1 /*-
2 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
12 *
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23 * SUCH DAMAGE.
24 */
25
26 #include <sys/cdefs.h>
27 __FBSDID("$FreeBSD$");
28
29 #define DEB(x)
30 #define DDB(x) x
31
32 /*
33 * Dynamic rule support for ipfw
34 */
35
36 #include "opt_ipfw.h"
37 #include "opt_inet.h"
38 #ifndef INET
39 #error IPFIREWALL requires INET.
40 #endif /* INET */
41 #include "opt_inet6.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/kernel.h>
48 #include <sys/ktr.h>
49 #include <sys/lock.h>
50 #include <sys/rmlock.h>
51 #include <sys/socket.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 #include <net/ethernet.h> /* for ETHERTYPE_IP */
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/vnet.h>
58
59 #include <netinet/in.h>
60 #include <netinet/ip.h>
61 #include <netinet/ip_var.h> /* ip_defttl */
62 #include <netinet/ip_fw.h>
63 #include <netinet/tcp_var.h>
64 #include <netinet/udp.h>
65
66 #include <netinet/ip6.h> /* IN6_ARE_ADDR_EQUAL */
67 #ifdef INET6
68 #include <netinet6/in6_var.h>
69 #include <netinet6/ip6_var.h>
70 #endif
71
72 #include <netpfil/ipfw/ip_fw_private.h>
73
74 #include <machine/in_cksum.h> /* XXX for in_cksum */
75
76 #ifdef MAC
77 #include <security/mac/mac_framework.h>
78 #endif
79
80 /*
81 * Description of dynamic rules.
82 *
83 * Dynamic rules are stored in lists accessed through a hash table
84 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
85 * be modified through the sysctl variable dyn_buckets which is
86 * updated when the table becomes empty.
87 *
88 * XXX currently there is only one list, ipfw_dyn.
89 *
90 * When a packet is received, its address fields are first masked
91 * with the mask defined for the rule, then hashed, then matched
92 * against the entries in the corresponding list.
93 * Dynamic rules can be used for different purposes:
94 * + stateful rules;
95 * + enforcing limits on the number of sessions;
96 * + in-kernel NAT (not implemented yet)
97 *
98 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
99 * measured in seconds and depending on the flags.
100 *
101 * The total number of dynamic rules is equal to UMA zone items count.
102 * The max number of dynamic rules is dyn_max. When we reach
103 * the maximum number of rules we do not create anymore. This is
104 * done to avoid consuming too much memory, but also too much
105 * time when searching on each packet (ideally, we should try instead
106 * to put a limit on the length of the list on each bucket...).
107 *
108 * Each dynamic rule holds a pointer to the parent ipfw rule so
109 * we know what action to perform. Dynamic rules are removed when
110 * the parent rule is deleted. This can be changed by dyn_keep_states
111 * sysctl.
112 *
113 * There are some limitations with dynamic rules -- we do not
114 * obey the 'randomized match', and we do not do multiple
115 * passes through the firewall. XXX check the latter!!!
116 */
117
118 struct ipfw_dyn_bucket {
119 struct mtx mtx; /* Bucket protecting lock */
120 ipfw_dyn_rule *head; /* Pointer to first rule */
121 };
122
123 /*
124 * Static variables followed by global ones
125 */
126 static VNET_DEFINE(struct ipfw_dyn_bucket *, ipfw_dyn_v);
127 static VNET_DEFINE(u_int32_t, dyn_buckets_max);
128 static VNET_DEFINE(u_int32_t, curr_dyn_buckets);
129 static VNET_DEFINE(struct callout, ipfw_timeout);
130 #define V_ipfw_dyn_v VNET(ipfw_dyn_v)
131 #define V_dyn_buckets_max VNET(dyn_buckets_max)
132 #define V_curr_dyn_buckets VNET(curr_dyn_buckets)
133 #define V_ipfw_timeout VNET(ipfw_timeout)
134
135 static VNET_DEFINE(uma_zone_t, ipfw_dyn_rule_zone);
136 #define V_ipfw_dyn_rule_zone VNET(ipfw_dyn_rule_zone)
137
138 #define IPFW_BUCK_LOCK_INIT(b) \
139 mtx_init(&(b)->mtx, "IPFW dynamic bucket", NULL, MTX_DEF)
140 #define IPFW_BUCK_LOCK_DESTROY(b) \
141 mtx_destroy(&(b)->mtx)
142 #define IPFW_BUCK_LOCK(i) mtx_lock(&V_ipfw_dyn_v[(i)].mtx)
143 #define IPFW_BUCK_UNLOCK(i) mtx_unlock(&V_ipfw_dyn_v[(i)].mtx)
144 #define IPFW_BUCK_ASSERT(i) mtx_assert(&V_ipfw_dyn_v[(i)].mtx, MA_OWNED)
145
146
147 static VNET_DEFINE(int, dyn_keep_states);
148 #define V_dyn_keep_states VNET(dyn_keep_states)
149
150 /*
151 * Timeouts for various events in handing dynamic rules.
152 */
153 static VNET_DEFINE(u_int32_t, dyn_ack_lifetime);
154 static VNET_DEFINE(u_int32_t, dyn_syn_lifetime);
155 static VNET_DEFINE(u_int32_t, dyn_fin_lifetime);
156 static VNET_DEFINE(u_int32_t, dyn_rst_lifetime);
157 static VNET_DEFINE(u_int32_t, dyn_udp_lifetime);
158 static VNET_DEFINE(u_int32_t, dyn_short_lifetime);
159
160 #define V_dyn_ack_lifetime VNET(dyn_ack_lifetime)
161 #define V_dyn_syn_lifetime VNET(dyn_syn_lifetime)
162 #define V_dyn_fin_lifetime VNET(dyn_fin_lifetime)
163 #define V_dyn_rst_lifetime VNET(dyn_rst_lifetime)
164 #define V_dyn_udp_lifetime VNET(dyn_udp_lifetime)
165 #define V_dyn_short_lifetime VNET(dyn_short_lifetime)
166
167 /*
168 * Keepalives are sent if dyn_keepalive is set. They are sent every
169 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
170 * seconds of lifetime of a rule.
171 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
172 * than dyn_keepalive_period.
173 */
174
175 static VNET_DEFINE(u_int32_t, dyn_keepalive_interval);
176 static VNET_DEFINE(u_int32_t, dyn_keepalive_period);
177 static VNET_DEFINE(u_int32_t, dyn_keepalive);
178 static VNET_DEFINE(time_t, dyn_keepalive_last);
179
180 #define V_dyn_keepalive_interval VNET(dyn_keepalive_interval)
181 #define V_dyn_keepalive_period VNET(dyn_keepalive_period)
182 #define V_dyn_keepalive VNET(dyn_keepalive)
183 #define V_dyn_keepalive_last VNET(dyn_keepalive_last)
184
185 static VNET_DEFINE(u_int32_t, dyn_max); /* max # of dynamic rules */
186
187 #define DYN_COUNT uma_zone_get_cur(V_ipfw_dyn_rule_zone)
188 #define V_dyn_max VNET(dyn_max)
189
190 /* for userspace, we emulate the uma_zone_counter with ipfw_dyn_count */
191 static int ipfw_dyn_count; /* number of objects */
192
193 #ifdef USERSPACE /* emulation of UMA object counters for userspace */
194 #define uma_zone_get_cur(x) ipfw_dyn_count
195 #endif /* USERSPACE */
196
197 static int last_log; /* Log ratelimiting */
198
199 static void ipfw_dyn_tick(void *vnetx);
200 static void check_dyn_rules(struct ip_fw_chain *, ipfw_range_tlv *, int, int);
201 #ifdef SYSCTL_NODE
202
203 static int sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS);
204 static int sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS);
205
206 SYSBEGIN(f2)
207
208 SYSCTL_DECL(_net_inet_ip_fw);
209 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_buckets,
210 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_buckets_max), 0,
211 "Max number of dyn. buckets");
212 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets,
213 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0,
214 "Current Number of dyn. buckets");
215 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_count,
216 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RD, 0, 0, sysctl_ipfw_dyn_count, "IU",
217 "Number of dyn. rules");
218 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_max,
219 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 0, 0, sysctl_ipfw_dyn_max, "IU",
220 "Max number of dyn. rules");
221 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime,
222 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0,
223 "Lifetime of dyn. rules for acks");
224 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime,
225 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0,
226 "Lifetime of dyn. rules for syn");
227 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime,
228 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0,
229 "Lifetime of dyn. rules for fin");
230 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime,
231 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0,
232 "Lifetime of dyn. rules for rst");
233 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime,
234 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0,
235 "Lifetime of dyn. rules for UDP");
236 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime,
237 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0,
238 "Lifetime of dyn. rules for other situations");
239 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive,
240 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0,
241 "Enable keepalives for dyn. rules");
242 SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, dyn_keep_states,
243 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keep_states), 0,
244 "Do not flush dynamic states on rule deletion");
245
246 SYSEND
247
248 #endif /* SYSCTL_NODE */
249
250
251 #ifdef INET6
252 static __inline int
hash_packet6(struct ipfw_flow_id * id)253 hash_packet6(struct ipfw_flow_id *id)
254 {
255 u_int32_t i;
256 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
257 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
258 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
259 (id->src_ip6.__u6_addr.__u6_addr32[3]) ^
260 (id->dst_port) ^ (id->src_port);
261 return i;
262 }
263 #endif
264
265 /*
266 * IMPORTANT: the hash function for dynamic rules must be commutative
267 * in source and destination (ip,port), because rules are bidirectional
268 * and we want to find both in the same bucket.
269 */
270 static __inline int
hash_packet(struct ipfw_flow_id * id,int buckets)271 hash_packet(struct ipfw_flow_id *id, int buckets)
272 {
273 u_int32_t i;
274
275 #ifdef INET6
276 if (IS_IP6_FLOW_ID(id))
277 i = hash_packet6(id);
278 else
279 #endif /* INET6 */
280 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
281 i &= (buckets - 1);
282 return i;
283 }
284
285 /**
286 * Print customizable flow id description via log(9) facility.
287 */
288 static void
print_dyn_rule_flags(struct ipfw_flow_id * id,int dyn_type,int log_flags,char * prefix,char * postfix)289 print_dyn_rule_flags(struct ipfw_flow_id *id, int dyn_type, int log_flags,
290 char *prefix, char *postfix)
291 {
292 struct in_addr da;
293 #ifdef INET6
294 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN];
295 #else
296 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN];
297 #endif
298
299 #ifdef INET6
300 if (IS_IP6_FLOW_ID(id)) {
301 ip6_sprintf(src, &id->src_ip6);
302 ip6_sprintf(dst, &id->dst_ip6);
303 } else
304 #endif
305 {
306 da.s_addr = htonl(id->src_ip);
307 inet_ntop(AF_INET, &da, src, sizeof(src));
308 da.s_addr = htonl(id->dst_ip);
309 inet_ntop(AF_INET, &da, dst, sizeof(dst));
310 }
311 log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n",
312 prefix, dyn_type, src, id->src_port, dst,
313 id->dst_port, DYN_COUNT, postfix);
314 }
315
316 #define print_dyn_rule(id, dtype, prefix, postfix) \
317 print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix)
318
319 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
320 #define TIME_LE(a,b) ((int)((a)-(b)) < 0)
321
322 /*
323 * Lookup a dynamic rule, locked version.
324 */
325 static ipfw_dyn_rule *
lookup_dyn_rule_locked(struct ipfw_flow_id * pkt,int i,int * match_direction,struct tcphdr * tcp)326 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int i, int *match_direction,
327 struct tcphdr *tcp)
328 {
329 /*
330 * Stateful ipfw extensions.
331 * Lookup into dynamic session queue.
332 */
333 #define MATCH_REVERSE 0
334 #define MATCH_FORWARD 1
335 #define MATCH_NONE 2
336 #define MATCH_UNKNOWN 3
337 int dir = MATCH_NONE;
338 ipfw_dyn_rule *prev, *q = NULL;
339
340 IPFW_BUCK_ASSERT(i);
341
342 for (prev = NULL, q = V_ipfw_dyn_v[i].head; q; prev = q, q = q->next) {
343 if (q->dyn_type == O_LIMIT_PARENT && q->count)
344 continue;
345
346 if (pkt->proto != q->id.proto || q->dyn_type == O_LIMIT_PARENT)
347 continue;
348
349 if (IS_IP6_FLOW_ID(pkt)) {
350 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.src_ip6) &&
351 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.dst_ip6) &&
352 pkt->src_port == q->id.src_port &&
353 pkt->dst_port == q->id.dst_port) {
354 dir = MATCH_FORWARD;
355 break;
356 }
357 if (IN6_ARE_ADDR_EQUAL(&pkt->src_ip6, &q->id.dst_ip6) &&
358 IN6_ARE_ADDR_EQUAL(&pkt->dst_ip6, &q->id.src_ip6) &&
359 pkt->src_port == q->id.dst_port &&
360 pkt->dst_port == q->id.src_port) {
361 dir = MATCH_REVERSE;
362 break;
363 }
364 } else {
365 if (pkt->src_ip == q->id.src_ip &&
366 pkt->dst_ip == q->id.dst_ip &&
367 pkt->src_port == q->id.src_port &&
368 pkt->dst_port == q->id.dst_port) {
369 dir = MATCH_FORWARD;
370 break;
371 }
372 if (pkt->src_ip == q->id.dst_ip &&
373 pkt->dst_ip == q->id.src_ip &&
374 pkt->src_port == q->id.dst_port &&
375 pkt->dst_port == q->id.src_port) {
376 dir = MATCH_REVERSE;
377 break;
378 }
379 }
380 }
381 if (q == NULL)
382 goto done; /* q = NULL, not found */
383
384 if (prev != NULL) { /* found and not in front */
385 prev->next = q->next;
386 q->next = V_ipfw_dyn_v[i].head;
387 V_ipfw_dyn_v[i].head = q;
388 }
389 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
390 uint32_t ack;
391 u_char flags = pkt->_flags & (TH_FIN | TH_SYN | TH_RST);
392
393 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
394 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
395 #define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8))
396 #define ACK_FWD 0x10000 /* fwd ack seen */
397 #define ACK_REV 0x20000 /* rev ack seen */
398
399 q->state |= (dir == MATCH_FORWARD) ? flags : (flags << 8);
400 switch (q->state & TCP_FLAGS) {
401 case TH_SYN: /* opening */
402 q->expire = time_uptime + V_dyn_syn_lifetime;
403 break;
404
405 case BOTH_SYN: /* move to established */
406 case BOTH_SYN | TH_FIN: /* one side tries to close */
407 case BOTH_SYN | (TH_FIN << 8):
408 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
409 if (tcp == NULL)
410 break;
411
412 ack = ntohl(tcp->th_ack);
413 if (dir == MATCH_FORWARD) {
414 if (q->ack_fwd == 0 ||
415 _SEQ_GE(ack, q->ack_fwd)) {
416 q->ack_fwd = ack;
417 q->state |= ACK_FWD;
418 }
419 } else {
420 if (q->ack_rev == 0 ||
421 _SEQ_GE(ack, q->ack_rev)) {
422 q->ack_rev = ack;
423 q->state |= ACK_REV;
424 }
425 }
426 if ((q->state & (ACK_FWD | ACK_REV)) ==
427 (ACK_FWD | ACK_REV)) {
428 q->expire = time_uptime + V_dyn_ack_lifetime;
429 q->state &= ~(ACK_FWD | ACK_REV);
430 }
431 break;
432
433 case BOTH_SYN | BOTH_FIN: /* both sides closed */
434 if (V_dyn_fin_lifetime >= V_dyn_keepalive_period)
435 V_dyn_fin_lifetime = V_dyn_keepalive_period - 1;
436 q->expire = time_uptime + V_dyn_fin_lifetime;
437 break;
438
439 default:
440 #if 0
441 /*
442 * reset or some invalid combination, but can also
443 * occur if we use keep-state the wrong way.
444 */
445 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
446 printf("invalid state: 0x%x\n", q->state);
447 #endif
448 if (V_dyn_rst_lifetime >= V_dyn_keepalive_period)
449 V_dyn_rst_lifetime = V_dyn_keepalive_period - 1;
450 q->expire = time_uptime + V_dyn_rst_lifetime;
451 break;
452 }
453 } else if (pkt->proto == IPPROTO_UDP) {
454 q->expire = time_uptime + V_dyn_udp_lifetime;
455 } else {
456 /* other protocols */
457 q->expire = time_uptime + V_dyn_short_lifetime;
458 }
459 done:
460 if (match_direction != NULL)
461 *match_direction = dir;
462 return (q);
463 }
464
465 ipfw_dyn_rule *
ipfw_lookup_dyn_rule(struct ipfw_flow_id * pkt,int * match_direction,struct tcphdr * tcp)466 ipfw_lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
467 struct tcphdr *tcp)
468 {
469 ipfw_dyn_rule *q;
470 int i;
471
472 i = hash_packet(pkt, V_curr_dyn_buckets);
473
474 IPFW_BUCK_LOCK(i);
475 q = lookup_dyn_rule_locked(pkt, i, match_direction, tcp);
476 if (q == NULL)
477 IPFW_BUCK_UNLOCK(i);
478 /* NB: return table locked when q is not NULL */
479 return q;
480 }
481
482 /*
483 * Unlock bucket mtx
484 * @p - pointer to dynamic rule
485 */
486 void
ipfw_dyn_unlock(ipfw_dyn_rule * q)487 ipfw_dyn_unlock(ipfw_dyn_rule *q)
488 {
489
490 IPFW_BUCK_UNLOCK(q->bucket);
491 }
492
493 static int
resize_dynamic_table(struct ip_fw_chain * chain,int nbuckets)494 resize_dynamic_table(struct ip_fw_chain *chain, int nbuckets)
495 {
496 int i, k, nbuckets_old;
497 ipfw_dyn_rule *q;
498 struct ipfw_dyn_bucket *dyn_v, *dyn_v_old;
499
500 /* Check if given number is power of 2 and less than 64k */
501 if ((nbuckets > 65536) || (!powerof2(nbuckets)))
502 return 1;
503
504 CTR3(KTR_NET, "%s: resize dynamic hash: %d -> %d", __func__,
505 V_curr_dyn_buckets, nbuckets);
506
507 /* Allocate and initialize new hash */
508 dyn_v = malloc(nbuckets * sizeof(ipfw_dyn_rule), M_IPFW,
509 M_WAITOK | M_ZERO);
510
511 for (i = 0 ; i < nbuckets; i++)
512 IPFW_BUCK_LOCK_INIT(&dyn_v[i]);
513
514 /*
515 * Call upper half lock, as get_map() do to ease
516 * read-only access to dynamic rules hash from sysctl
517 */
518 IPFW_UH_WLOCK(chain);
519
520 /*
521 * Acquire chain write lock to permit hash access
522 * for main traffic path without additional locks
523 */
524 IPFW_WLOCK(chain);
525
526 /* Save old values */
527 nbuckets_old = V_curr_dyn_buckets;
528 dyn_v_old = V_ipfw_dyn_v;
529
530 /* Skip relinking if array is not set up */
531 if (V_ipfw_dyn_v == NULL)
532 V_curr_dyn_buckets = 0;
533
534 /* Re-link all dynamic states */
535 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
536 while (V_ipfw_dyn_v[i].head != NULL) {
537 /* Remove from current chain */
538 q = V_ipfw_dyn_v[i].head;
539 V_ipfw_dyn_v[i].head = q->next;
540
541 /* Get new hash value */
542 k = hash_packet(&q->id, nbuckets);
543 q->bucket = k;
544 /* Add to the new head */
545 q->next = dyn_v[k].head;
546 dyn_v[k].head = q;
547 }
548 }
549
550 /* Update current pointers/buckets values */
551 V_curr_dyn_buckets = nbuckets;
552 V_ipfw_dyn_v = dyn_v;
553
554 IPFW_WUNLOCK(chain);
555
556 IPFW_UH_WUNLOCK(chain);
557
558 /* Start periodic callout on initial creation */
559 if (dyn_v_old == NULL) {
560 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, curvnet, 0);
561 return (0);
562 }
563
564 /* Destroy all mutexes */
565 for (i = 0 ; i < nbuckets_old ; i++)
566 IPFW_BUCK_LOCK_DESTROY(&dyn_v_old[i]);
567
568 /* Free old hash */
569 free(dyn_v_old, M_IPFW);
570
571 return 0;
572 }
573
574 /**
575 * Install state of type 'type' for a dynamic session.
576 * The hash table contains two type of rules:
577 * - regular rules (O_KEEP_STATE)
578 * - rules for sessions with limited number of sess per user
579 * (O_LIMIT). When they are created, the parent is
580 * increased by 1, and decreased on delete. In this case,
581 * the third parameter is the parent rule and not the chain.
582 * - "parent" rules for the above (O_LIMIT_PARENT).
583 */
584 static ipfw_dyn_rule *
add_dyn_rule(struct ipfw_flow_id * id,int i,u_int8_t dyn_type,struct ip_fw * rule)585 add_dyn_rule(struct ipfw_flow_id *id, int i, u_int8_t dyn_type, struct ip_fw *rule)
586 {
587 ipfw_dyn_rule *r;
588
589 IPFW_BUCK_ASSERT(i);
590
591 r = uma_zalloc(V_ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
592 if (r == NULL) {
593 if (last_log != time_uptime) {
594 last_log = time_uptime;
595 log(LOG_DEBUG,
596 "ipfw: Cannot allocate dynamic state, "
597 "consider increasing net.inet.ip.fw.dyn_max\n");
598 }
599 return NULL;
600 }
601 ipfw_dyn_count++;
602
603 /*
604 * refcount on parent is already incremented, so
605 * it is safe to use parent unlocked.
606 */
607 if (dyn_type == O_LIMIT) {
608 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
609 if ( parent->dyn_type != O_LIMIT_PARENT)
610 panic("invalid parent");
611 r->parent = parent;
612 rule = parent->rule;
613 }
614
615 r->id = *id;
616 r->expire = time_uptime + V_dyn_syn_lifetime;
617 r->rule = rule;
618 r->dyn_type = dyn_type;
619 IPFW_ZERO_DYN_COUNTER(r);
620 r->count = 0;
621
622 r->bucket = i;
623 r->next = V_ipfw_dyn_v[i].head;
624 V_ipfw_dyn_v[i].head = r;
625 DEB(print_dyn_rule(id, dyn_type, "add dyn entry", "total");)
626 return r;
627 }
628
629 /**
630 * lookup dynamic parent rule using pkt and rule as search keys.
631 * If the lookup fails, then install one.
632 */
633 static ipfw_dyn_rule *
lookup_dyn_parent(struct ipfw_flow_id * pkt,int * pindex,struct ip_fw * rule)634 lookup_dyn_parent(struct ipfw_flow_id *pkt, int *pindex, struct ip_fw *rule)
635 {
636 ipfw_dyn_rule *q;
637 int i, is_v6;
638
639 is_v6 = IS_IP6_FLOW_ID(pkt);
640 i = hash_packet( pkt, V_curr_dyn_buckets );
641 *pindex = i;
642 IPFW_BUCK_LOCK(i);
643 for (q = V_ipfw_dyn_v[i].head ; q != NULL ; q=q->next)
644 if (q->dyn_type == O_LIMIT_PARENT &&
645 rule== q->rule &&
646 pkt->proto == q->id.proto &&
647 pkt->src_port == q->id.src_port &&
648 pkt->dst_port == q->id.dst_port &&
649 (
650 (is_v6 &&
651 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
652 &(q->id.src_ip6)) &&
653 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
654 &(q->id.dst_ip6))) ||
655 (!is_v6 &&
656 pkt->src_ip == q->id.src_ip &&
657 pkt->dst_ip == q->id.dst_ip)
658 )
659 ) {
660 q->expire = time_uptime + V_dyn_short_lifetime;
661 DEB(print_dyn_rule(pkt, q->dyn_type,
662 "lookup_dyn_parent found", "");)
663 return q;
664 }
665
666 /* Add virtual limiting rule */
667 return add_dyn_rule(pkt, i, O_LIMIT_PARENT, rule);
668 }
669
670 /**
671 * Install dynamic state for rule type cmd->o.opcode
672 *
673 * Returns 1 (failure) if state is not installed because of errors or because
674 * session limitations are enforced.
675 */
676 int
ipfw_install_state(struct ip_fw_chain * chain,struct ip_fw * rule,ipfw_insn_limit * cmd,struct ip_fw_args * args,uint32_t tablearg)677 ipfw_install_state(struct ip_fw_chain *chain, struct ip_fw *rule,
678 ipfw_insn_limit *cmd, struct ip_fw_args *args, uint32_t tablearg)
679 {
680 ipfw_dyn_rule *q;
681 int i;
682
683 DEB(print_dyn_rule(&args->f_id, cmd->o.opcode, "install_state", "");)
684
685 i = hash_packet(&args->f_id, V_curr_dyn_buckets);
686
687 IPFW_BUCK_LOCK(i);
688
689 q = lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL);
690
691 if (q != NULL) { /* should never occur */
692 DEB(
693 if (last_log != time_uptime) {
694 last_log = time_uptime;
695 printf("ipfw: %s: entry already present, done\n",
696 __func__);
697 })
698 IPFW_BUCK_UNLOCK(i);
699 return (0);
700 }
701
702 /*
703 * State limiting is done via uma(9) zone limiting.
704 * Save pointer to newly-installed rule and reject
705 * packet if add_dyn_rule() returned NULL.
706 * Note q is currently set to NULL.
707 */
708
709 switch (cmd->o.opcode) {
710 case O_KEEP_STATE: /* bidir rule */
711 q = add_dyn_rule(&args->f_id, i, O_KEEP_STATE, rule);
712 break;
713
714 case O_LIMIT: { /* limit number of sessions */
715 struct ipfw_flow_id id;
716 ipfw_dyn_rule *parent;
717 uint32_t conn_limit;
718 uint16_t limit_mask = cmd->limit_mask;
719 int pindex;
720
721 conn_limit = IP_FW_ARG_TABLEARG(chain, cmd->conn_limit, limit);
722
723 DEB(
724 if (cmd->conn_limit == IP_FW_TARG)
725 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u "
726 "(tablearg)\n", __func__, conn_limit);
727 else
728 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n",
729 __func__, conn_limit);
730 )
731
732 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0;
733 id.proto = args->f_id.proto;
734 id.addr_type = args->f_id.addr_type;
735 id.fib = M_GETFIB(args->m);
736
737 if (IS_IP6_FLOW_ID (&(args->f_id))) {
738 bzero(&id.src_ip6, sizeof(id.src_ip6));
739 bzero(&id.dst_ip6, sizeof(id.dst_ip6));
740
741 if (limit_mask & DYN_SRC_ADDR)
742 id.src_ip6 = args->f_id.src_ip6;
743 if (limit_mask & DYN_DST_ADDR)
744 id.dst_ip6 = args->f_id.dst_ip6;
745 } else {
746 if (limit_mask & DYN_SRC_ADDR)
747 id.src_ip = args->f_id.src_ip;
748 if (limit_mask & DYN_DST_ADDR)
749 id.dst_ip = args->f_id.dst_ip;
750 }
751 if (limit_mask & DYN_SRC_PORT)
752 id.src_port = args->f_id.src_port;
753 if (limit_mask & DYN_DST_PORT)
754 id.dst_port = args->f_id.dst_port;
755
756 /*
757 * We have to release lock for previous bucket to
758 * avoid possible deadlock
759 */
760 IPFW_BUCK_UNLOCK(i);
761
762 if ((parent = lookup_dyn_parent(&id, &pindex, rule)) == NULL) {
763 printf("ipfw: %s: add parent failed\n", __func__);
764 IPFW_BUCK_UNLOCK(pindex);
765 return (1);
766 }
767
768 if (parent->count >= conn_limit) {
769 if (V_fw_verbose && last_log != time_uptime) {
770 last_log = time_uptime;
771 char sbuf[24];
772 last_log = time_uptime;
773 snprintf(sbuf, sizeof(sbuf),
774 "%d drop session",
775 parent->rule->rulenum);
776 print_dyn_rule_flags(&args->f_id,
777 cmd->o.opcode,
778 LOG_SECURITY | LOG_DEBUG,
779 sbuf, "too many entries");
780 }
781 IPFW_BUCK_UNLOCK(pindex);
782 return (1);
783 }
784 /* Increment counter on parent */
785 parent->count++;
786 IPFW_BUCK_UNLOCK(pindex);
787
788 IPFW_BUCK_LOCK(i);
789 q = add_dyn_rule(&args->f_id, i, O_LIMIT, (struct ip_fw *)parent);
790 if (q == NULL) {
791 /* Decrement index and notify caller */
792 IPFW_BUCK_UNLOCK(i);
793 IPFW_BUCK_LOCK(pindex);
794 parent->count--;
795 IPFW_BUCK_UNLOCK(pindex);
796 return (1);
797 }
798 break;
799 }
800 default:
801 printf("ipfw: %s: unknown dynamic rule type %u\n",
802 __func__, cmd->o.opcode);
803 }
804
805 if (q == NULL) {
806 IPFW_BUCK_UNLOCK(i);
807 return (1); /* Notify caller about failure */
808 }
809
810 /* XXX just set lifetime */
811 lookup_dyn_rule_locked(&args->f_id, i, NULL, NULL);
812
813 IPFW_BUCK_UNLOCK(i);
814 return (0);
815 }
816
817 /*
818 * Generate a TCP packet, containing either a RST or a keepalive.
819 * When flags & TH_RST, we are sending a RST packet, because of a
820 * "reset" action matched the packet.
821 * Otherwise we are sending a keepalive, and flags & TH_
822 * The 'replyto' mbuf is the mbuf being replied to, if any, and is required
823 * so that MAC can label the reply appropriately.
824 */
825 struct mbuf *
ipfw_send_pkt(struct mbuf * replyto,struct ipfw_flow_id * id,u_int32_t seq,u_int32_t ack,int flags)826 ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq,
827 u_int32_t ack, int flags)
828 {
829 struct mbuf *m = NULL; /* stupid compiler */
830 int len, dir;
831 struct ip *h = NULL; /* stupid compiler */
832 #ifdef INET6
833 struct ip6_hdr *h6 = NULL;
834 #endif
835 struct tcphdr *th = NULL;
836
837 MGETHDR(m, M_NOWAIT, MT_DATA);
838 if (m == NULL)
839 return (NULL);
840
841 M_SETFIB(m, id->fib);
842 #ifdef MAC
843 if (replyto != NULL)
844 mac_netinet_firewall_reply(replyto, m);
845 else
846 mac_netinet_firewall_send(m);
847 #else
848 (void)replyto; /* don't warn about unused arg */
849 #endif
850
851 switch (id->addr_type) {
852 case 4:
853 len = sizeof(struct ip) + sizeof(struct tcphdr);
854 break;
855 #ifdef INET6
856 case 6:
857 len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
858 break;
859 #endif
860 default:
861 /* XXX: log me?!? */
862 FREE_PKT(m);
863 return (NULL);
864 }
865 dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN);
866
867 m->m_data += max_linkhdr;
868 m->m_flags |= M_SKIP_FIREWALL;
869 m->m_pkthdr.len = m->m_len = len;
870 m->m_pkthdr.rcvif = NULL;
871 bzero(m->m_data, len);
872
873 switch (id->addr_type) {
874 case 4:
875 h = mtod(m, struct ip *);
876
877 /* prepare for checksum */
878 h->ip_p = IPPROTO_TCP;
879 h->ip_len = htons(sizeof(struct tcphdr));
880 if (dir) {
881 h->ip_src.s_addr = htonl(id->src_ip);
882 h->ip_dst.s_addr = htonl(id->dst_ip);
883 } else {
884 h->ip_src.s_addr = htonl(id->dst_ip);
885 h->ip_dst.s_addr = htonl(id->src_ip);
886 }
887
888 th = (struct tcphdr *)(h + 1);
889 break;
890 #ifdef INET6
891 case 6:
892 h6 = mtod(m, struct ip6_hdr *);
893
894 /* prepare for checksum */
895 h6->ip6_nxt = IPPROTO_TCP;
896 h6->ip6_plen = htons(sizeof(struct tcphdr));
897 if (dir) {
898 h6->ip6_src = id->src_ip6;
899 h6->ip6_dst = id->dst_ip6;
900 } else {
901 h6->ip6_src = id->dst_ip6;
902 h6->ip6_dst = id->src_ip6;
903 }
904
905 th = (struct tcphdr *)(h6 + 1);
906 break;
907 #endif
908 }
909
910 if (dir) {
911 th->th_sport = htons(id->src_port);
912 th->th_dport = htons(id->dst_port);
913 } else {
914 th->th_sport = htons(id->dst_port);
915 th->th_dport = htons(id->src_port);
916 }
917 th->th_off = sizeof(struct tcphdr) >> 2;
918
919 if (flags & TH_RST) {
920 if (flags & TH_ACK) {
921 th->th_seq = htonl(ack);
922 th->th_flags = TH_RST;
923 } else {
924 if (flags & TH_SYN)
925 seq++;
926 th->th_ack = htonl(seq);
927 th->th_flags = TH_RST | TH_ACK;
928 }
929 } else {
930 /*
931 * Keepalive - use caller provided sequence numbers
932 */
933 th->th_seq = htonl(seq);
934 th->th_ack = htonl(ack);
935 th->th_flags = TH_ACK;
936 }
937
938 switch (id->addr_type) {
939 case 4:
940 th->th_sum = in_cksum(m, len);
941
942 /* finish the ip header */
943 h->ip_v = 4;
944 h->ip_hl = sizeof(*h) >> 2;
945 h->ip_tos = IPTOS_LOWDELAY;
946 h->ip_off = htons(0);
947 h->ip_len = htons(len);
948 h->ip_ttl = V_ip_defttl;
949 h->ip_sum = 0;
950 break;
951 #ifdef INET6
952 case 6:
953 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6),
954 sizeof(struct tcphdr));
955
956 /* finish the ip6 header */
957 h6->ip6_vfc |= IPV6_VERSION;
958 h6->ip6_hlim = IPV6_DEFHLIM;
959 break;
960 #endif
961 }
962
963 return (m);
964 }
965
966 /*
967 * Queue keepalive packets for given dynamic rule
968 */
969 static struct mbuf **
ipfw_dyn_send_ka(struct mbuf ** mtailp,ipfw_dyn_rule * q)970 ipfw_dyn_send_ka(struct mbuf **mtailp, ipfw_dyn_rule *q)
971 {
972 struct mbuf *m_rev, *m_fwd;
973
974 m_rev = (q->state & ACK_REV) ? NULL :
975 ipfw_send_pkt(NULL, &(q->id), q->ack_rev - 1, q->ack_fwd, TH_SYN);
976 m_fwd = (q->state & ACK_FWD) ? NULL :
977 ipfw_send_pkt(NULL, &(q->id), q->ack_fwd - 1, q->ack_rev, 0);
978
979 if (m_rev != NULL) {
980 *mtailp = m_rev;
981 mtailp = &(*mtailp)->m_nextpkt;
982 }
983 if (m_fwd != NULL) {
984 *mtailp = m_fwd;
985 mtailp = &(*mtailp)->m_nextpkt;
986 }
987
988 return (mtailp);
989 }
990
991 /*
992 * This procedure is used to perform various maintance
993 * on dynamic hash list. Currently it is called every second.
994 */
995 static void
ipfw_dyn_tick(void * vnetx)996 ipfw_dyn_tick(void * vnetx)
997 {
998 struct ip_fw_chain *chain;
999 int check_ka = 0;
1000 #ifdef VIMAGE
1001 struct vnet *vp = vnetx;
1002 #endif
1003
1004 CURVNET_SET(vp);
1005
1006 chain = &V_layer3_chain;
1007
1008 /* Run keepalive checks every keepalive_period iff ka is enabled */
1009 if ((V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) &&
1010 (V_dyn_keepalive != 0)) {
1011 V_dyn_keepalive_last = time_uptime;
1012 check_ka = 1;
1013 }
1014
1015 check_dyn_rules(chain, NULL, check_ka, 1);
1016
1017 callout_reset_on(&V_ipfw_timeout, hz, ipfw_dyn_tick, vnetx, 0);
1018
1019 CURVNET_RESTORE();
1020 }
1021
1022
1023 /*
1024 * Walk thru all dynamic states doing generic maintance:
1025 * 1) free expired states
1026 * 2) free all states based on deleted rule / set
1027 * 3) send keepalives for states if needed
1028 *
1029 * @chain - pointer to current ipfw rules chain
1030 * @rule - delete all states originated by given rule if != NULL
1031 * @set - delete all states originated by any rule in set @set if != RESVD_SET
1032 * @check_ka - perform checking/sending keepalives
1033 * @timer - indicate call from timer routine.
1034 *
1035 * Timer routine must call this function unlocked to permit
1036 * sending keepalives/resizing table.
1037 *
1038 * Others has to call function with IPFW_UH_WLOCK held.
1039 * Additionally, function assume that dynamic rule/set is
1040 * ALREADY deleted so no new states can be generated by
1041 * 'deleted' rules.
1042 *
1043 * Write lock is needed to ensure that unused parent rules
1044 * are not freed by other instance (see stage 2, 3)
1045 */
1046 static void
check_dyn_rules(struct ip_fw_chain * chain,ipfw_range_tlv * rt,int check_ka,int timer)1047 check_dyn_rules(struct ip_fw_chain *chain, ipfw_range_tlv *rt,
1048 int check_ka, int timer)
1049 {
1050 struct mbuf *m0, *m, *mnext, **mtailp;
1051 struct ip *h;
1052 int i, dyn_count, new_buckets = 0, max_buckets;
1053 int expired = 0, expired_limits = 0, parents = 0, total = 0;
1054 ipfw_dyn_rule *q, *q_prev, *q_next;
1055 ipfw_dyn_rule *exp_head, **exptailp;
1056 ipfw_dyn_rule *exp_lhead, **expltailp;
1057
1058 KASSERT(V_ipfw_dyn_v != NULL, ("%s: dynamic table not allocated",
1059 __func__));
1060
1061 /* Avoid possible LOR */
1062 KASSERT(!check_ka || timer, ("%s: keepalive check with lock held",
1063 __func__));
1064
1065 /*
1066 * Do not perform any checks if we currently have no dynamic states
1067 */
1068 if (DYN_COUNT == 0)
1069 return;
1070
1071 /* Expired states */
1072 exp_head = NULL;
1073 exptailp = &exp_head;
1074
1075 /* Expired limit states */
1076 exp_lhead = NULL;
1077 expltailp = &exp_lhead;
1078
1079 /*
1080 * We make a chain of packets to go out here -- not deferring
1081 * until after we drop the IPFW dynamic rule lock would result
1082 * in a lock order reversal with the normal packet input -> ipfw
1083 * call stack.
1084 */
1085 m0 = NULL;
1086 mtailp = &m0;
1087
1088 /* Protect from hash resizing */
1089 if (timer != 0)
1090 IPFW_UH_WLOCK(chain);
1091 else
1092 IPFW_UH_WLOCK_ASSERT(chain);
1093
1094 #define NEXT_RULE() { q_prev = q; q = q->next ; continue; }
1095
1096 /* Stage 1: perform requested deletion */
1097 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1098 IPFW_BUCK_LOCK(i);
1099 for (q = V_ipfw_dyn_v[i].head, q_prev = q; q ; ) {
1100 /* account every rule */
1101 total++;
1102
1103 /* Skip parent rules at all */
1104 if (q->dyn_type == O_LIMIT_PARENT) {
1105 parents++;
1106 NEXT_RULE();
1107 }
1108
1109 /*
1110 * Remove rules which are:
1111 * 1) expired
1112 * 2) matches deletion range
1113 */
1114 if ((TIME_LEQ(q->expire, time_uptime)) ||
1115 (rt != NULL && ipfw_match_range(q->rule, rt))) {
1116 if (TIME_LE(time_uptime, q->expire) &&
1117 q->dyn_type == O_KEEP_STATE &&
1118 V_dyn_keep_states != 0) {
1119 /*
1120 * Do not delete state if
1121 * it is not expired and
1122 * dyn_keep_states is ON.
1123 * However we need to re-link it
1124 * to any other stable rule
1125 */
1126 q->rule = chain->default_rule;
1127 NEXT_RULE();
1128 }
1129
1130 /* Unlink q from current list */
1131 q_next = q->next;
1132 if (q == V_ipfw_dyn_v[i].head)
1133 V_ipfw_dyn_v[i].head = q_next;
1134 else
1135 q_prev->next = q_next;
1136
1137 q->next = NULL;
1138
1139 /* queue q to expire list */
1140 if (q->dyn_type != O_LIMIT) {
1141 *exptailp = q;
1142 exptailp = &(*exptailp)->next;
1143 DEB(print_dyn_rule(&q->id, q->dyn_type,
1144 "unlink entry", "left");
1145 )
1146 } else {
1147 /* Separate list for limit rules */
1148 *expltailp = q;
1149 expltailp = &(*expltailp)->next;
1150 expired_limits++;
1151 DEB(print_dyn_rule(&q->id, q->dyn_type,
1152 "unlink limit entry", "left");
1153 )
1154 }
1155
1156 q = q_next;
1157 expired++;
1158 continue;
1159 }
1160
1161 /*
1162 * Check if we need to send keepalive:
1163 * we need to ensure if is time to do KA,
1164 * this is established TCP session, and
1165 * expire time is within keepalive interval
1166 */
1167 if ((check_ka != 0) && (q->id.proto == IPPROTO_TCP) &&
1168 ((q->state & BOTH_SYN) == BOTH_SYN) &&
1169 (TIME_LEQ(q->expire, time_uptime +
1170 V_dyn_keepalive_interval)))
1171 mtailp = ipfw_dyn_send_ka(mtailp, q);
1172
1173 NEXT_RULE();
1174 }
1175 IPFW_BUCK_UNLOCK(i);
1176 }
1177
1178 /* Stage 2: decrement counters from O_LIMIT parents */
1179 if (expired_limits != 0) {
1180 /*
1181 * XXX: Note that deleting set with more than one
1182 * heavily-used LIMIT rules can result in overwhelming
1183 * locking due to lack of per-hash value sorting
1184 *
1185 * We should probably think about:
1186 * 1) pre-allocating hash of size, say,
1187 * MAX(16, V_curr_dyn_buckets / 1024)
1188 * 2) checking if expired_limits is large enough
1189 * 3) If yes, init hash (or its part), re-link
1190 * current list and start decrementing procedure in
1191 * each bucket separately
1192 */
1193
1194 /*
1195 * Small optimization: do not unlock bucket until
1196 * we see the next item resides in different bucket
1197 */
1198 if (exp_lhead != NULL) {
1199 i = exp_lhead->parent->bucket;
1200 IPFW_BUCK_LOCK(i);
1201 }
1202 for (q = exp_lhead; q != NULL; q = q->next) {
1203 if (i != q->parent->bucket) {
1204 IPFW_BUCK_UNLOCK(i);
1205 i = q->parent->bucket;
1206 IPFW_BUCK_LOCK(i);
1207 }
1208
1209 /* Decrease parent refcount */
1210 q->parent->count--;
1211 }
1212 if (exp_lhead != NULL)
1213 IPFW_BUCK_UNLOCK(i);
1214 }
1215
1216 /*
1217 * We protectet ourselves from unused parent deletion
1218 * (from the timer function) by holding UH write lock.
1219 */
1220
1221 /* Stage 3: remove unused parent rules */
1222 if ((parents != 0) && (expired != 0)) {
1223 for (i = 0 ; i < V_curr_dyn_buckets ; i++) {
1224 IPFW_BUCK_LOCK(i);
1225 for (q = V_ipfw_dyn_v[i].head, q_prev = q ; q ; ) {
1226 if (q->dyn_type != O_LIMIT_PARENT)
1227 NEXT_RULE();
1228
1229 if (q->count != 0)
1230 NEXT_RULE();
1231
1232 /* Parent rule without consumers */
1233
1234 /* Unlink q from current list */
1235 q_next = q->next;
1236 if (q == V_ipfw_dyn_v[i].head)
1237 V_ipfw_dyn_v[i].head = q_next;
1238 else
1239 q_prev->next = q_next;
1240
1241 q->next = NULL;
1242
1243 /* Add to expired list */
1244 *exptailp = q;
1245 exptailp = &(*exptailp)->next;
1246
1247 DEB(print_dyn_rule(&q->id, q->dyn_type,
1248 "unlink parent entry", "left");
1249 )
1250
1251 expired++;
1252
1253 q = q_next;
1254 }
1255 IPFW_BUCK_UNLOCK(i);
1256 }
1257 }
1258
1259 #undef NEXT_RULE
1260
1261 if (timer != 0) {
1262 /*
1263 * Check if we need to resize hash:
1264 * if current number of states exceeds number of buckes in hash,
1265 * grow hash size to the minimum power of 2 which is bigger than
1266 * current states count. Limit hash size by 64k.
1267 */
1268 max_buckets = (V_dyn_buckets_max > 65536) ?
1269 65536 : V_dyn_buckets_max;
1270
1271 dyn_count = DYN_COUNT;
1272
1273 if ((dyn_count > V_curr_dyn_buckets * 2) &&
1274 (dyn_count < max_buckets)) {
1275 new_buckets = V_curr_dyn_buckets;
1276 while (new_buckets < dyn_count) {
1277 new_buckets *= 2;
1278
1279 if (new_buckets >= max_buckets)
1280 break;
1281 }
1282 }
1283
1284 IPFW_UH_WUNLOCK(chain);
1285 }
1286
1287 /* Finally delete old states ad limits if any */
1288 for (q = exp_head; q != NULL; q = q_next) {
1289 q_next = q->next;
1290 uma_zfree(V_ipfw_dyn_rule_zone, q);
1291 ipfw_dyn_count--;
1292 }
1293
1294 for (q = exp_lhead; q != NULL; q = q_next) {
1295 q_next = q->next;
1296 uma_zfree(V_ipfw_dyn_rule_zone, q);
1297 ipfw_dyn_count--;
1298 }
1299
1300 /*
1301 * The rest code MUST be called from timer routine only
1302 * without holding any locks
1303 */
1304 if (timer == 0)
1305 return;
1306
1307 /* Send keepalive packets if any */
1308 for (m = m0; m != NULL; m = mnext) {
1309 mnext = m->m_nextpkt;
1310 m->m_nextpkt = NULL;
1311 h = mtod(m, struct ip *);
1312 if (h->ip_v == 4)
1313 ip_output(m, NULL, NULL, 0, NULL, NULL);
1314 #ifdef INET6
1315 else
1316 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
1317 #endif
1318 }
1319
1320 /* Run table resize without holding any locks */
1321 if (new_buckets != 0)
1322 resize_dynamic_table(chain, new_buckets);
1323 }
1324
1325 /*
1326 * Deletes all dynamic rules originated by given rule or all rules in
1327 * given set. Specify RESVD_SET to indicate set should not be used.
1328 * @chain - pointer to current ipfw rules chain
1329 * @rr - delete all states originated by rules in matched range.
1330 *
1331 * Function has to be called with IPFW_UH_WLOCK held.
1332 * Additionally, function assume that dynamic rule/set is
1333 * ALREADY deleted so no new states can be generated by
1334 * 'deleted' rules.
1335 */
1336 void
ipfw_expire_dyn_rules(struct ip_fw_chain * chain,ipfw_range_tlv * rt)1337 ipfw_expire_dyn_rules(struct ip_fw_chain *chain, ipfw_range_tlv *rt)
1338 {
1339
1340 check_dyn_rules(chain, rt, 0, 0);
1341 }
1342
1343 /*
1344 * Check if rule contains at least one dynamic opcode.
1345 *
1346 * Returns 1 if such opcode is found, 0 otherwise.
1347 */
1348 int
ipfw_is_dyn_rule(struct ip_fw * rule)1349 ipfw_is_dyn_rule(struct ip_fw *rule)
1350 {
1351 int cmdlen, l;
1352 ipfw_insn *cmd;
1353
1354 l = rule->cmd_len;
1355 cmd = rule->cmd;
1356 cmdlen = 0;
1357 for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) {
1358 cmdlen = F_LEN(cmd);
1359
1360 switch (cmd->opcode) {
1361 case O_LIMIT:
1362 case O_KEEP_STATE:
1363 case O_PROBE_STATE:
1364 case O_CHECK_STATE:
1365 return (1);
1366 }
1367 }
1368
1369 return (0);
1370 }
1371
1372 void
ipfw_dyn_init(struct ip_fw_chain * chain)1373 ipfw_dyn_init(struct ip_fw_chain *chain)
1374 {
1375
1376 V_ipfw_dyn_v = NULL;
1377 V_dyn_buckets_max = 256; /* must be power of 2 */
1378 V_curr_dyn_buckets = 256; /* must be power of 2 */
1379
1380 V_dyn_ack_lifetime = 300;
1381 V_dyn_syn_lifetime = 20;
1382 V_dyn_fin_lifetime = 1;
1383 V_dyn_rst_lifetime = 1;
1384 V_dyn_udp_lifetime = 10;
1385 V_dyn_short_lifetime = 5;
1386
1387 V_dyn_keepalive_interval = 20;
1388 V_dyn_keepalive_period = 5;
1389 V_dyn_keepalive = 1; /* do send keepalives */
1390 V_dyn_keepalive_last = time_uptime;
1391
1392 V_dyn_max = 16384; /* max # of dynamic rules */
1393
1394 V_ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule",
1395 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
1396 UMA_ALIGN_PTR, 0);
1397
1398 /* Enforce limit on dynamic rules */
1399 uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max);
1400
1401 callout_init(&V_ipfw_timeout, 1);
1402
1403 /*
1404 * This can potentially be done on first dynamic rule
1405 * being added to chain.
1406 */
1407 resize_dynamic_table(chain, V_curr_dyn_buckets);
1408 }
1409
1410 void
ipfw_dyn_uninit(int pass)1411 ipfw_dyn_uninit(int pass)
1412 {
1413 int i;
1414
1415 if (pass == 0) {
1416 callout_drain(&V_ipfw_timeout);
1417 return;
1418 }
1419
1420 if (V_ipfw_dyn_v != NULL) {
1421 /*
1422 * Skip deleting all dynamic states -
1423 * uma_zdestroy() does this more efficiently;
1424 */
1425
1426 /* Destroy all mutexes */
1427 for (i = 0 ; i < V_curr_dyn_buckets ; i++)
1428 IPFW_BUCK_LOCK_DESTROY(&V_ipfw_dyn_v[i]);
1429 free(V_ipfw_dyn_v, M_IPFW);
1430 V_ipfw_dyn_v = NULL;
1431 }
1432
1433 uma_zdestroy(V_ipfw_dyn_rule_zone);
1434 }
1435
1436 #ifdef SYSCTL_NODE
1437 /*
1438 * Get/set maximum number of dynamic states in given VNET instance.
1439 */
1440 static int
sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS)1441 sysctl_ipfw_dyn_max(SYSCTL_HANDLER_ARGS)
1442 {
1443 int error;
1444 unsigned int nstates;
1445
1446 nstates = V_dyn_max;
1447
1448 error = sysctl_handle_int(oidp, &nstates, 0, req);
1449 /* Read operation or some error */
1450 if ((error != 0) || (req->newptr == NULL))
1451 return (error);
1452
1453 V_dyn_max = nstates;
1454 uma_zone_set_max(V_ipfw_dyn_rule_zone, V_dyn_max);
1455
1456 return (0);
1457 }
1458
1459 /*
1460 * Get current number of dynamic states in given VNET instance.
1461 */
1462 static int
sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS)1463 sysctl_ipfw_dyn_count(SYSCTL_HANDLER_ARGS)
1464 {
1465 int error;
1466 unsigned int nstates;
1467
1468 nstates = DYN_COUNT;
1469
1470 error = sysctl_handle_int(oidp, &nstates, 0, req);
1471
1472 return (error);
1473 }
1474 #endif
1475
1476 /*
1477 * Returns size of dynamic states in legacy format
1478 */
1479 int
ipfw_dyn_len(void)1480 ipfw_dyn_len(void)
1481 {
1482
1483 return (V_ipfw_dyn_v == NULL) ? 0 :
1484 (DYN_COUNT * sizeof(ipfw_dyn_rule));
1485 }
1486
1487 /*
1488 * Returns number of dynamic states.
1489 * Used by dump format v1 (current).
1490 */
1491 int
ipfw_dyn_get_count(void)1492 ipfw_dyn_get_count(void)
1493 {
1494
1495 return (V_ipfw_dyn_v == NULL) ? 0 : DYN_COUNT;
1496 }
1497
1498 static void
export_dyn_rule(ipfw_dyn_rule * src,ipfw_dyn_rule * dst)1499 export_dyn_rule(ipfw_dyn_rule *src, ipfw_dyn_rule *dst)
1500 {
1501
1502 memcpy(dst, src, sizeof(*src));
1503 memcpy(&(dst->rule), &(src->rule->rulenum), sizeof(src->rule->rulenum));
1504 /*
1505 * store set number into high word of
1506 * dst->rule pointer.
1507 */
1508 memcpy((char *)&dst->rule + sizeof(src->rule->rulenum),
1509 &(src->rule->set), sizeof(src->rule->set));
1510 /*
1511 * store a non-null value in "next".
1512 * The userland code will interpret a
1513 * NULL here as a marker
1514 * for the last dynamic rule.
1515 */
1516 memcpy(&dst->next, &dst, sizeof(dst));
1517 dst->expire =
1518 TIME_LEQ(dst->expire, time_uptime) ? 0 : dst->expire - time_uptime;
1519 }
1520
1521 /*
1522 * Fills int buffer given by @sd with dynamic states.
1523 * Used by dump format v1 (current).
1524 *
1525 * Returns 0 on success.
1526 */
1527 int
ipfw_dump_states(struct ip_fw_chain * chain,struct sockopt_data * sd)1528 ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd)
1529 {
1530 ipfw_dyn_rule *p;
1531 ipfw_obj_dyntlv *dst, *last;
1532 ipfw_obj_ctlv *ctlv;
1533 int i;
1534 size_t sz;
1535
1536 if (V_ipfw_dyn_v == NULL)
1537 return (0);
1538
1539 IPFW_UH_RLOCK_ASSERT(chain);
1540
1541 ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv));
1542 if (ctlv == NULL)
1543 return (ENOMEM);
1544 sz = sizeof(ipfw_obj_dyntlv);
1545 ctlv->head.type = IPFW_TLV_DYNSTATE_LIST;
1546 ctlv->objsize = sz;
1547 last = NULL;
1548
1549 for (i = 0 ; i < V_curr_dyn_buckets; i++) {
1550 IPFW_BUCK_LOCK(i);
1551 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) {
1552 dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, sz);
1553 if (dst == NULL) {
1554 IPFW_BUCK_UNLOCK(i);
1555 return (ENOMEM);
1556 }
1557
1558 export_dyn_rule(p, &dst->state);
1559 dst->head.length = sz;
1560 dst->head.type = IPFW_TLV_DYN_ENT;
1561 last = dst;
1562 }
1563 IPFW_BUCK_UNLOCK(i);
1564 }
1565
1566 if (last != NULL) /* mark last dynamic rule */
1567 last->head.flags = IPFW_DF_LAST;
1568
1569 return (0);
1570 }
1571
1572 /*
1573 * Fill given buffer with dynamic states (legacy format).
1574 * IPFW_UH_RLOCK has to be held while calling.
1575 */
1576 void
ipfw_get_dynamic(struct ip_fw_chain * chain,char ** pbp,const char * ep)1577 ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep)
1578 {
1579 ipfw_dyn_rule *p, *last = NULL;
1580 char *bp;
1581 int i;
1582
1583 if (V_ipfw_dyn_v == NULL)
1584 return;
1585 bp = *pbp;
1586
1587 IPFW_UH_RLOCK_ASSERT(chain);
1588
1589 for (i = 0 ; i < V_curr_dyn_buckets; i++) {
1590 IPFW_BUCK_LOCK(i);
1591 for (p = V_ipfw_dyn_v[i].head ; p != NULL; p = p->next) {
1592 if (bp + sizeof *p <= ep) {
1593 ipfw_dyn_rule *dst =
1594 (ipfw_dyn_rule *)bp;
1595
1596 export_dyn_rule(p, dst);
1597 last = dst;
1598 bp += sizeof(ipfw_dyn_rule);
1599 }
1600 }
1601 IPFW_BUCK_UNLOCK(i);
1602 }
1603
1604 if (last != NULL) /* mark last dynamic rule */
1605 bzero(&last->next, sizeof(last));
1606 *pbp = bp;
1607 }
1608 /* end of file */
1609