946 goto resize;
947 }
948 old = n;
949 n = kzalloc(sizeof(*n) +
950 (old->size + AHASH_INIT_SIZE) * set->dsize,
951 GFP_ATOMIC);
952 if (!n) {
953 ret = -ENOMEM;
954 goto unlock;
955 }
956 memcpy(n, old, sizeof(struct hbucket) +
957 old->size * set->dsize);
958 n->size = old->size + AHASH_INIT_SIZE;
959 t->hregion[r].ext_size +=
960 ext_size(AHASH_INIT_SIZE, set->dsize);
961 }
962
963 copy_elem:
964 j = n->pos++;
965 data = ahash_data(n, j, set->dsize);
966 copy_data:
967 t->hregion[r].elements++;
968 #ifdef IP_SET_HASH_WITH_NETS
969 for (i = 0; i < IPSET_NET_COUNT; i++)
970 mtype_add_cidr(set, h, NCIDR_PUT(DCIDR_GET(d->cidr, i)), i);
971 #endif
972 memcpy(data, d, sizeof(struct mtype_elem));
973 overwrite_extensions:
974 #ifdef IP_SET_HASH_WITH_NETS
975 mtype_data_set_flags(data, flags);
976 #endif
977 if (SET_WITH_COUNTER(set))
978 ip_set_init_counter(ext_counter(data, set), ext);
979 if (SET_WITH_COMMENT(set))
980 ip_set_init_comment(set, ext_comment(data, set), ext);
981 if (SET_WITH_SKBINFO(set))
982 ip_set_init_skbinfo(ext_skbinfo(data, set), ext);
983 /* Must come last for the case when timed out entry is reused */
984 if (SET_WITH_TIMEOUT(set))
985 ip_set_timeout_set(ext_timeout(data, set), ext->timeout);
986 smp_mb__before_atomic();
987 set_bit(j, n->used);
988 if (old != ERR_PTR(-ENOENT)) {
989 rcu_assign_pointer(hbucket(t, key), n);
990 if (old)
991 kfree_rcu(old, rcu);
992 }
993 ret = 0;
994 resize:
995 spin_unlock_bh(&t->hregion[r].lock);
996 if (atomic_read(&t->ref) && ext->target) {
997 /* Resize is in process and kernel side add, save values */
998 struct mtype_resize_ad *x;
999
1000 x = kzalloc(sizeof(struct mtype_resize_ad), GFP_ATOMIC);
1001 if (!x)
1002 /* Don't bother */
1003 goto out;
1004 x->ad = IPSET_ADD;
1005 memcpy(&x->d, value, sizeof(struct mtype_elem));
1006 memcpy(&x->ext, ext, sizeof(struct ip_set_ext));
1007 memcpy(&x->mext, mext, sizeof(struct ip_set_ext));
1008 x->flags = flags;
1009 spin_lock_bh(&set->lock);
1010 list_add_tail(&x->list, &h->ad);
1011 spin_unlock_bh(&set->lock);
1012 }
1013 goto out;
1014
1015 set_full:
1016 if (net_ratelimit())
1017 pr_warn("Set %s is full, maxelem %u reached\n",
1018 set->name, maxelem);
1019 ret = -IPSET_ERR_HASH_FULL;
1020 unlock:
1021 spin_unlock_bh(&t->hregion[r].lock);
1022 out:
1023 if (atomic_dec_and_test(&t->uref) && atomic_read(&t->ref)) {
1024 pr_debug("Table destroy after resize by add: %p\n", t);
1025 mtype_ahash_destroy(set, t, false);
1026 }
1027 return ret;
1028 }
1029
1030 /* Delete an element from the hash and free up space if possible.
1031 */
1032 static int
mtype_del(struct ip_set * set,void * value,const struct ip_set_ext * ext,struct ip_set_ext * mext,u32 flags)1033 mtype_del(struct ip_set *set, void *value, const struct ip_set_ext *ext,
1034 struct ip_set_ext *mext, u32 flags)
1035 {
1036 struct htype *h = set->data;
1037 struct htable *t;
1038 const struct mtype_elem *d = value;
1039 struct mtype_elem *data;
1040 struct hbucket *n;
1041 struct mtype_resize_ad *x = NULL;
1042 int i, j, k, r, ret = -IPSET_ERR_EXIST;
1043 u32 key, multi = 0;
1044 size_t dsize = set->dsize;
1045
1046 /* Userspace add and resize is excluded by the mutex.
1047 * Kernespace add does not trigger resize.
1048 */
1049 rcu_read_lock_bh();
1050 t = rcu_dereference_bh(h->table);
1051 key = HKEY(value, h->initval, t->htable_bits);
1052 r = ahash_region(key, t->htable_bits);
1053 atomic_inc(&t->uref);
1054 rcu_read_unlock_bh();
1055
1056 spin_lock_bh(&t->hregion[r].lock);
1057 n = rcu_dereference_bh(hbucket(t, key));
1058 if (!n)
1059 goto out;
1060 for (i = 0, k = 0; i < n->pos; i++) {
1061 if (!test_bit(i, n->used)) {
1062 k++;
1063 continue;
1064 }
1065 data = ahash_data(n, i, dsize);
1066 if (!mtype_data_equal(data, d, &multi))
1067 continue;
1068 if (SET_ELEM_EXPIRED(set, data))
1069 goto out;
1070
1071 ret = 0;
1072 clear_bit(i, n->used);
1073 smp_mb__after_atomic();
1074 if (i + 1 == n->pos)
1075 n->pos--;
1076 t->hregion[r].elements--;
1077 #ifdef IP_SET_HASH_WITH_NETS
1078 for (j = 0; j < IPSET_NET_COUNT; j++)
1079 mtype_del_cidr(set, h,
1080 NCIDR_PUT(DCIDR_GET(d->cidr, j)), j);
1081 #endif
1082 ip_set_ext_destroy(set, data);
1083
1084 if (atomic_read(&t->ref) && ext->target) {
1085 /* Resize is in process and kernel side del,
1086 * save values
1087 */
1088 x = kzalloc(sizeof(struct mtype_resize_ad),
1089 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 (n->pos == 0 && k == 0) {
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